issue_owner_repo
listlengths 2
2
| issue_body
stringlengths 0
261k
⌀ | issue_title
stringlengths 1
925
| issue_comments_url
stringlengths 56
81
| issue_comments_count
int64 0
2.5k
| issue_created_at
stringlengths 20
20
| issue_updated_at
stringlengths 20
20
| issue_html_url
stringlengths 37
62
| issue_github_id
int64 387k
2.46B
| issue_number
int64 1
127k
|
|---|---|---|---|---|---|---|---|---|---|
[
"hwchase17",
"langchain"
]
| ### Feature request
First of all BIG thanks for the working you are doing guys!!!
It will be nice if you can add a stop generate response implementation.
### Motivation
I'm always frustrated when I need to wait until the final response returned and then ask new question.
Sometimes my chatbot sending long JSON response that takes 55sec end-to-end it will be nice if I could stop it manually.
### Your contribution
No | Python - Stop generate response functionality | https://api.github.com/repos/langchain-ai/langchain/issues/10874/comments | 1 | 2023-09-21T06:14:41Z | 2023-12-28T16:05:22Z | https://github.com/langchain-ai/langchain/issues/10874 | 1,906,192,130 | 10,874 |
[
"hwchase17",
"langchain"
]
| ### System Info
```
try:
# Skip any validation in case of forced collection recreate.
if force_recreate:
raise ValueError
# Get the vector configuration of the existing collection and vector, if it
# was specified. If the old configuration does not match the current one,
# an exception is being thrown.
collection_info = client.get_collection(collection_name=collection_name)
current_vector_config = collection_info.config.params.vectors
if isinstance(current_vector_config, dict) and vector_name is not None:
if vector_name not in current_vector_config:
raise QdrantException(
f"Existing Qdrant collection {collection_name} does not "
f"contain vector named {vector_name}. Did you mean one of the "
f"existing vectors: {', '.join(current_vector_config.keys())}? "
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
current_vector_config = current_vector_config.get(
vector_name
) # type: ignore[assignment]
elif isinstance(current_vector_config, dict) and vector_name is None:
raise QdrantException(
f"Existing Qdrant collection {collection_name} uses named vectors. "
f"If you want to reuse it, please set `vector_name` to any of the "
f"existing named vectors: "
f"{', '.join(current_vector_config.keys())}." # noqa
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
elif (
not isinstance(current_vector_config, dict) and vector_name is not None
):
raise QdrantException(
f"Existing Qdrant collection {collection_name} doesn't use named "
f"vectors. If you want to reuse it, please set `vector_name` to "
f"`None`. If you want to recreate the collection, set "
f"`force_recreate` parameter to `True`."
)
# Check if the vector configuration has the same dimensionality.
if current_vector_config.size != vector_size: # type: ignore[union-attr]
raise QdrantException(
f"Existing Qdrant collection is configured for vectors with "
f"{current_vector_config.size} " # type: ignore[union-attr]
f"dimensions. Selected embeddings are {vector_size}-dimensional. "
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
current_distance_func = (
current_vector_config.distance.name.upper() # type: ignore[union-attr]
)
if current_distance_func != distance_func:
raise QdrantException(
f"Existing Qdrant collection is configured for "
f"{current_vector_config.distance} " # type: ignore[union-attr]
f"similarity. Please set `distance_func` parameter to "
f"`{distance_func}` if you want to reuse it. If you want to "
f"recreate the collection, set `force_recreate` parameter to "
f"`True`."
)
except (UnexpectedResponse, RpcError, ValueError):
vectors_config = rest.VectorParams(
size=vector_size,
distance=rest.Distance[distance_func],
)
# If vector name was provided, we're going to use the named vectors feature
# with just a single vector.
if vector_name is not None:
vectors_config = { # type: ignore[assignment]
vector_name: vectors_config,
}
client.recreate_collection(
collection_name=collection_name,
vectors_config=vectors_config,
shard_number=shard_number,
replication_factor=replication_factor,
write_consistency_factor=write_consistency_factor,
on_disk_payload=on_disk_payload,
hnsw_config=hnsw_config,
optimizers_config=optimizers_config,
wal_config=wal_config,
quantization_config=quantization_config,
init_from=init_from,
timeout=timeout, # type: ignore[arg-type]
)
```
I want to know which idiot wrote this code.
Recreate the collection, then won't all the data be lost?
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Not all exceptions should be handled by recreate collection
### Expected behavior
Not all exceptions should be handled by recreate collection | Qdrant collection issue | https://api.github.com/repos/langchain-ai/langchain/issues/10872/comments | 1 | 2023-09-21T05:52:25Z | 2023-09-21T07:01:10Z | https://github.com/langchain-ai/langchain/issues/10872 | 1,906,164,528 | 10,872 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain 0.0.295
python 3.11
os macOS 12.5
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
The model appears to be missing from the _default_params method in the baidu_qianfan_endpoint.py file
`
@property
def _default_params(self) -> Dict[str, Any]:
"""Get the default parameters for calling OpenAI API."""
normal_params = {
"stream": self.streaming,
"request_timeout": self.request_timeout,
"top_p": self.top_p,
"temperature": self.temperature,
"penalty_score": self.penalty_score,
}
return {**normal_params, **self.model_kwargs}
`
Line 183 of the chat_completion.py file in the qianfan module failed to be executed.
`
def do(
self,
model: Optional[str] = None,
endpoint: Optional[str] = None,
stream: bool = False,
retry_count: int = 1,
request_timeout: float = 60,
backoff_factor: float = 0,
**kwargs,
) -> QfResponse:
"""
if model is EB, use EB SDK to deal with the request
"""
if "messages" in kwargs and isinstance(kwargs["messages"], QfMessages):
kwargs["messages"] = kwargs["messages"]._to_list()
if not GLOBAL_CONFIG.DISABLE_EB_SDK:
if model in ["ERNIE-Bot-turbo", "ERNIE-Bot"]:
import erniebot #line 183
erniebot.ak = self._client._auth._ak
erniebot.sk = self._client._auth._sk
erniebot.access_token = self._client._auth.access_token()
# compat with eb sdk
if model == "ERNIE-Bot":
model = "ernie-bot-3.5"
return erniebot.ChatCompletion.create(
model=model.lower(), messages=kwargs["messages"], stream=stream
)
return super().do(
model,
endpoint,
stream,
retry_count,
request_timeout,
backoff_factor,
**kwargs,
)
`
but even with model，baidu will return the following error
`
File "/Users/xxx/Workspace/xxx/xxx/chatbot/venv/lib/python3.11/site-packages/erniebot/backends.py", line 113, in handle_response
raise errors.InvalidParameterError(emsg)
erniebot.errors.InvalidParameterError: the length of messages must be an odd number
`
because there are two messages in the requested messages
### Expected behavior
able to return correct results | The model appears to be missing from the _default_params method in the baidu_qianfan_endpoint.py file | https://api.github.com/repos/langchain-ai/langchain/issues/10867/comments | 7 | 2023-09-21T02:30:24Z | 2024-01-30T00:41:06Z | https://github.com/langchain-ai/langchain/issues/10867 | 1,905,974,168 | 10,867 |
[
"hwchase17",
"langchain"
]
| ### System Info
- OS: Ubuntu 20.04
- `langhcain==0.0.297`
- ` chromadb==0.4.12`
- Python 3.9.18
### Who can help?
@hwchase17
@agola11
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```python
from langchain.document_loaders import TextLoader
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import Chroma
loader = TextLoader("state_of_the_union.txt")
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
texts = text_splitter.split_documents(documents)
embeddings = embeddings = HuggingFaceEmbeddings(model_name='all-MiniLM-L6-v2', encode_kwargs={'normalize_embeddings': True})
docsearch = Chroma.from_documents(texts, embeddings)
retriever = docsearch.as_retriever(search_type="similarity_score_threshold", search_kwargs={'score_threshold': 0.2})
print(retriever.get_relevant_documents('Ketanji'))
```
The above code returns negative similarity scores for all retrieved results:
```python
lib/python3.9/site-packages/langchain/vectorstores/base.py:257: UserWarning: Relevance scores must be between 0 and 1, got
[(Document(page_content='Tonight. I call on the Senate to: Pass the Freedom to Vote Act. Pass the John Lewis Voting Rights Act. And while you’re at it, pass the Disclose Act so Americans can know who is funding our elections. \n\nTonight, I’d like to honor someone who has dedicated his life to serve this country: Justice Stephen Breyer—an Army veteran, Constitutional scholar, and retiring Justice of the United States Supreme Court. Justice Breyer, thank you for your service. \n\nOne of the most serious constitutional responsibilities a President has is nominating someone to serve on the United States Supreme Court. \n\nAnd I did that 4 days ago, when I nominated Circuit Court of Appeals Judge Ketanji Brown Jackson. One of our nation’s top legal minds, who will continue Justice Breyer’s legacy of excellence.', metadata={'source': 'state_of_the_union.txt'}), -0.18782109124725155),
(Document(page_content='Madam Speaker, Madam Vice President, our First Lady and Second Gentleman. Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans. \n\nLast year COVID-19 kept us apart. This year we are finally together again. \n\nTonight, we meet as Democrats Republicans and Independents. But most importantly as Americans. \n\nWith a duty to one another to the American people to the Constitution. \n\nAnd with an unwavering resolve that freedom will always triumph over tyranny. \n\nSix days ago, Russia’s Vladimir Putin sought to shake the foundations of the free world thinking he could make it bend to his menacing ways. But he badly miscalculated. \n\nHe thought he could roll into Ukraine and the world would roll over. Instead he met a wall of strength he never imagined. \n\nHe met the Ukrainian people. \n\nFrom President Zelenskyy to every Ukrainian, their fearlessness, their courage, their determination, inspires the world.', metadata={'source': 'state_of_the_union.txt'}), -0.2858684850499025),
(Document(page_content='We are inflicting pain on Russia and supporting the people of Ukraine. Putin is now isolated from the world more than ever. \n\nTogether with our allies –we are right now enforcing powerful economic sanctions. \n\nWe are cutting off Russia’s largest banks from the international financial system. \n\nPreventing Russia’s central bank from defending the Russian Ruble making Putin’s $630 Billion “war fund” worthless. \n\nWe are choking off Russia’s access to technology that will sap its economic strength and weaken its military for years to come. \n\nTonight I say to the Russian oligarchs and corrupt leaders who have bilked billions of dollars off this violent regime no more. \n\nThe U.S. Department of Justice is assembling a dedicated task force to go after the crimes of Russian oligarchs. \n\nWe are joining with our European allies to find and seize your yachts your luxury apartments your private jets. We are coming for your ill-begotten gains.', metadata={'source': 'state_of_the_union.txt'}), -0.3382525501830016),
(Document(page_content='For that purpose we’ve mobilized American ground forces, air squadrons, and ship deployments to protect NATO countries including Poland, Romania, Latvia, Lithuania, and Estonia. \n\nAs I have made crystal clear the United States and our Allies will defend every inch of territory of NATO countries with the full force of our collective power. \n\nAnd we remain clear-eyed. The Ukrainians are fighting back with pure courage. But the next few days weeks, months, will be hard on them. \n\nPutin has unleashed violence and chaos. But while he may make gains on the battlefield – he will pay a continuing high price over the long run. \n\nAnd a proud Ukrainian people, who have known 30 years of independence, have repeatedly shown that they will not tolerate anyone who tries to take their country backwards. \n\nTo all Americans, I will be honest with you, as I’ve always promised. A Russian dictator, invading a foreign country, has costs around the world.', metadata={'source': 'state_of_the_union.txt'}), -0.3629898842731978)]
```
### Expected behavior
Score for each `Document` should be between 0 and 1. | When `search_type="similarity_score_threshold`, retriever returns negative scores | https://api.github.com/repos/langchain-ai/langchain/issues/10864/comments | 23 | 2023-09-20T23:14:16Z | 2024-08-01T22:17:48Z | https://github.com/langchain-ai/langchain/issues/10864 | 1,905,836,300 | 10,864 |
[
"hwchase17",
"langchain"
]
| ### System Info
Langhchain version: 0.0.296
Python version: 3.9.7
Platform: Windows 10
### Who can help?
I need to implement a fallback for my RAG application based on Langchain conversational retrievers because my application needs to use GPT 4 only for RAG and GPT 3.5 for generic questions (eg.: ChatGPT). A way to implement this behaviour is by means of fallbacks, because infortunately create_conversational_retrieval_agent() enforces the same LLM for all the tools. So, I wrote the following code:
```
short_llm = ChatOpenAI(
deployment_id=DEPLOYMENT_NAME_GPT_3_5_4K,
engine=DEPLOYMENT_NAME_GPT_3_5_4K,
model_name=model_gpt_3_5_4k,
temperature=0,
openai_api_base=BASE_URL,
openai_api_key=API_KEY,
streaming=True,
callbacks=[StreamingStdOutCallbackHandler()]
)
long_llm = ChatOpenAI(
deployment_id=DEPLOYMENT_NAME_GPT_4_8K,
engine=DEPLOYMENT_NAME_GPT_4_8K,
model_name=model_gpt_4_8k,
temperature=0,
openai_api_base=BASE_URL,
openai_api_key=API_KEY,
streaming=True,
callbacks=[StreamingStdOutCallbackHandler()]
)
chat_llm = short_llm.with_fallbacks([long_llm])
```
And to create the agent:
`agent_executor = create_conversational_retrieval_agent(chat_llm, tools, verbose=True, sytem_message=system_prompt, remember_intermediate_steps=False)`
But the following error occurs:
```
C:\ProgramData\Anaconda3\lib\site-packages\langchain\agents\agent_toolkits\conversational_retrieval\openai_functions.py in create_conversational_retrieval_agent(llm, tools, remember_intermediate_steps, memory_key, system_message, verbose, max_token_limit, **kwargs)
58
59 if not isinstance(llm, ChatOpenAI):
---> 60 raise ValueError("Only supported with ChatOpenAI models.")
61 if remember_intermediate_steps:
62 memory: BaseMemory = AgentTokenBufferMemory(
ValueError: Only supported with ChatOpenAI models.
```
In fact, chat_llm has the following type: langchain.schema.runnable.base.RunnableWithFallbacks. It apparently does not inherit ChatOpenAI .
Is there any workaround? My requirement is to use long LLM for RAG/retriever tools and a short one for normal conversation with ChatGPT 3.5
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```
short_llm = ChatOpenAI(
deployment_id=DEPLOYMENT_NAME_GPT_3_5_4K,
engine=DEPLOYMENT_NAME_GPT_3_5_4K,
model_name=model_gpt_3_5_4k,
temperature=0,
openai_api_base=BASE_URL,
openai_api_key=API_KEY,
streaming=True,
callbacks=[StreamingStdOutCallbackHandler()]
)
long_llm = ChatOpenAI(
deployment_id=DEPLOYMENT_NAME_GPT_4_8K,
engine=DEPLOYMENT_NAME_GPT_4_8K,
model_name=model_gpt_4_8k,
temperature=0,
openai_api_base=BASE_URL,
openai_api_key=API_KEY,
streaming=True,
callbacks=[StreamingStdOutCallbackHandler()]
)
chat_llm = short_llm.with_fallbacks([long_llm])
```
And to create the agent:
`agent_executor = create_conversational_retrieval_agent(chat_llm, tools, verbose=True, sytem_message=system_prompt, remember_intermediate_steps=False)`
### Expected behavior
Conversational retrieval agents should be able to work with fallbacks. | create_conversational_retrieval_agent does not work with fallbacks | https://api.github.com/repos/langchain-ai/langchain/issues/10852/comments | 7 | 2023-09-20T19:39:26Z | 2024-01-30T00:41:06Z | https://github.com/langchain-ai/langchain/issues/10852 | 1,905,609,497 | 10,852 |
[
"hwchase17",
"langchain"
]
| ### integration tests.
I have tried to follow the readme in langchain/tests to run the integration tests but got these errors :
`ModuleNotFoundError: No module named 'qdrant_client'`
Using poetry with python version = 3.10
I already tried the `poetry install` and` poetry install --with test_integration` but does't helped. The readme seems don't have much information around this and couldn't find any document that what is guideline to run the integration tests.
The unit tests are running fine.
Is it maybe something wrong with my env or I might missing a step before running integration tests?
### Suggestion:
_No response_ | Issue: running integration tests | https://api.github.com/repos/langchain-ai/langchain/issues/10849/comments | 1 | 2023-09-20T16:50:05Z | 2023-09-21T11:13:05Z | https://github.com/langchain-ai/langchain/issues/10849 | 1,905,383,407 | 10,849 |
[
"hwchase17",
"langchain"
]
| ### System Info
Python 3.9.18
langchain 0.0.292
transformers 4.33.2
torch 2.0.1
os macOs Ventura 13.5.2
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```
from transformers import pipeline
from langchain import PromptTemplate
from langchain.llms.huggingface_pipeline import HuggingFacePipeline
from transformers import AutoModelForCausalLM, AutoTokenizer
from langchain.chains import LLMChain
model = AutoModelForCausalLM.from_pretrained("microsoft/phi-1_5",
trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(
"microsoft/phi-1_5", trust_remote_code=True, use_cache=True, return_attention_mask=False)
question = "Who won the FIFA World Cup in the year 1994? "
template = """Question: {question}
Answer: Let's think step by step."""
prompt = PromptTemplate(template=template, input_variables=["question"])
# llm_pipeline = HuggingFacePipeline(pipeline=generate)
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
llm_pipeline = HuggingFacePipeline(pipeline=pipe)
generator_chain = LLMChain(llm=llm_pipeline, prompt=prompt)
generator_outputs = generator_chain.run(question)
print(generator_outputs)
```
ValueError: The following `model_kwargs` are not used by the model: ['attention_mask'] (note: typos in the generate arguments will also show up in this list)
### Expected behavior
Return the response from the model | ValueError: The following `model_kwargs` are not used by the model while running Microsoft/Phi-1.5 using HuggingFacePipeline | https://api.github.com/repos/langchain-ai/langchain/issues/10848/comments | 3 | 2023-09-20T16:49:29Z | 2023-12-27T16:04:18Z | https://github.com/langchain-ai/langchain/issues/10848 | 1,905,382,623 | 10,848 |
[
"hwchase17",
"langchain"
]
| ### System Info
Linux, Python 3.11.5
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
1. `python -m venv .venv`
2. `. .venv/bin/activate`
3. `pip install -e .`
### Expected behavior
To install package in a local virtualenv.
What happens:
```
% pip install -e .
Obtaining file:///home/iskren/src/deepinfra/libs/langchain
Installing build dependencies ... done
Checking if build backend supports build_editable ... done
Getting requirements to build editable ... error
error: subprocess-exited-with-error
× Getting requirements to build editable did not run successfully.
│ exit code: 1
╰─> [14 lines of output]
error: Multiple top-level packages discovered in a flat-layout: ['libs', 'langchain'].
To avoid accidental inclusion of unwanted files or directories,
setuptools will not proceed with this build.
If you are trying to create a single distribution with multiple packages
on purpose, you should not rely on automatic discovery.
Instead, consider the following options:
1. set up custom discovery (`find` directive with `include` or `exclude`)
2. use a `src-layout`
3. explicitly set `py_modules` or `packages` with a list of names
To find more information, look for "package discovery" on setuptools docs.
[end of output]
note: This error originates from a subprocess, and is likely not a problem with pip.
error: subprocess-exited-with-error
× Getting requirements to build editable did not run successfully.
│ exit code: 1
╰─> See above for output.
note: This error originates from a subprocess, and is likely not a problem with pip.
``` | Installing from source (pip install -e .) fails with multiple top-level packags found | https://api.github.com/repos/langchain-ai/langchain/issues/10844/comments | 3 | 2023-09-20T15:51:04Z | 2024-01-22T06:20:00Z | https://github.com/langchain-ai/langchain/issues/10844 | 1,905,287,813 | 10,844 |
[
"hwchase17",
"langchain"
]
| ### System Info
LANGCHAIN - 0.0.0286
OS - Windows
### Who can help?
@eyurtsev
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [X] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
To reproduce, take a HTML document that has text between STRONG tags of HTML tags and extract using directoryloader
### Expected behavior
It should extract all the text from HTML | Directory loader does not extract text between <strong> </strong> tags in HTML | https://api.github.com/repos/langchain-ai/langchain/issues/10841/comments | 3 | 2023-09-20T14:50:31Z | 2024-03-13T20:01:58Z | https://github.com/langchain-ai/langchain/issues/10841 | 1,905,168,973 | 10,841 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain=0.0.294
python=3.10
### Who can help?
@hwchase17 @agola11
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```
loader = TextLoader("state_of_the_union.txt")
documents = loader.load()
pattern = "\.|,|\n\n"
text_splitter = CharacterTextSplitter(separator=pattern, is_separator_regex=True,
chunk_size=100, chunk_overlap=0,
keep_separator=False)
docs = text_splitter.split_documents(documents)
for doc in docs:
print(doc.page_content)
```
the output would be
```
'Madam Speaker\\.|,|\n\n Madam Vice President\\.|,|\n\n our First Lady and Second Gentleman'
'Members of Congress and the Cabinet\\.|,|\n\n Justices of the Supreme Court\\.|,|\n\n My fellow Americans'
'\\.|,|\n\nLast year COVID-19 kept us apart\\.|,|\n\n This year we are finally together again\\.|,|'
```
which includes my regex expression. I even set keep_seperator=False.
with keep_seperator=True, the seperator become the leading character, which is still stange.
```
'Madam Speaker, Madam Vice President, our First Lady and Second Gentleman'
'. Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.'
'Last year COVID-19 kept us apart. This year we are finally together again.'
```
### Expected behavior
```
'Madam Speaker, Madam Vice President, our First Lady and Second Gentleman.'
'Members of Congress and the Cabinet. Justices of the Supreme Court. My fellow Americans.'
'Last year COVID-19 kept us apart. This year we are finally together again.'
``` | Confused behavior of CharacterTextSplitter with regex | https://api.github.com/repos/langchain-ai/langchain/issues/10840/comments | 5 | 2023-09-20T14:16:53Z | 2024-04-09T16:13:27Z | https://github.com/langchain-ai/langchain/issues/10840 | 1,905,102,656 | 10,840 |
[
"hwchase17",
"langchain"
]
| ### Discussed in https://github.com/langchain-ai/langchain/discussions/10835
<div type='discussions-op-text'>
<sup>Originally posted by **astro-siddhesh** September 20, 2023</sup>
import pandas as pd
import streamlit as st
from langchain.chat_models import ChatOpenAI
from langchain.agents.agent_toolkits import create_pandas_dataframe_agent
from langchain.agents.agent_types import AgentType
df = pd.read_csv('/Users/siddheshphapale/Desktop/project/sqlcsv.csv')
llm = ChatOpenAI(openai_api_key= "my key" , temperature=0 ,max_tokens= 500 , verbose= False)
agent = create_pandas_dataframe_agent(llm, df, agent_type=AgentType.OPENAI_FUNCTIONS)
from langsmith import Client
client = Client()
def send_feedback(run_id, score):
client.create_feedback(run_id, "user_score", score=score)
st.set_page_config(page_title='🦜🔗 Ask the CSV App')
st.title('🦜🔗 Ask the CSV App')
query_text = st.text_input('Enter your question:', placeholder = 'Who was in cabin C128?')
# Form input and query
result = None
with st.form('myform', clear_on_submit=True):
submitted = st.form_submit_button('Submit')
if submitted:
with st.spinner('Calculating...'):
response = agent({"input": query_text}, include_run_info=True)
result = response["output"]
run_id = response["__run"].run_id
if result is not None:
st.info(result)
col_blank, col_text, col1, col2 = st.columns([10, 2,1,1])
with col_text:
st.text("Feedback:")
with col1:
st.button("👍", on_click=send_feedback, args=(run_id, 1))
with col2:
st.button("👎", on_click=send_feedback, args=(run_id, 0))
<img width="600" alt="Screenshot 2023-09-20 at 5 21 38 PM" src="https://github.com/langchain-ai/langchain/assets/118797304/536155f7-fb12-4fa2-8dc8-80aea70e7389">
</div> | below code is not able to show grpahs at frontend , rather it returns code , | https://api.github.com/repos/langchain-ai/langchain/issues/10836/comments | 2 | 2023-09-20T11:52:08Z | 2023-12-27T16:04:28Z | https://github.com/langchain-ai/langchain/issues/10836 | 1,904,823,416 | 10,836 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
in docs，has a simple demo to use LongContextReorder to reorder docs.
but ,if i want use it with RetrievalQAChain ,how can i use it. thx
### Suggestion:
_No response_ | how to use LongContextReorder with RetrievalQAChain | https://api.github.com/repos/langchain-ai/langchain/issues/10834/comments | 9 | 2023-09-20T11:22:57Z | 2024-03-29T16:06:10Z | https://github.com/langchain-ai/langchain/issues/10834 | 1,904,777,227 | 10,834 |
[
"hwchase17",
"langchain"
]
| ### System Info
My systeminfo:
langchain 0.0.295
Python 3.11
### Who can help?
_No response_
### Information
- [X] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
That is my code where I set up my agent with custom tools, each tool has a name, description and access to a specific vector store with a collection name.
```python
def process_prompt(user_prompt: str, session_id="allgemein", debug=False):
openai.api_key = config.openai_api_key
llm = ChatOpenAI(temperature=0, model="gpt-3.5-turbo-16k", max_tokens=4096)
service_context = ServiceContext.from_defaults(llm=llm)
used_tools = []
for tool in tools_to_use:
vector_store = get_vector_store(tool.get("collection"))
index = VectorStoreIndex.from_vector_store(vector_store=vector_store, service_context=service_context)
filters = None
vector_store_info = VectorStoreInfo(
content_info=tool.get("description"),
metadata_info=[]
)
retriever = VectorIndexAutoRetriever(index, vector_store_info=vector_store_info)
used_tools.append(
Tool(
name=tool.get("name"),
description=tool.get("description"),
return_direct=tool.get("return_direct"),
func=lambda q: str(retriever.retrieve(q)),
)
)
message_history = PostgresChatMessageHistory(
connection_string=f"postgresql://postgres:{config.postgres}, session_id=session_id, table_name="chat_history",
)
memory = ConversationBufferMemory(
memory_key="chat_history", chat_memory=message_history
)
prompt = PromptTemplate(input_variables=["input", "chat_history"], template=summary_prompt_template)
summary_chain = LLMChain(
llm=llm,
prompt=prompt,
verbose=True,
memory=memory, # <--- this is the only change
)
used_tools.append(
Tool(
name="Summary Tool",
description="useful for retrieving memory summaries",
func=summary_chain.run,
)
)
prompt = CustomPromptTemplate(template=get_prompt_template(), tools=used_tools,
input_variables=["input", "intermediate_steps", "chat_history"])
output_parser = CustomOutputParser()
llm_chain = LLMChain(llm=llm, prompt=prompt)
tool_names = [tool.name for tool in used_tools]
agent = LLMSingleActionAgent(
llm_chain=llm_chain,
output_parser=output_parser,
stop=["\nObservation:"], #
allowed_tools=tool_names,
handle_parsing_errors=True,
)
agent_executor = AgentExecutor.from_agent_and_tools(agent=agent, tools=used_tools, verbose=debug,
memory=memory, return_intermediate_steps=False)
response = agent_executor.run(user_prompt)
```
Here is my code for my custom prompt template and output parser:
```python
class CustomPromptTemplate(BaseChatPromptTemplate):
template: str
tools: List[Tool]
def format_messages(self, **kwargs) -> list[HumanMessage]:
# Get the intermediate steps (AgentAction, Observation tuples)
# Format them in a particular way
intermediate_steps = kwargs.pop("intermediate_steps")
thoughts = ""
for action, observation in intermediate_steps:
thoughts += action.log
thoughts += f"\nObservation: {observation}\nThought: "
# Set the agent_scratchpad variable to that value
kwargs["agent_scratchpad"] = thoughts
# print(thoughts)
# Create a tools variable from the list of tools provided
kwargs["tools"] = "\n".join([f"{tool.name}: {tool.description}" for tool in self.tools])
# Create a list of tool names for the tools provided
kwargs["tool_names"] = ", ".join([tool.name for tool in self.tools])
formatted = self.template.format(**kwargs)
return [HumanMessage(content=formatted)]
class CustomOutputParser(AgentOutputParser):
def parse(self, llm_output: str) -> Union[AgentAction, AgentFinish]:
# Check if agent should finish
if "Final Answer:" in llm_output:
# print(llm_output)
return AgentFinish(
# Return values is generally always a dictionary with a single `output` key
# It is not recommended to try anything else at the moment :)
return_values={"output": llm_output.split("Final Answer:")[-1]},
log=llm_output,
)
# Parse out the action and action input
regex = r"Action\s*\d*\s*:(.*?)\nAction\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
match = re.search(regex, llm_output, re.DOTALL)
if match:
action = match.group(1).strip()
action_input = match.group(2)
# Return the action and action input
return AgentAction(tool=action, tool_input=action_input.strip(" ").strip('"'), log=llm_output)
else:
# raise ValueError(f"Could not parse output: {llm_output}")
return AgentFinish(return_values={"output": llm_output.split("Thought:")[-1]}, log=llm_output)
```
### Expected behavior
The agent should use exactly the tool what is defined in the action, Instead he always writes:
Example:
```
action: Notion Tool
Input Action: [some input]
Thought: I use the Notion Tool to retrieve some information about the input.
....
```
But when I print the agent_scratchpad it shows that he retrieved information from an other Tool instead of that what he uses. | LLMSingleActionAgent mixes up tools | https://api.github.com/repos/langchain-ai/langchain/issues/10833/comments | 5 | 2023-09-20T11:14:36Z | 2023-10-18T15:11:47Z | https://github.com/langchain-ai/langchain/issues/10833 | 1,904,764,017 | 10,833 |
[
"hwchase17",
"langchain"
]
| ### System Info
The GoogleDrive loader currently uses the PyPDF2 library instead of the PyPDF.
Since PyPDF2 was already merged to the original one, it shouldn't be used anymore.
[https://data.safetycli.com/v/59234/f17/?utm_source=pyupio&utm_medium=redirect&utm_campaign=pyup_rd&utm_id=081](url)
This is the vulnerability issue that gets picked up, note that it is fixed in PyPDF
Also I tried to change the loader using `file_loader_cls` but it just doesn't work
### Who can help?
_No response_
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [X] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
No functionality issue but https://data.safetycli.com/v/59234/f17/?utm_source=pyupio&utm_medium=redirect&utm_campaign=pyup_rd&utm_id=0817&utm_content=data showcases the bug
### Expected behavior
This error should not pop up | PyPDF2 used in Google Drive loader has vulnerability issues | https://api.github.com/repos/langchain-ai/langchain/issues/10832/comments | 4 | 2023-09-20T10:26:26Z | 2023-12-27T16:04:33Z | https://github.com/langchain-ai/langchain/issues/10832 | 1,904,680,537 | 10,832 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
`
llm = QianfanChatEndpoint(
model="ChatGLM2-6B-32K",
qianfan_ak="xxx",
qianfan_sk="xxx"
)
embedding = ErnieEmbeddings(
qianfan_ak="xxx",
qianfan_sk="xxx"
)
db = Chroma.from_documents(texts, embedding)
retriever = db.as_retriever()
qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever)
qa.run("请问事项的名称是什么？")
`
an error occurred in the above code.
`
File "/Users/xxx/Workspace/xxx/smart_gov_nav/chatbot/venv/lib/python3.11/site-packages/langchain/chat_models/baidu_qianfan_endpoint.py", line 63, in convert_message_to_dict
raise TypeError(f"Got unknown type {message}")
TypeError: Got unknown type content="Use the following pieces of context to answer the users question. \nIf you don't know the answer, just say that you don't know, don't try to make up an answer.\n----------------\n办事大厅名称\n " additional_kwargs={}
`
Because the type of message in the convert_message_to_dict argument is SystemMessage. That's why this error occurred. How to use it correctly?
### Suggestion:
_No response_ | Issue: use QianfanChatEndpoint and RetrievalQA Error | https://api.github.com/repos/langchain-ai/langchain/issues/10831/comments | 5 | 2023-09-20T10:18:23Z | 2024-01-30T00:41:08Z | https://github.com/langchain-ai/langchain/issues/10831 | 1,904,666,457 | 10,831 |
[
"hwchase17",
"langchain"
]
| ### System Info
Hello! I'm using Conversational Retrieval Agents in a Nodejs environment with Faiss vectorStore. The problem is that the agent does not search for simurality. Vectors are normally well created and saved in a folder by Faiss.
It should be noted that the chatbot communicates well but does not take vectors into account.
Here are the snippets of my code :
**creation and export of the chain**
`const llm = new ChatOpenAI({
temperature: 0,
maxTokens: 300,
modelName: "gpt-4"
});
const memory = new OpenAIAgentTokenBufferMemory({
llm: llm,
memoryKey: "chat_history",
returnMessages: true,
inputKey:"input",
outputKey:"output",
});
const makeChain = async (vectorstore,basePrompt) => {
const retriever = vectorstore.asRetriever();
const tool = createRetrieverTool(retriever, {
name: "search_state_of_union",
description: "Searches and returns documents regarding the state-of-the-union.",
});
const tools = [tool]
const chain = await initializeAgentExecutorWithOptions(tools,llm,{
verbose: true,
agentType: "openai-functions",
memory:memory,
returnIntermediateSteps: true,
agentArgs: {
prefix:basePrompt +` `+`Do your best to answer the questions. Feel free to use any tools available to look up relevant information, only if necessary.`,
},
});
return chain;
};
`
**Using of the chain**
` const vectorStorePath = process.env.VECTOR_STORE_PATH || path.join(__dirname, '..', 'vectorStore');
const directory = path.join(vectorStorePath, chatbot.name);
console.log(directory);
const vectorStore=await FaissStore.load(
directory,
new OpenAIEmbeddings()
);
/**const resultTwo = await vectorStore.similaritySearch("nestor", 1);
console.log(resultTwo); */
// Sanitize the message
const sanitizedMessage = message.trim().replaceAll('\n', ' ')
//create chain
const chain = await makeChain(vectorStore, chatbot.basePrompt);
//Ask a question using chat history
const response = await chain.call({
input: sanitizedMessage,
chat_history: chatHistory || []
});
`
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
**Create and export the chain**
`const llm = new ChatOpenAI({
temperature: 0,
maxTokens: 300,
modelName: "gpt-4"
});
const memory = new OpenAIAgentTokenBufferMemory({
llm: llm,
memoryKey: "chat_history",
returnMessages: true,
inputKey:"input",
outputKey:"output",
});
const makeChain = async (vectorstore,basePrompt) => {
const retriever = vectorstore.asRetriever();
const tool = createRetrieverTool(retriever, {
name: "search_state_of_union",
description: "Searches and returns documents regarding the state-of-the-union.",
});
const tools = [tool]
const chain = await initializeAgentExecutorWithOptions(tools,llm,{
verbose: true,
agentType: "openai-functions",
memory:memory,
returnIntermediateSteps: true,
agentArgs: {
prefix:basePrompt +` `+`Do your best to answer the questions. Feel free to use any tools available to look up relevant information, only if necessary.`,
},
});
return chain;
};
`
**Using of the chain**
` const vectorStorePath = process.env.VECTOR_STORE_PATH || path.join(__dirname, '..', 'vectorStore');
const directory = path.join(vectorStorePath, chatbot.name);
console.log(directory);
const vectorStore=await FaissStore.load(
directory,
new OpenAIEmbeddings()
);
/**const resultTwo = await vectorStore.similaritySearch("nestor", 1);
console.log(resultTwo); */
// Sanitize the message
const sanitizedMessage = message.trim().replaceAll('\n', ' ')
//create chain
const chain = await makeChain(vectorStore, chatbot.basePrompt);
//Ask a question using chat history
const response = await chain.call({
input: sanitizedMessage,
chat_history: chatHistory || []
});
`
### Expected behavior
Help me find a solution | Simirality search not working for Conversational Retrieval Agents | https://api.github.com/repos/langchain-ai/langchain/issues/10829/comments | 3 | 2023-09-20T10:06:42Z | 2023-12-27T16:04:43Z | https://github.com/langchain-ai/langchain/issues/10829 | 1,904,646,977 | 10,829 |
[
"hwchase17",
"langchain"
]
| ### System Info
LangChain version: tested with both versions 0.0.294 and 0.0.295.
Python Version: 3.10.10 running on Linux x86_64 (Ubuntu 22.04).
Tested with OpenAI library, both versions 0.27.8 and 0.28.0.
### Who can help?
@hwchase17 @agola11
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [x] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Run the following snippet:
```python
from langchain import OpenAI
llm = OpenAI(
model="gpt-3.5-turbo-instruct",
verbose=True,
max_tokens=-1,
)
llm("I am ")
```
The following error is returned (just the relevant stack trace):
```
File .venv/lib/python3.10/site-packages/langchain/llms/base.py:831, in BaseLLM.__call__(self, prompt, stop, callbacks, tags, metadata, **kwargs)
824 if not isinstance(prompt, str):
825 raise ValueError(
826 "Argument `prompt` is expected to be a string. Instead found "
827 f"{type(prompt)}. If you want to run the LLM on multiple prompts, use "
828 "`generate` instead."
829 )
830 return (
--> 831 self.generate(
832 [prompt],
833 stop=stop,
834 callbacks=callbacks,
835 tags=tags,
836 metadata=metadata,
837 **kwargs,
838 )
839 .generations[0][0]
840 .text
841 )
File .venv/lib/python3.10/site-packages/langchain/llms/base.py:627, in BaseLLM.generate(self, prompts, stop, callbacks, tags, metadata, **kwargs)
618 raise ValueError(
619 "Asked to cache, but no cache found at `langchain.cache`."
620 )
621 run_managers = [
622 callback_manager.on_llm_start(
623 dumpd(self), [prompt], invocation_params=params, options=options
624 )[0]
625 for callback_manager, prompt in zip(callback_managers, prompts)
626 ]
--> 627 output = self._generate_helper(
628 prompts, stop, run_managers, bool(new_arg_supported), **kwargs
629 )
630 return output
631 if len(missing_prompts) > 0:
File .venv/lib/python3.10/site-packages/langchain/llms/base.py:529, in BaseLLM._generate_helper(self, prompts, stop, run_managers, new_arg_supported, **kwargs)
527 for run_manager in run_managers:
528 run_manager.on_llm_error(e)
--> 529 raise e
530 flattened_outputs = output.flatten()
531 for manager, flattened_output in zip(run_managers, flattened_outputs):
File .venv/lib/python3.10/site-packages/langchain/llms/base.py:516, in BaseLLM._generate_helper(self, prompts, stop, run_managers, new_arg_supported, **kwargs)
506 def _generate_helper(
507 self,
508 prompts: List[str],
(...)
512 **kwargs: Any,
513 ) -> LLMResult:
514 try:
515 output = (
--> 516 self._generate(
517 prompts,
518 stop=stop,
519 # TODO: support multiple run managers
520 run_manager=run_managers[0] if run_managers else None,
521 **kwargs,
522 )
523 if new_arg_supported
524 else self._generate(prompts, stop=stop)
525 )
526 except BaseException as e:
527 for run_manager in run_managers:
File .venv/lib/python3.10/site-packages/langchain/llms/openai.py:357, in BaseOpenAI._generate(self, prompts, stop, run_manager, **kwargs)
355 params = self._invocation_params
356 params = {**params, **kwargs}
--> 357 sub_prompts = self.get_sub_prompts(params, prompts, stop)
358 choices = []
359 token_usage: Dict[str, int] = {}
File .venv/lib/python3.10/site-packages/langchain/llms/openai.py:459, in BaseOpenAI.get_sub_prompts(self, params, prompts, stop)
455 if len(prompts) != 1:
456 raise ValueError(
457 "max_tokens set to -1 not supported for multiple inputs."
458 )
--> 459 params["max_tokens"] = self.max_tokens_for_prompt(prompts[0])
460 sub_prompts = [
461 prompts[i : i + self.batch_size]
462 for i in range(0, len(prompts), self.batch_size)
463 ]
464 return sub_prompts
File .venv/lib/python3.10/site-packages/langchain/llms/openai.py:616, in BaseOpenAI.max_tokens_for_prompt(self, prompt)
602 """Calculate the maximum number of tokens possible to generate for a prompt.
603
604 Args:
(...)
613 max_tokens = openai.max_token_for_prompt("Tell me a joke.")
614 """
615 num_tokens = self.get_num_tokens(prompt)
--> 616 return self.max_context_size - num_tokens
File .venv/lib/python3.10/site-packages/langchain/llms/openai.py:599, in BaseOpenAI.max_context_size(self)
596 @property
597 def max_context_size(self) -> int:
598 """Get max context size for this model."""
--> 599 return self.modelname_to_contextsize(self.model_name)
File .venv/lib/python3.10/site-packages/langchain/llms/openai.py:589, in BaseOpenAI.modelname_to_contextsize(modelname)
586 context_size = model_token_mapping.get(modelname, None)
588 if context_size is None:
--> 589 raise ValueError(
590 f"Unknown model: {modelname}. Please provide a valid OpenAI model name."
591 "Known models are: " + ", ".join(model_token_mapping.keys())
592 )
594 return context_size
ValueError: Unknown model: gpt-3.5-turbo-instruct. Please provide a valid OpenAI model name.Known models are: gpt-4, gpt-4-0314, gpt-4-0613, gpt-4-32k, gpt-4-32k-0314, gpt-4-32k-0613, gpt-3.5-turbo, gpt-3.5-turbo-0301, gpt-3.5-turbo-0613, gpt-3.5-turbo-16k, gpt-3.5-turbo-16k-0613, text-ada-001, ada, text-babbage-001, babbage, text-curie-001, curie, davinci, text-davinci-003, text-davinci-002, code-davinci-002, code-davinci-001, code-cushman-002, code-cushman-001
```
Originally got this through a `LLMChain`, but tracked it down to the LLM itself.
### Expected behavior
Should be processing the input correctly. Can confirm this works with other models like `text-davinci-003`. | Providing `-1` to `max_tokens` while creating an OpenAI LLM using the `gpt-3.5-turbo-instruct` is failing | https://api.github.com/repos/langchain-ai/langchain/issues/10822/comments | 2 | 2023-09-20T07:58:46Z | 2023-09-20T08:10:49Z | https://github.com/langchain-ai/langchain/issues/10822 | 1,904,404,073 | 10,822 |
[
"hwchase17",
"langchain"
]
| ### Feature request
When using MongoDBAtlasVectorSearch, the user is limited to only first populate a MongoDB collection of documents using either add_document or add_text methods. Those methods create documents on the given collection using a fixed structure. What if the user already has a collection on MongoDB with custom structure (still containing the field 'embedding' which is used to measure similarity)? This is not currently supported and the user would have to implement a wrapper around MongoDBAtlasVectorSearch to achieve that.
For instance the following code won't work (docs will be empty):
```
from langchain.vectorstores import MongoDBAtlasVectorSearch
from langchain.embeddings import HuggingFaceEmbeddings
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
index_name = "index_name_from_MongoDB"
docsearch = MongoDBAtlasVectorSearch(embeddings, collection=collection, index_name=index_name)
query = "What did the president say about Ketanji Brown Jackson"
docs = docsearch.similarity_search(query)
print(docs[0].page_content)
```
This is because the collection is not populated using any of the original methods of MongoDBAtlasVectorSearch, e.g. add_document or add_text, hence the structure of the collection might be different (containing additional fields and missing the 'text' field for example). | MongoDBAtlasVectorSearch: Make It Work with Existing MongoDB Collection (rather than having to create a new one) | https://api.github.com/repos/langchain-ai/langchain/issues/10820/comments | 2 | 2023-09-20T06:52:26Z | 2024-02-06T16:29:31Z | https://github.com/langchain-ai/langchain/issues/10820 | 1,904,284,433 | 10,820 |
[
"hwchase17",
"langchain"
]
| ### System Info
Mac OS ventura 13.4.1
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
langchain/document_loaders/url_selenium.py
def load(self) -> List[Document]:
"""Load the specified URLs using Selenium and create Document instances.
Returns:
List[Document]: A list of Document instances with loaded content.
"""
from unstructured.partition.html import partition_html
docs: List[Document] = list()
driver = self._get_driver()
for url in self.urls:
try:
driver.get(url)
sleep(30) // need wait
page_content = driver.page_source
elements = partition_html(text=page_content)
text = "\n\n".join([str(el) for el in elements])
metadata = {"source": url}
docs.append(Document(page_content=text, metadata=metadata))
except Exception as e:
if self.continue_on_failure:
logger.error(f"Error fetching or processing {url}, exception: {e}")
else:
raise e
driver.quit()
return docs
### Expected behavior
timer is config parms, user can use
SeleniumURLLoader({waittime: 30}) | SeleniumURLLoader load asynchronous js, need wait | https://api.github.com/repos/langchain-ai/langchain/issues/10814/comments | 4 | 2023-09-20T04:00:24Z | 2024-05-10T16:06:50Z | https://github.com/langchain-ai/langchain/issues/10814 | 1,904,069,804 | 10,814 |
[
"hwchase17",
"langchain"
]
| ### System Info
I'm using latest langchain version 0.0.295. It turns out langchain SQLAgent always connects to Postgres public schema. I need SQLAgent to check all schemas to generate SQL query. How to configure SQLAgent to connect to all Postgres schemas?
Tried both pg_uri = f"postgresql+psycopg2://{username}:{password}@{host}:{port}/{mydatabase}?currentSchema=myschema", pg_uri = f"postgresql+psycopg2://{username}:{password}@{host}:{port}/{mydatabase}?search_path=myschema" when creating SQLDatabase: db = SQLDatabase.from_uri(pg_uri). It doesn't work.
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
When creating SQLDatabase for postgres, I can't specify schema. It only connects to postgres public schema. As a result, SQLAgent doesn't produce meaningful result.
# pg_uri = f"postgresql+psycopg2://{username}:{password}@{host}:{port}/{mydatabase}?search_path=myschema"
pg_uri = f"postgresql+psycopg2://{username}:{password}@{host}:{port}/{mydatabase}?currentSchema=myschema"
db = SQLDatabase.from_uri(pg_uri)
### Expected behavior
Expect SQLAgent or SQLDatabase can configure schemas for Postgres so that SQLAgent can generate meaningful SQL query. | SQLAgent fails to connect to Postgres multiple schemas | https://api.github.com/repos/langchain-ai/langchain/issues/10811/comments | 3 | 2023-09-20T01:19:04Z | 2023-12-27T16:04:48Z | https://github.com/langchain-ai/langchain/issues/10811 | 1,903,925,576 | 10,811 |
[
"hwchase17",
"langchain"
]
| ### System Info
Python 3.11.0
LangChain==0.0.295
Azure-ai-vision==0.15.1b1
Streamlit
### Who can help?
@agola11 @hwchase17
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [X] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Important to note that the below error is INTERMITTEND using the exact same input every single time. Sometimes it works for an hour without error, then it will throw the below error a few times and return to normal. Other times it throws this error like 20 times before it will return to normal and start working again. It is like it gets confused about what inputs the agent expect at random.
Also note that this is a simplified version. The reason I use create_conversational_retrieval_agent is because I pass multiple tools like O365, SQLDatabase ect. But I can still produce the error without them.
**Error:**
TypeError: AzureCogsImageAnalysisTool._run() got an unexpected keyword argument 'url'
Traceback:
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 552, in _run_script
exec(code, module.__dict__)
File "C:\_Dev\AZURE-AGENT-TEST\home.py", line 409, in <module>
main()
File "C:\_Dev\AZURE-AGENT-TEST\home.py", line 256, in main
handle_userinput(user_question, image_url)
File "C:\_Dev\AZURE-AGENT-TEST\home.py", line 151, in handle_userinput
response = st.session_state.conversation({'input': user_question + " " + '"' + file_url + '"' })
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\langchain\chains\base.py", line 292, in __call__
raise e
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\langchain\chains\base.py", line 286, in __call__
self._call(inputs, run_manager=run_manager)
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\langchain\agents\agent.py", line 1122, in _call
next_step_output = self._take_next_step(
^^^^^^^^^^^^^^^^^^^^^
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\langchain\agents\agent.py", line 977, in _take_next_step
observation = tool.run(
^^^^^^^^^
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\langchain\tools\base.py", line 356, in run
raise e
File "C:\_Dev\AZURE-AGENT-TEST\venv\Lib\site-packages\langchain\tools\base.py", line 328, in run
self._run(*tool_args, run_manager=run_manager, **tool_kwargs)
**Initialize the agent:**
toolkit = AzureCognitiveServicesToolkit()
tools = toolkit.get_tools()
llm = ChatOpenAI(temperature=0.7)
st.session_state.conversation = create_conversational_retrieval_agent(llm,
tools,
verbose=False,
max_iterations=3,
early_stopping_method='generate',
AgentType=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION)
**Then handle the user input:**
user_question = "What can I make with these ingredients?"
file_url = "https://images.openai.com/blob/9ad5a2ab-041f-475f-ad6a-b51899c50182/ingredients.png"
response = st.session_state.conversation({'input': user_question + " " + '"' + file_url + '"' })
st.write(response)
### Expected behavior
I expect it to not throw this error intermittently. | INTERMITTENT ERROR: AzureCogsImageAnalysisTool._run() got an unexpected keyword argument 'url' | https://api.github.com/repos/langchain-ai/langchain/issues/10810/comments | 3 | 2023-09-20T00:25:58Z | 2023-12-27T16:04:53Z | https://github.com/langchain-ai/langchain/issues/10810 | 1,903,890,762 | 10,810 |
[
"hwchase17",
"langchain"
]
| ### Issue with current documentation:
While following the Quickstart guide in the LangChain documentation, I encountered an AttributeError when trying to use the `predict` method with an instance of the OpenAI class. It seems like the `predict` method is not available in the OpenAI class as described in the guide.
I'm running the following code from the Langchain Quickstart Guide, LLMs section:
```
from langchain.llms import OpenAI
from langchain.chat_models import ChatOpenAI
llm = OpenAI()
chat_model = ChatOpenAI()
llm.predict("hi!")
>>> "Hi"
chat_model.predict("hi!")
>>> "Hi"
```
Below are the steps I took before running:
1. Installed LangChain using `pip install langchain`.
2. Installed OpenAI Python package using `pip install openai`.
3. Set up the OpenAI API key as described in the guide.
4. Created an instance of the OpenAI class and attempted to call the `predict` method with a string argument, as shown in the guide.
**Expected behavior:** Based on the Quickstart guide, I expected to be able to call the `predict` method with a string argument and receive a string as the output.
**Actual behavior:** When attempting to call the `predict` method, I received the following error message:
AttributeError: 'OpenAI' object has no attribute 'predict'
**Environment:**
- LangChain version: (0.0.112)
- OpenAI Python package version: (0.27.2)
- Python version: 3.9.5
- Operating System: macOS
I checked the available methods of the OpenAI class using the `dir()` function and confirmed that the `predict` method is not available. I also tried passing a list of strings to the `generate` method as an alternative, based on the methods that are available, but this did not produce the expected result.
### Idea or request for content:
_No response_ | OpenAI class missing `predict` method as described in the Quickstart guide | https://api.github.com/repos/langchain-ai/langchain/issues/10809/comments | 1 | 2023-09-19T23:51:29Z | 2023-09-20T00:00:51Z | https://github.com/langchain-ai/langchain/issues/10809 | 1,903,867,154 | 10,809 |
[
"hwchase17",
"langchain"
]
| ### System Info
LangChain version: 0.0.295 (just upgraded to this version to use gpt-3.5-turbo-instruct)
### Who can help?
@hwchase17 @agola
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Minimal code to reproduce:
```python
# load OpenAI API Key
from langchain.llms import OpenAI
llm = OpenAI(temperature=0.1, model_name="gpt-3.5-turbo-instruct", max_tokens=-1)
llm("give me a list of Chinese dishes and their recipes")
```
Error message:
>```ValueError: Unknown model: gpt-3.5-turbo-instruct. Please provide a valid OpenAI model name.Known models are: gpt-4, gpt-4-0314, gpt-4-0613, gpt-4-32k, gpt-4-32k-0314, gpt-4-32k-0613, gpt-3.5-turbo, gpt-3.5-turbo-0301, gpt-3.5-turbo-0613, gpt-3.5-turbo-16k, gpt-3.5-turbo-16k-0613, text-ada-001, ada, text-babbage-001, babbage, text-curie-001, curie, davinci, text-davinci-003, text-davinci-002, code-davinci-002, code-davinci-001, code-cushman-002, code-cushman-001```
Cause of the error: looks like it's because the `model_token_mapping` is missing an entry for `gpt-3.5-turbo-instruct`: https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/llms/openai.py#L555
### Expected behavior
The code succeeds without error | Error when using gpt-3.5-turbo-instruct: model_token_mapping is missing an entry for gpt-3.5-turbo-instruct | https://api.github.com/repos/langchain-ai/langchain/issues/10806/comments | 1 | 2023-09-19T23:26:18Z | 2023-09-20T00:03:17Z | https://github.com/langchain-ai/langchain/issues/10806 | 1,903,852,127 | 10,806 |
[
"hwchase17",
"langchain"
]
| Using the GrobidParser with `segment_sentences=False` results in `langchain/document_loaders/parsers/grobid.py", line 87, in process_xml
chunk_bboxes[0][0]["page"],
IndexError: list index out of range` for some PDFs.
From my own testing with the offending PDFs I suspect the following if-block in the GrobidParser is indented incorrectly, it needs one more level of indentation. Adding this indentation solved the issue on my end.
https://github.com/langchain-ai/langchain/blob/20c6ade2fc24c458a080eaa468add7c61dc8dbc3/langchain/document_loaders/parsers/grobid.py#L80-L93
If you need a reproduction to verify, let me know. | GrobidParser: process_xml chunk_bboxes[0][0]["page"], IndexError: list index out of range | https://api.github.com/repos/langchain-ai/langchain/issues/10801/comments | 4 | 2023-09-19T21:09:13Z | 2024-03-27T13:14:05Z | https://github.com/langchain-ai/langchain/issues/10801 | 1,903,726,835 | 10,801 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
According to the [ReadTheDocsLoader documentation](https://api.python.langchain.com/en/latest/document_loaders/langchain.document_loaders.readthedocs.ReadTheDocsLoader.html), this loader iterates through all the files located under a specified path and extracts the actual content of the files by retrieving `main` html tags.
```python
from langchain.document_loaders import ReadTheDocsLoader
loader = ReadTheDocsLoader("docs", encoding="utf-8", features="html.parser")
docs = loader.load()
len(docs)
```
Question: In my HTML files, I don't have main HTML tags. Is it possible to configure the ReadTheDocsLoader to extract content from files by capturing all HTML tags instead of just the main HTML tags? Alternatively, are there alternative loaders available that can traverse all files within a specified path and extract content by capturing all HTML tags?
| Query: Is It Possible to Extract Content Using All HTML Tags in ReadTheDocsLoader? | https://api.github.com/repos/langchain-ai/langchain/issues/10798/comments | 4 | 2023-09-19T20:15:53Z | 2023-12-26T16:04:46Z | https://github.com/langchain-ai/langchain/issues/10798 | 1,903,665,132 | 10,798 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
**I tried to use create_pandas_dataframe_agent with GPT4ALL to query a csv. This is the output:**
```
Enter a query: Analyse the data and give precise answer to the question that is given below. Provide answer only to what
is asked. Answer should be with respect to the whole document. Question: How many employees are male"
> Entering new AgentExecutor chain...
Thought: I will use pandas dataframe to get the summary of the data
Action: I will use pandas dataframe to get the summary of the data
Observation: The summary of the data is given below
... (this Observation/Action/Observation can repeat N times)
Final Answer: The summary of the data shows that there are 4 male employees in the company.Traceback (most recent call last):
File "C:\project\privateGPT\privateGPT\analyticsGPT.py", line 102, in <module>
main()
File "C:\project\privateGPT\privateGPT\analyticsGPT.py", line 73, in main
agent.run(query)
File "C:\Python311\Lib\site-packages\langchain\chains\base.py", line 475, in run
return self(args[0], callbacks=callbacks, tags=tags, metadata=metadata)[
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\site-packages\langchain\chains\base.py", line 282, in __call__
raise e
File "C:\Python311\Lib\site-packages\langchain\chains\base.py", line 276, in __call__
self._call(inputs, run_manager=run_manager)
File "C:\Python311\Lib\site-packages\langchain\agents\agent.py", line 1036, in _call
next_step_output = self._take_next_step(
^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\site-packages\langchain\agents\agent.py", line 844, in _take_next_step
raise e
File "C:\Python311\Lib\site-packages\langchain\agents\agent.py", line 833, in _take_next_step
output = self.agent.plan(
^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\site-packages\langchain\agents\agent.py", line 457, in plan
return self.output_parser.parse(full_output)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\site-packages\langchain\agents\mrkl\output_parser.py", line 61, in parse
raise OutputParserException(
langchain.schema.output_parser.OutputParserException: Could not parse LLM output: `
Thought: I will use pandas dataframe to get the summary of the data
Action: I will use pandas dataframe to get the summary of the data`
```
Given below is my code snippet:
```
#!/usr/bin/env python3
from dotenv import load_dotenv
from langchain.chains import RetrievalQA
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from langchain.vectorstores import Chroma
from langchain.llms import GPT4All, LlamaCpp
import chromadb
import os
import argparse
import time
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from langchain.agents import create_pandas_dataframe_agent
if not load_dotenv():
print("Could not load .env file or it is empty. Please check if it exists and is readable.")
exit(1)
embeddings_model_name = os.environ.get("EMBEDDINGS_MODEL_NAME")
persist_directory = os.environ.get('PERSIST_DIRECTORY')
model_type = os.environ.get('MODEL_TYPE')
model_path = os.environ.get('MODEL_PATH')
model_n_ctx = os.environ.get('MODEL_N_CTX')
model_n_batch = int(os.environ.get('MODEL_N_BATCH',8))
target_source_chunks = int(os.environ.get('TARGET_SOURCE_CHUNKS',4))
from constants import CHROMA_SETTINGS
def main():
args = parse_arguments()
callbacks = [] if args.mute_stream else [StreamingStdOutCallbackHandler()]
if model_type == "LlamaCpp":
llm = LlamaCpp(model_path=model_path, max_tokens=model_n_ctx, n_batch=model_n_batch, callbacks=callbacks, verbose=False)
elif model_type == "GPT4All":
llm = GPT4All(model=model_path, max_tokens=model_n_ctx, backend='gptj', n_batch=model_n_batch, callbacks=callbacks, verbose=False)
else:
raise Exception(f"Model type {model_type} is not supported. Please choose one of the following: LlamaCpp, GPT4All")
df = pd.read_csv('C:/project/privateGPT/privateGPT/source_documents/employees.csv')
agent = create_pandas_dataframe_agent(llm,
df, verbose=True, handle_parsing_errors="Check your output and make sure it conforms!")
while True:
query = input("\nEnter a query: ")
if query == "exit":
break
if query.strip() == "":
continue
# Get the answer from the chain
start = time.time()
agent.run(query)
end = time.time()
def parse_arguments():
parser = argparse.ArgumentParser(description='privateGPT: Ask questions to your documents without an internet connection, '
'using the power of LLMs.')
parser.add_argument("--hide-source", "-S", action='store_true',
help='Use this flag to disable printing of source documents used for answers.')
parser.add_argument("--mute-stream", "-M",
action='store_true',
help='Use this flag to disable the streaming StdOut callback for LLMs.')
return parser.parse_args()
if __name__ == "__main__":
main()
```
### Suggestion:
_No response_ | Issue: output_parser raises error when using create_pandas_dataframe_agent with GPT4ALL model | https://api.github.com/repos/langchain-ai/langchain/issues/10792/comments | 2 | 2023-09-19T17:03:06Z | 2023-12-26T16:04:51Z | https://github.com/langchain-ai/langchain/issues/10792 | 1,903,398,229 | 10,792 |
[
"hwchase17",
"langchain"
]
| ### System Info
I'm building a chatbot to introduce matching job openings from files to users. When the user input "I want to find a job", it is modified to "What kind of job are you looking for?" which ask the model what job he is looking for, but of course, it won't get an answer.
Human input: I want to find a job
Human input in Prompt after formatting: What kind of job are you looking for?
Chatbot: As an AI, I don't have personal preferences or the ability to look for a job.
Here is my prompt and conversation:
```
template = """
You are an AI assistant specifically tasked with finding matching
job opportunities in our job data based on user requests.
###
Context from data: {context}
###
{chat_history}
Human: {question}
Chatbot:"""
prompt = PromptTemplate(
input_variables=["question", "chat_history", "context"],
template=template
)
qa = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=db.as_retriever(),
combine_docs_chain_kwargs={"prompt": prompt},
memory=memory,
verbose=True
)
```
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [X] Document Loaders
- [X] Vector Stores / Retrievers
- [X] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
template = """
You are an AI assistant specifically tasked with finding matching
job opportunities in our job data based on user requests.
###
Context from data: {context}
###
{chat_history}
Human: {question}
Chatbot:"""
prompt = PromptTemplate(
input_variables=["question", "chat_history", "context"],
template=template
)
qa = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=db.as_retriever(),
combine_docs_chain_kwargs={"prompt": prompt},
memory=memory,
verbose=True
)
### Expected behavior
the prompt should be what user have input | Prompt was changed to different meaning | https://api.github.com/repos/langchain-ai/langchain/issues/10789/comments | 2 | 2023-09-19T15:51:05Z | 2023-09-27T13:52:06Z | https://github.com/langchain-ai/langchain/issues/10789 | 1,903,281,810 | 10,789 |
[
"hwchase17",
"langchain"
]
| ### System Info
Having basicaly this demo code and the issue is that it's not working reliably!
Sometimes OPENAI_FUNCTIONS just doesn't look into tools and telling that he can't find the documents.
Seems like this happens because OpenAI decides on it's own if he rather use tools or replies on his own knowledge.
I also don't want to fill the memory with all the data every time I am starting a new conversation, this does not work with huge data files.
Any other way to solve this?
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
from pydantic import BaseModel, Field
from langchain.chat_models import ChatOpenAI
from langchain.agents import Tool
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import FAISS
from langchain.document_loaders import PyPDFLoader
from langchain.chains import RetrievalQA
class DocumentInput(BaseModel):
question: str = Field()
llm = ChatOpenAI(temperature=0, model="gpt-3.5-turbo-0613")
tools = []
files = [
# https://abc.xyz/investor/static/pdf/2023Q1_alphabet_earnings_release.pdf
{
"name": "alphabet-earnings",
"path": "/Users/harrisonchase/Downloads/2023Q1_alphabet_earnings_release.pdf",
},
# https://digitalassets.tesla.com/tesla-contents/image/upload/IR/TSLA-Q1-2023-Update
{
"name": "tesla-earnings",
"path": "/Users/harrisonchase/Downloads/TSLA-Q1-2023-Update.pdf",
},
]
for file in files:
loader = PyPDFLoader(file["path"])
pages = loader.load_and_split()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
docs = text_splitter.split_documents(pages)
embeddings = OpenAIEmbeddings()
retriever = FAISS.from_documents(docs, embeddings).as_retriever()
# Wrap retrievers in a Tool
tools.append(
Tool(
args_schema=DocumentInput,
name=file["name"],
description=f"useful when you want to answer questions about {file['name']}",
func=RetrievalQA.from_chain_type(llm=llm, retriever=retriever),
)
)
llm = ChatOpenAI(
temperature=0,
model="gpt-3.5-turbo-0613",
)
agent = initialize_agent(
agent=AgentType.OPENAI_FUNCTIONS,
tools=tools,
llm=llm,
verbose=True,
)
agent({"input": "did alphabet or tesla have more revenue?"})
### Expected behavior
Always checking the tools before answering a question. | Document Comparison, OPENAI_FUNCTIONS does not always look into the tools | https://api.github.com/repos/langchain-ai/langchain/issues/10787/comments | 6 | 2023-09-19T15:22:10Z | 2023-12-26T16:04:57Z | https://github.com/langchain-ai/langchain/issues/10787 | 1,903,231,986 | 10,787 |
[
"hwchase17",
"langchain"
]
| ### System Info
`Python 3.11.4`
`langchain==0.0.271`
`lark==1.1.7`
`openai==0.27.8`
### Who can help?
Getting the following error:
```sh
retriever = SelfQueryRetriever.from_llm(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/site-packages/langchain/retrievers/self_query/base.py", line 135, in from_llm
structured_query_translator = _get_builtin_translator(vectorstore)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/site-packages/langchain/retrievers/self_query/base.py", line 45, in _get_builtin_translator
raise ValueError(
ValueError: Self query retriever with Vector Store type <class 'langchain.vectorstores.supabase.SupabaseVectorStore'> not supported.
```
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
- initialize a SupabaseVectorStore. I want to use existing index, embeddings already created and stored in table `documents`
table definition:
```sql
create table
public.documents (
id bigserial,
content text null,
metadata jsonb null,
embedding public.vector null,
constraint documents_pkey primary key (id)
) tablespace pg_default;
```
```python
document_content_description = "Brief summary of a movie"
embeddings = OpenAIEmbeddings()
vectorstore = SupabaseVectorStore(
client=supabase,
embedding=embeddings,
table_name="documents",
query_name="match_documents",
)
llm = OpenAI(temperature=0)
retriever = SelfQueryRetriever.from_llm(
llm, vectorstore, document_content_description, metadata_field_info, verbose=True
)
retriever.get_relevant_documents("What are some movies about dinosaurs")
```
### Expected behavior
return relevant documents | initializing and using a Supabase vector store using SupabaseVectorStore | https://api.github.com/repos/langchain-ai/langchain/issues/10784/comments | 2 | 2023-09-19T14:46:15Z | 2023-12-26T16:07:16Z | https://github.com/langchain-ai/langchain/issues/10784 | 1,903,159,588 | 10,784 |
[
"hwchase17",
"langchain"
]
| ### System Info
My systeminfo:
langchain 0.0.294
Python 3.11
### Who can help?
Im not sure, maybe @hwchase17 and @agola11 can help
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [X] Memory
- [X] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
That is my code:
```python
openai.api_key = config.openai_api_key
llm = ChatOpenAI(temperature=1, model="gpt-3.5-turbo-16k", max_tokens=1024)
used_tools = []
for tool in tools_to_use:
vector_store = get_vector_store(tool.get("collection"))
index = VectorStoreIndex.from_vector_store(vector_store=vector_store, service_context=service_context)
vector_store_info = VectorStoreInfo(
content_info=tool.get("description"),
metadata_info=[]
)
retriever = VectorIndexAutoRetriever(index, vector_store_info=vector_store_info)
used_tools.append(
Tool(
name=tool.get("name"),
description=tool.get("description"),
func=lambda q: str(retriever.retrieve(q)),
)
)
message_history = PostgresChatMessageHistory(
connection_string=f"postgresql://postgres:{postgres_str}", session_id=session_id, table_name="chat_history",
)
prompt = CustomPromptTemplate(template=get_prompt_template(), tools=used_tools,
input_variables=["input", "intermediate_steps", "chat_history"])
output_parser = CustomOutputParser()
memory = ConversationBufferMemory(
memory_key="chat_history", chat_memory=message_history
)
llm_chain = LLMChain(llm=llm, prompt=prompt)
tool_names = [tool.name for tool in used_tools]
agent = LLMSingleActionAgent(
llm_chain=llm_chain,
output_parser=output_parser,
stop=["\nFinal Answer:"], # Observation
allowed_tools=tool_names,
handle_parsing_errors=True,
)
agent_executor = AgentExecutor.from_agent_and_tools(agent=agent, tools=used_tools, verbose=debug,
memory=memory, return_intermediate_steps=True)
response = agent_executor(user_prompt)
```
### Expected behavior
If i print the params, it shows my chat history (so it's successfully loaded into code and into the agent).
```
{'intermediate_steps': [], 'stop': ['\nFinal Answer:'], 'input': 'What was the last thing I said?', 'chat_history': 'Human: ....?\nAI:....']
```
It should also process the chat history correctly.
But I always get this answer to my prompt:
```
Since the previous chat history is not available, I can't tell what you said last. Please give me more information or ask your question again so I can help you further.
```
| Message History not working with LLMSingleActionAgent | https://api.github.com/repos/langchain-ai/langchain/issues/10783/comments | 7 | 2023-09-19T13:59:39Z | 2023-09-20T13:14:16Z | https://github.com/langchain-ai/langchain/issues/10783 | 1,903,062,021 | 10,783 |
[
"hwchase17",
"langchain"
]
| ### Feature request
This is the current implementation. You get your Vectorstore and turn it into a retriever. You pass search kwargs into the as_retriever method.
```
retriever = db.as_retriever(search_type="similarity_score_threshold", search_kwargs={"score_threshold": .5})
docs = retriever.get_relevant_documents("what did he say about ketanji brown jackson")
```
Is this as intended? I think it´s not the best way. We use PGVector and use it in a API where we pass metadata to do filtered search in the vectorstore.
How I think it should look like:
```
docs = retriever.get_relevant_documents("what did he say about ketanji brown jackson", search_kwargs={"score_threshold": .5})
```
So instead of using it in the constructor, I would rather prefer it in `get_relevant_documents` method.
### Motivation
I don´t want to create a new retriever instance every request and rather have `get_relevant_documents` more dynamic.
### Your contribution
As far as I can see this should be doable by passing some kwargs around. I could probably implement this if I am allowed to. | Vector store-backed retriever -> allow using search_kwargs in get_relevant_documents function (PGVector) | https://api.github.com/repos/langchain-ai/langchain/issues/10781/comments | 2 | 2023-09-19T13:19:02Z | 2023-12-26T16:05:07Z | https://github.com/langchain-ai/langchain/issues/10781 | 1,902,965,145 | 10,781 |
[
"hwchase17",
"langchain"
]
| ### Feature request
Most of the tools are currently sync tools, so if an async agent executor calls them tools raise `NotImplementedError`
My suggestion is to change `_arun` method of the `langchain.tools.base.BaseTool` class with a bolierplate code asyncer. An [example is given in the gh issues](https://github.com/langchain-ai/langchain/issues/5011#issuecomment-1725356727).
Asyncer is a small lib developped by @tiangolo (FastApi creator) that can convert a sync code to async (runs in worker threads)
This way by default we would have an async version of any tool. Tool implementors can always override `_arun` method in their implementation providing more native async support. But otherwise we would have a default async implementation.
### Motivation
In production systems we need to scale the requests to our langchain app, thus this means servin cconcurrently to many users. In python for IO Bound operations this is best done with async libs and frameworks. However all parts of the code has to be async, including LLMs + tools ..etc
Since many tools lack support for `_arun` method, this enhancement will provide default async support for all tools.
### Your contribution
Yes, I'll create a PR for this. | Make tools async by default unless user overrides `_arun` | https://api.github.com/repos/langchain-ai/langchain/issues/10779/comments | 2 | 2023-09-19T12:52:13Z | 2023-09-19T13:04:35Z | https://github.com/langchain-ai/langchain/issues/10779 | 1,902,915,680 | 10,779 |
[
"hwchase17",
"langchain"
]
| ### System Info
Using this Code I am getting this error:
TypeError: Chain.__call__() got an unexpected keyword argument 'question'
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
from pydantic import BaseModel, Field
from langchain.chat_models import ChatOpenAI
from langchain.agents import Tool
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import FAISS
from langchain.document_loaders import PyPDFLoader
from langchain.chains import RetrievalQA
from langchain.agents import initialize_agent
from langchain.agents import AgentType
from langchain.tools import StructuredTool
from dotenv import load_dotenv
class DocumentInput(BaseModel):
question: str = Field()
load_dotenv()
llm = ChatOpenAI(temperature=0, model="gpt-3.5-turbo-0613")
tools = []
files = [
# https://abc.xyz/investor/static/pdf/2023Q1_alphabet_earnings_release.pdf
{
"name": "alphabet-earnings",
"path": "/Users/harrisonchase/Downloads/2023Q1_alphabet_earnings_release.pdf",
},
# https://digitalassets.tesla.com/tesla-contents/image/upload/IR/TSLA-Q1-2023-Update
{
"name": "tesla-earnings",
"path": "/Users/harrisonchase/Downloads/TSLA-Q1-2023-Update.pdf",
},
]
for file in files:
loader = PyPDFLoader(file["path"])
pages = loader.load_and_split()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
docs = text_splitter.split_documents(pages)
embeddings = OpenAIEmbeddings()
retriever = FAISS.from_documents(docs, embeddings).as_retriever()
# Wrap retrievers in a Tool
tools.append(
StructuredTool.from_function(
args_schema=DocumentInput,
name=file["name"],
description=f"useful when you want to answer questions about {file['name']}",
func=RetrievalQA.from_chain_type(llm=llm, retriever=retriever),
)
)
llm = ChatOpenAI(
temperature=0,
model="gpt-3.5-turbo-0613",
)
agent = initialize_agent(
agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION,
tools=tools,
llm=llm,
verbose=True,
)
agent({"input": "did alphabet or tesla have more revenue?"})
### Expected behavior
Compare the two documents! | Using StructuredTool with agent | https://api.github.com/repos/langchain-ai/langchain/issues/10778/comments | 19 | 2023-09-19T12:21:53Z | 2024-06-23T14:28:42Z | https://github.com/langchain-ai/langchain/issues/10778 | 1,902,863,147 | 10,778 |
[
"hwchase17",
"langchain"
]
| ### System Info
I am developing a macOS application that uses Swift to call Python. To achieve this, I have integrated PythonKit, which also works properly. When I followed the integration of Ollama guided by Langchain, I encountered a problem and reported an error in 'from langchain.llms import Ollama'. The subsequent code cannot execute properly, but I ran it on the terminal without any issues. By setting environment variables, I managed to run both the Python environment and the terminal as 3.11.5, And I can successfully find olama.py under langchain/llms under site packages. I have tried installing TextLoader, which is directly installed under site packages and can import TextLoader normally. I guess it may not be possible to find deeper level packages. Can anyone encounter similar problems and help me solve them or provide some ideas? Thank you
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
from langchain.llms import Ollama
print('hello')
### Expected behavior
print('hello') | In xcode, use swift to call Python code and introduce Ollama error | https://api.github.com/repos/langchain-ai/langchain/issues/10773/comments | 5 | 2023-09-19T10:22:23Z | 2023-12-27T16:05:03Z | https://github.com/langchain-ai/langchain/issues/10773 | 1,902,666,640 | 10,773 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain==0.0.249
Python 3.10.9
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
import re
import json
from langchain.document_loaders import CSVLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import FAISS
from langchain.text_splitter import SpacyTextSplitter
from langchain import OpenAI, VectorDBQA
from langchain.document_loaders import TextLoader
from langchain.agents import initialize_agent, tool
from langchain.memory import ConversationBufferMemory
from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain
# VectorDBQA 更新为 RetrievalQA
from langchain.chains import RetrievalQA
llm = OpenAI(temperature=0)
loader = TextLoader('./data/faq/ecommerce_faq.txt')
documents = loader.load()
text_splitter = SpacyTextSplitter(chunk_size=256, pipeline="zh_core_web_sm")
texts = text_splitter.split_documents(documents)
embeddings = OpenAIEmbeddings()
docsearch = FAISS.from_documents(texts, embeddings)
faq_chain = VectorDBQA.from_chain_type(
llm=llm, vectorstore=docsearch, verbose=True)
product_loader = CSVLoader('./data/faq/ecommerce_products.csv')
product_documents = product_loader.load()
product_text_splitter = CharacterTextSplitter(chunk_size=1024, separator="\n")
product_texts = product_text_splitter.split_documents(product_documents)
product_search = FAISS.from_documents(product_texts, OpenAIEmbeddings())
product_chain = VectorDBQA.from_chain_type(
llm=llm, vectorstore=product_search, verbose=True)
ORDER_1 = "20230101ABC"
ORDER_2 = "20230101EFG"
ORDER_1_DETAIL = {
"order_number": ORDER_1,
"status": "已发货",
"shipping_date": "2023-01-03",
"estimated_delivered_date": "2023-01-05",
}
ORDER_2_DETAIL = {
"order_number": ORDER_2,
"status": "未发货",
"shipping_date": None,
"estimated_delivered_date": None,
}
answer_order_info = PromptTemplate(
template="请把下面的订单信息回复给用户： \n\n {order}?", input_variables=["order"]
)
answer_order_llm = LLMChain(llm=ChatOpenAI(
temperature=0), prompt=answer_order_info)
@tool("Search Order", return_direct=True)
def search_order(input: str) -> str:
"""useful for when you need to answer questions about customers orders"""
pattern = r"\d+[A-Z]+"
match = re.search(pattern, input)
order_number = input
if match:
order_number = match.group(0)
else:
return "请问您的订单号是多少？"
if order_number == ORDER_1:
return answer_order_llm.run(json.dumps(ORDER_1_DETAIL))
elif order_number == ORDER_2:
return answer_order_llm.run(json.dumps(ORDER_2_DETAIL))
else:
return f"对不起，根据{input}没有找到您的订单"
@tool("FAQ")
def faq(intput: str) -> str:
""""useful for when you need to answer questions about shopping policies, like return policy, shipping policy, etc."""
return faq_chain.run(intput)
@tool("Recommend Product")
def recommend_product(input: str) -> str:
""""useful for when you need to search and recommend products and recommend it to the user"""
return product_chain.run(input)
tools = [
search_order,
recommend_product,
faq
]
chatllm = ChatOpenAI(temperature=0)
memory = ConversationBufferMemory(
memory_key="chat_history", return_messages=True)
conversation_agent = initialize_agent(
tools,
chatllm,
agent="conversational-react-description",
memory=memory,
verbose=True
)
question = "请问你们的货，能送到三亚吗？大概需要几天？"
result = conversation_agent.run(question)
print(result)
### Expected behavior
Hope no error is reported | lib/python3.10/site-packages/langchain/agents/conversational/output_parser.py", line 26, in parse raise OutputParserException(f"Could not parse LLM output: `{text}`") langchain.schema.output_parser.OutputParserException: Could not parse LLM output: `Do I need to use a tool? No | https://api.github.com/repos/langchain-ai/langchain/issues/10770/comments | 3 | 2023-09-19T08:05:47Z | 2023-12-26T16:05:21Z | https://github.com/langchain-ai/langchain/issues/10770 | 1,902,432,993 | 10,770 |
[
"hwchase17",
"langchain"
]
| ### System Info
LangChain v0.0.294
### Who can help?
@hwchase17
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [X] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
I've recently encountered an issue that took a bit of time to isolate and identify. I have a service that utilizes a chain for conversations, where multiple instances are created of the same chain to manage responses to all users. However, I observed an issue where messages started to get mixed up.
Initially, I suspected the issue might be related to the websocket, but upon further investigation, I realized that the root cause seems to be the way `BaseChatMessageHistory` is declared. It appears that the messages list is implemented as a class attribute rather than an instance attribute. This seems to be causing memory to be shared across all instances, which doesn't align with the expected behavior and is causing the message mix-up that I observed.
I wanted to reach out to understand if there is a specific reason for implementing the messages list as a class attribute? In my perspective, modifying it to be an instance attribute might prevent such issues from occurring, and I believe it could potentially be considered a bug in its current state. I'm open to collaborating on a fix and would be happy to submit a PR to address this.
### Expected behavior
Just make it an instance attribute. | Issue with BaseChatMessageHistory Attribute Scope | https://api.github.com/repos/langchain-ai/langchain/issues/10764/comments | 2 | 2023-09-19T02:02:29Z | 2023-12-28T16:05:52Z | https://github.com/langchain-ai/langchain/issues/10764 | 1,902,072,269 | 10,764 |
[
"hwchase17",
"langchain"
]
| ### System Info
Python Version: `Python 3.10.12`
LangChain Version: `master`
### Who can help?
@hwchase17
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Following [Getting Started - Installing - From Source](https://python.langchain.com/docs/get_started/installation#from-source)
```
python3 -m pip install -e .
```
Yields an install of `UNKNOWN` package due to missing `[project]` section in the `pyproject.toml`
```
Successfully built UNKNOWN
Installing collected packages: UNKNOWN
Attempting uninstall: UNKNOWN
Found existing installation: UNKNOWN 0.0.0
Uninstalling UNKNOWN-0.0.0:
Successfully uninstalled UNKNOWN-0.0.0
Successfully installed UNKNOWN-0.0.0
```
### Expected behavior
```
Successfully built langchain
Installing collected packages: langchain
Successfully installed langchain-0.0.294
``` | pip install from source creates UNKNOWN package | https://api.github.com/repos/langchain-ai/langchain/issues/10760/comments | 2 | 2023-09-19T01:16:26Z | 2023-09-19T04:59:02Z | https://github.com/langchain-ai/langchain/issues/10760 | 1,902,034,213 | 10,760 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
I am using LLMSingleActionAgent with ReAct Prompt. Most of the times it fails to give the proper response to the input. For example, when I greet the agent, it is not greeting back most of the times. Below is the response snippet
**Below is my Agent code:**
agent = LLMSingleActionAgent(
llm_chain=llm_chain,
output_parser=output_parser,
stop=["\nObservation:"],
allowed_tools=tool_names
)
**Part of Prompt:**
You have access to the following tools. you can use tools, your Previous conversation history to answer the questions
{tools}
Previous Conversation history:
{history}
Use the following format:
Question: the input question you must answer
Thought: you should always think about what to do
Action: the action to take, can be one of [{tool_names}] or {history}
Action Input: the input to the action
Observation: the result of the action
Thought: I now know the final answer
Final Answer: the final answer to the original input question
**Output Parser:**
class CustomOutputParser(AgentOutputParser):
def parse(self, llm_output: str) -> Union[AgentAction, AgentFinish]:
# Check if agent should finish
if "Final Answer:" in llm_output:
return AgentFinish(
# Return values is generally always a dictionary with a single `output` key
# It is not recommended to try anything else at the moment :)
return_values={"output": llm_output.split("Final Answer:")[-1].strip()},
log=llm_output,
)
# Parse out the action and action input
regex = r"Action\s*\d*\s*:(.*?)\nAction\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
match = re.search(regex, llm_output, re.DOTALL)
if not match:
raise ValueError(f"Could not parse LLM output: `{llm_output}`")
action = match.group(1).strip()
action_input = match.group(2)
# Return the action and action input
return AgentAction(tool=action, tool_input=action_input.strip(" ").strip('"'), log=llm_output)
### Suggestion:
_No response_ | LLMSingleActionAgent Doesnot provide Expected response | https://api.github.com/repos/langchain-ai/langchain/issues/10749/comments | 2 | 2023-09-18T21:51:30Z | 2023-12-25T16:05:44Z | https://github.com/langchain-ai/langchain/issues/10749 | 1,901,803,694 | 10,749 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Recently, we re-deployed the llam2-13-B chat model on azure ML Studio, whenever we are trying to do in-contex learning (RAG QA), always replied back with out of context response.
Currently we are passing request payload as using langchain:
System prompt: sys msg + context
Human prompt: question
Response: out of context response
2nd variation:
Human prompt: context + question
Response: junk answers like \n
Any suggestions who have experienced?
And, we are also followed the required input format.
### Suggestion:
_No response_ | LLaMa-2-13B-chat deployed on azureml studio giving out of context response | https://api.github.com/repos/langchain-ai/langchain/issues/10747/comments | 3 | 2023-09-18T20:12:19Z | 2023-12-26T16:05:31Z | https://github.com/langchain-ai/langchain/issues/10747 | 1,901,665,767 | 10,747 |
[
"hwchase17",
"langchain"
]
| ### Feature request
I propose to add the Python client of gradient.ai as a `LLM`.
`gradient_ai.py` under https://github.com/langchain-ai/langchain/tree/master/libs/langchain/langchain/llms
```python
class GradientAI(LLM):
# client for Large Language Models on gradient.ai
#
```
See: API docs
https://docs.gradient.ai/reference/listmodels
### Motivation
Gradient.ai offers fine-tuning and inference of models without paying for the server costs.
I fine-tuned a model with a specific RAG template based on llama2 on gradient.ai, now i want to use it from Langchain.
### Your contribution
I would submit a PR. | Support of Gradient.ai LLM | https://api.github.com/repos/langchain-ai/langchain/issues/10745/comments | 1 | 2023-09-18T18:42:19Z | 2023-09-22T09:33:42Z | https://github.com/langchain-ai/langchain/issues/10745 | 1,901,534,677 | 10,745 |
[
"hwchase17",
"langchain"
]
| ### System Info
LangChain 0.0.292
Python 3.11.5
GPT4ALL with LLAMA q4_0 3b model running on CPU
### Who can help?
@agola11
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
My code is:
```from langchain.llms import GPT4All
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from langchain.agents import AgentType, initialize_agent
from langchain.tools import Tool
from pydantic import BaseModel, Field
class JokeInput(BaseModel):
confidence: float = Field(default=0.0)
import pyjokes
import langchain
langchain.debug = True
import os
os.environ["LANGCHAIN_TRACING"] = "true" # If you want to trace the execution of the program, set to "true"
# There are many CallbackHandlers supported, such as
# from langchain.callbacks.streamlit import StreamlitCallbackHandler
callbacks = [StreamingStdOutCallbackHandler()]
model = GPT4All(model=".cache/llama-2-7b-chat.ggmlv3.q4_0.bin", n_threads=6)
# Generate text. Tokens are streamed through the callback manager.
tools = []
tools.append(
Tool.from_function(
func=pyjokes.get_joke,
name="Joke",
description="useful for when you need to tell the user a joke",
args_schema=JokeInput
)
)
# Init the agent
agent = initialize_agent(
tools, model, agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION, verbose=True
)
# Start the agent
while True:
user_prompt = input("You: ")
agent(user_prompt, callbacks=callbacks)```
```
### Expected behavior
It should use the Joke tool and return a joke, but instead it gives me this error:
```
Found model file at .cache/llama-2-7b-chat.ggmlv3.q4_0.bin
llama.cpp: loading model from .cache/llama-2-7b-chat.ggmlv3.q4_0.bin
llama_model_load_internal: format = ggjt v3 (latest)
llama_model_load_internal: n_vocab = 32000
llama_model_load_internal: n_ctx = 2048
llama_model_load_internal: n_embd = 4096
llama_model_load_internal: n_mult = 256
llama_model_load_internal: n_head = 32
llama_model_load_internal: n_head_kv = 32
llama_model_load_internal: n_layer = 32
llama_model_load_internal: n_rot = 128
llama_model_load_internal: n_gqa = 1
llama_model_load_internal: rnorm_eps = 5.0e-06
llama_model_load_internal: n_ff = 11008
llama_model_load_internal: freq_base = 10000.0
llama_model_load_internal: freq_scale = 1
llama_model_load_internal: ftype = 2 (mostly Q4_0)
llama_model_load_internal: model size = 7B
llama_model_load_internal: ggml ctx size = 3615.73 MB
llama_model_load_internal: mem required = 4013.73 MB (+ 1024.00 MB per state)
llama_new_context_with_model: kv self size = 1024.00 MB
You: Tell me a joke
Failed to load default session, using empty session: HTTPConnectionPool(host='localhost', port=8000): Max retries exceeded with url: /sessions?name=default (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7ff8fc87d290>: Failed to establish a new connection: [Errno 111] Connection refused'))
[chain/start] [1:chain:AgentExecutor] Entering Chain run with input:
{
"input": "Tell me a joke"
}
[chain/start] [1:chain:AgentExecutor > 2:chain:LLMChain] Entering Chain run with input:
{
"input": "Tell me a joke",
"agent_scratchpad": "",
"stop": [
"Observation:"
]
}
[llm/start] [1:chain:AgentExecutor > 2:chain:LLMChain > 3:llm:GPT4All] Entering LLM run with input:
{
"prompts": [
"System: Answer the following questions as best you can. You have access to the following tools:\n\nJoke: useful for when you need to tell the user a joke\n\nThe way you use the tools is by specifying a json blob.\nSpecifically, this json should have a `action` key (with the name of the tool to use) and a `action_input` key (with the input to the tool going here).\n\nThe only values that should be in the \"action\" field are: Joke\n\nThe $JSON_BLOB should only contain a SINGLE action, do NOT return a list of multiple actions. Here is an example of a valid $JSON_BLOB:\n\n```\n{\n \"action\": $TOOL_NAME,\n \"action_input\": $INPUT\n}\n```\n\nALWAYS use the following format:\n\nQuestion: the input question you must answer\nThought: you should always think about what to do\nAction:\n```\n$JSON_BLOB\n```\nObservation: the result of the action\n... (this Thought/Action/Observation can repeat N times)\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n\nBegin! Reminder to always use the exact characters `Final Answer` when responding.\nHuman: Tell me a joke"
]
}
$JSON_BLOB[llm/end] [1:chain:AgentExecutor > 2:chain:LLMChain > 3:llm:GPT4All] [18.05s] Exiting LLM run with output:
{
"generations": [
[
{
"text": "$JSON_BLOB",
"generation_info": null
}
]
],
"llm_output": null,
"run": null
}
[chain/end] [1:chain:AgentExecutor > 2:chain:LLMChain] [18.05s] Exiting Chain run with output:
{
"text": "$JSON_BLOB"
}
[chain/error] [1:chain:AgentExecutor] [18.05s] Chain run errored with error:
"OutputParserException('Could not parse LLM output: $JSON_BLOB')"
Failed to persist run: HTTPConnectionPool(host='localhost', port=8000): Max retries exceeded with url: /chain-runs (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7ff8fc24f9d0>: Failed to establish a new connection: [Errno 111] Connection refused'))
Traceback (most recent call last):
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/agents/chat/output_parser.py", line 27, in parse
raise ValueError("action not found")
ValueError: action not found
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/mnt/0A956B927E2FFFE8/Code/jarvis/langchain_try.py", line 41, in <module>
agent(user_prompt, callbacks=callbacks)
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/chains/base.py", line 292, in __call__
raise e
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/chains/base.py", line 286, in __call__
self._call(inputs, run_manager=run_manager)
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/agents/agent.py", line 1122, in _call
next_step_output = self._take_next_step(
^^^^^^^^^^^^^^^^^^^^^
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/agents/agent.py", line 930, in _take_next_step
raise e
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/agents/agent.py", line 919, in _take_next_step
output = self.agent.plan(
^^^^^^^^^^^^^^^^
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/agents/agent.py", line 532, in plan
return self.output_parser.parse(full_output)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mnt/0A956B927E2FFFE8/Code/jarvis/.venv/lib/python3.11/site-packages/langchain/agents/chat/output_parser.py", line 42, in parse
raise OutputParserException(f"Could not parse LLM output: {text}")
langchain.schema.output_parser.OutputParserException: Could not parse LLM output: $JSON_BLOB
``` | Cannot use LangChain Tools with GPT4ALL and LLaMA model | https://api.github.com/repos/langchain-ai/langchain/issues/10744/comments | 4 | 2023-09-18T18:28:42Z | 2024-05-10T16:06:45Z | https://github.com/langchain-ai/langchain/issues/10744 | 1,901,513,550 | 10,744 |
[
"hwchase17",
"langchain"
]
| ### Feature request
Return the Youtube video links in full format like `https://www.youtube.com/watch?v=VIDEO_ID`
Currently the links are like `/watch?v=VIDEO_ID`
Return the links as List like `['link1, 'link2']`
Currently it is returning the whole list as string ` "['link1, 'link2']" `
### Motivation
If the links returned are exact same as **direct links to youtube in a list** rather than a string, i can avoid the hustle and bustle of processing it agian to convert to the required format
### Your contribution
I will change the code a bit and pull it. | Update return parameter of YouTubeSearchTool | https://api.github.com/repos/langchain-ai/langchain/issues/10742/comments | 1 | 2023-09-18T17:47:53Z | 2023-09-20T00:04:08Z | https://github.com/langchain-ai/langchain/issues/10742 | 1,901,452,804 | 10,742 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
I am using langchain with Pinecone to push the embedding into the Pincone vector db. using below code. But I see that there is too many IDs created for the single PDF file in Pincone Browswer(My PDF file more than 250 pages). I want to update my PDF for the same Index. How can I do ?
`from langchain.document_loaders import UnstructuredPDFLoader, OnlinePDFLoader, PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma, Pinecone
from langchain.embeddings.openai import OpenAIEmbeddings
import pinecone
import os
from langchain.llms import OpenAI
from langchain.chains.question_answering import load_qa_chain
loader = UnstructuredPDFLoader("example.pdf")
data = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=0)
texts = text_splitter.split_documents(data)
type(texts)
OPENAI_API_KEY = "sk-xxxxx"
PINECONE_API_KEY = "xxxx"
PINECONE_API_ENV = "us-west-gcp-free"
embeddings = OpenAIEmbeddings(openai_api_key=OPENAI_API_KEY)
pinecone.init(
api_key=PINECONE_API_KEY,
environment=PINECONE_API_ENV
)
index_name = "langchaintest"
docsearch = Pinecone.from_texts([t.page_content for t in texts], embeddings, index_name=index_name)`
I want to update my vectors Due to multiple IDs I am not able to get the ID to update the New one PDF file
### Suggestion:
ID should be one for the single PDF file. | Issue: How to update Pinecone vectors data using langchain | https://api.github.com/repos/langchain-ai/langchain/issues/10739/comments | 5 | 2023-09-18T13:31:20Z | 2024-02-11T16:14:12Z | https://github.com/langchain-ai/langchain/issues/10739 | 1,900,951,737 | 10,739 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Hello everyone,
I am facing an issue with a Python function that I have created to handle some chat functionality with the ChatGPT model. The function is part of a bigger project that aims to use Pinecone to search and retrieve similar documents.
Here is a simplified version of my code:
> def pineconeGPT(system, user, questions, answers, instruction, metadata={}):
llm = ChatOpenAI(model='gpt-3.5-turbo-16k', temperature=0, openai_api_key="key")
embeddings = OpenAIEmbeddings(
openai_api_key="key", model="text-embedding-ada-002")
pinecone.init(api_key='key', environment='region')
index = Pinecone.from_existing_index('index_name', embeddings)
history = ChatMessageHistory()
if len(questions) > 0:
for i in range(len(questions)):
history.add_ai_message(questions[i])
history.add_user_message(answers[i])
retriever = index.as_retriever(
search_type="similarity",
search_kwargs={'filter': metadata}
)
pineconeIndex = create_retriever_tool(
retriever,
"search_pinecone",
"Searches and returns documents regarding the similarity."
)
tools = [pineconeIndex]
system_message = SystemMessage(
content=(system)
prompt = OpenAIFunctionsAgent.create_prompt(
system_message=system_message
)
agent = OpenAIFunctionsAgent(llm=llm, tools=tools, prompt=prompt)
agent_executor = AgentExecutor(
agent=agent,
tools=tools,
memory=ConversationBufferMemory(chat_memory=history),
verbose=True,
return_intermediate_steps=False,
return_only_outputs=False)
result = agent_executor({"input" : instruction})
The history is included in the result but for example:
If the question is: What is your name?
Answer: My name is Bob
And the instruction says "what is my name"
ChatGPT response is saying I have not provided name.
This suggest that ChatGPT has not being passed the history by the agent.
I would appreciate any insights on what might be causing this issue and how to resolve it. Thank you in advance!
### Suggestion:
_No response_ | Issue: <Langchain Retriever Agent Function Not Passing “ChatMessageHistory” When Called in Console - Python> | https://api.github.com/repos/langchain-ai/langchain/issues/10737/comments | 2 | 2023-09-18T13:17:46Z | 2023-09-21T10:06:11Z | https://github.com/langchain-ai/langchain/issues/10737 | 1,900,924,280 | 10,737 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain 0.0.292
The type of text_splitter must be TextSplitter and not RecursiveCharacterTextSplitter
```
class WebResearchRetriever(BaseRetriever):
...
text_splitter: **TextSplitter** = Field(
RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=50),
description="Text splitter for splitting web pages into chunks",
)
```
### Who can help?
@hwchase17
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Use WebResearchRetriever with a simple TextRetriver. It's run.
Now `make lint` => Error
### Expected behavior
No error with `make lint`
Pull request [here](https://github.com/langchain-ai/langchain/pull/10734) | Typing error in WebResearchRetriever | https://api.github.com/repos/langchain-ai/langchain/issues/10736/comments | 2 | 2023-09-18T12:58:58Z | 2023-09-19T07:29:37Z | https://github.com/langchain-ai/langchain/issues/10736 | 1,900,884,697 | 10,736 |
[
"hwchase17",
"langchain"
]
| ### System Info
```
langchain==0.0.291
chromadb==0.4.10
Python 3.10.0
22.6.0 Darwin Kernel
```
### Who can help?
@agola11 @hwchase17 bug map_reduce - reproducible
### Information
- [X] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [X] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [X] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
## Description
The bug arrises when using `map_reduce` with `RetrievalQAWithSourcesChain.from_chain_type` and `Chroma,` and adding more than 1 document.
Specifically, we have `paper.pdf` and `paper2.pdf`.
`paper.pdf` talks about `'x=3'`
`paper2.pdf` talks about `'y=5'`
*This is key*: Embedding and searching for each one individually with just *one* works perfectly.
If you embed and add the two (or anything more than 1) back to back, and you search with `map_reduce` get this error:
(NOTE: *this is also key*, for the same embedding, if you switch to using `RetrievalQAWithSourcesChain.from_chain_type` with `stuff`, it also *works perfectly*)
## Example of Error:
```
--------------> PDF1
[Document(page_content='This is a sample paper!This paper talks about x.In this case, we are going to say that "x = 3"', metadata={'source': '/Users/user/Desktop/paper.pdf', 'page': 0, 'start_index': 0})]
Number of requested results 4 is greater than number of elements in index 1, updating n_results = 1
========================
{'question': 'what is x?', 'answer': 'x = 3.\n', 'sources': '/Users/user/Desktop/paper.pdf', 'source_documents': [Document(page_content='This is a sample paper!This paper talks about x.In this case, we are going to say that "x = 3"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper.pdf', 'start_index': 0})]}
========================
--------------> PDF2
[Document(page_content='This is another sample paper! This paper talks about y. In this case, we are going to say that "y = 5"', metadata={'source': '/Users/user/Desktop/paper2.pdf', 'page': 0, 'start_index': 0})]
Number of requested results 4 is greater than number of elements in index 2, updating n_results = 2
========================
{'question': 'what is y?', 'answer': 'There is no relevant text provided to answer the question "what is y?"\n', 'sources': '/Users/user/Desktop/paper.pdf', 'source_documents': [Document(page_content='This is another sample paper! This paper talks about y. In this case, we are going to say that "y = 5"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper2.pdf', 'start_index': 0}), Document(page_content='This is a sample paper!This paper talks about x.In this case, we are going to say that "x = 3"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper.pdf', 'start_index': 0})]}
========================
```
Again, -- each one by itself works. (both `paper.pdf` and `paper2.pdf` produces the correct answer individually). Also, moving to the `stuff` chain, produces the correct result for both or more documents.
```
--------------> PDF1
[Document(page_content='This is a sample paper!\r\nThis paper talks about x.\r\nIn this case, we are going to say that "x = 3"', metadata={'source': '/Users/user/Desktop/paper.pdf', 'page': 0, 'start_index': 0})]
Number of requested results 4 is greater than number of elements in index 1, updating n_results = 1
========================
{'question': 'what is x?', 'answer': 'x = 3\n', 'sources': '/Users/user/Desktop/paper.pdf', 'source_documents': [Document(page_content='This is a sample paper!\r\nThis paper talks about x.\r\nIn this case, we are going to say that "x = 3"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper.pdf', 'start_index': 0})]}
========================
--------------> PDF2
[Document(page_content='This is another sample paper!\r\nThis paper talks about y.\r\nIn this case, we are going to say that "y = 5"', metadata={'source': '/Users/user/Desktop/paper2.pdf', 'page': 0, 'start_index': 0})]
Number of requested results 4 is greater than number of elements in index 2, updating n_results = 2
========================
{'question': 'what is y?', 'answer': 'y = 5\n', 'sources': '/Users/user/Desktop/paper2.pdf', 'source_documents': [Document(page_content='This is another sample paper!\r\nThis paper talks about y.\r\nIn this case, we are going to say that "y = 5"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper2.pdf', 'start_index': 0}), Document(page_content='This is a sample paper!\r\nThis paper talks about x.\r\nIn this case, we are going to say that "x = 3"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper.pdf', 'start_index': 0})]}
========================
```
## Reproduce:
Create a llm (OpenAI or AzureChatOpenAI):
1.)
```
llm = AzureChatOpenAI(
temperature=0,
openai_api_base=openai_api_base,
openai_api_version=openai_api_version,
openai_api_key=openai_api_key,
openai_api_type=openai_api_type,
deployment_name=openai_model
)
```
2.) Create embeddings:
```
embeddings = OpenAIEmbeddings(deployment=openai_embedding_model, chunk_size=1)
text_splitter = RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=20, length_function = len, add_start_index = True)
```
3.) Setup Chroma and Retriever
```
db = Chroma(embedding_function=embeddings)
# Again a note: if you replace `map_reduce` with `stuff` -- works perfectly!
qa = RetrievalQAWithSourcesChain.from_chain_type(llm=llm, chain_type="map_reduce", retriever=db.as_retriever(search_kwargs={"k": 4}), return_source_documents=True)
```
4.) Embed the files `db.add_documents()` and search!
```
NOTE: I have tried with multiple PDF loaders and the same problem exists even with plain text -- so this is not the issue
print("--------------> PDF1")
loader = PyPDFium2Loader('/Users/user/Desktop/paper.pdf')
docs = loader.load_and_split(text_splitter)
print(docs)
db.add_documents(docs)
text = 'what is x?'
result = qa({"question": text})
print("========================")
print(result)
print("========================")
print("--------------> PDF2")
loader = PyPDFium2Loader('/Users/user/Desktop/paper2.pdf')
docs = loader.load_and_split(text_splitter)
print(docs)
db.add_documents(docs)
text = 'what is y?'
result = qa({"question": text})
print("========================")
print(result)
print("========================")
```
## Key Summary:
* As you will see, the correct documents are retrieved -- it's just the injections/`map_reduce` that's causing a problem
* With `map_reduce` - if there is a single `Document()` it works fine -- for both `paper.pdf` and `paper2.pdf`
* With `map_reduce`, more than 1 document causes the issue.
* If you use the same embeddings and the same code with `stuff` chain it works perfectly with multiple documents.
## Files Attached for testing:
[paper.pdf](https://github.com/langchain-ai/langchain/files/12647756/paper.pdf)
[paper2.pdf](https://github.com/langchain-ai/langchain/files/12647750/paper2.pdf)
see a self-contained example here:
https://github.com/langchain-ai/langchain/issues/10735#issuecomment-1723595341
### Expected behavior
For `RetrievalQAWithSourcesChain.from_chain_type` to be able to inject the correct documents that it finds with `map_reduce` (similar to `stuff` which works):
```
--------------> PDF1
[Document(page_content='This is a sample paper!\r\nThis paper talks about x.\r\nIn this case, we are going to say that "x = 3"', metadata={'source': '/Users/user/Desktop/paper.pdf', 'page': 0, 'start_index': 0})]
Number of requested results 4 is greater than number of elements in index 1, updating n_results = 1
========================
{'question': 'what is x?', 'answer': 'x = 3\n', 'sources': '/Users/user/Desktop/paper.pdf', 'source_documents': [Document(page_content='This is a sample paper!\r\nThis paper talks about x.\r\nIn this case, we are going to say that "x = 3"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper.pdf', 'start_index': 0})]}
========================
--------------> PDF2
[Document(page_content='This is another sample paper!\r\nThis paper talks about y.\r\nIn this case, we are going to say that "y = 5"', metadata={'source': '/Users/user/Desktop/paper2.pdf', 'page': 0, 'start_index': 0})]
Number of requested results 4 is greater than number of elements in index 2, updating n_results = 2
========================
{'question': 'what is y?', 'answer': 'y = 5\n', 'sources': '/Users/user/Desktop/paper2.pdf', 'source_documents': [Document(page_content='This is another sample paper!\r\nThis paper talks about y.\r\nIn this case, we are going to say that "y = 5"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper2.pdf', 'start_index': 0}), Document(page_content='This is a sample paper!\r\nThis paper talks about x.\r\nIn this case, we are going to say that "x = 3"', metadata={'page': 0, 'source': '/Users/user/Desktop/paper.pdf', 'start_index': 0})]}
========================
``` | bug: map_reduce (RetrievalQAWithSourcesChain.from_chain_type) with embeddings - reproducible | https://api.github.com/repos/langchain-ai/langchain/issues/10735/comments | 6 | 2023-09-18T12:38:22Z | 2024-03-25T16:05:56Z | https://github.com/langchain-ai/langchain/issues/10735 | 1,900,847,417 | 10,735 |
[
"hwchase17",
"langchain"
]
| ### System Info
pydantic '1.10.0' or '2.3.0'
langchain '0.0.292'
python '3.11.5'
### Who can help?
@hwchase17
@agola11
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
from langchain.chains.openai_functions.openapi import get_openapi_chain
chain = get_openapi_chain("https://www.klarna.com/us/shopping/public/openai/v0/api-docs/")
chain("What are some options for a men's large blue button down shirt")
Error message when running with pydantic 1
Unable to parse spec from source https://chat-web3-plugin.alchemy.com/openapi.yaml error
Error message when running with pydantic 2
AttributeError: type object 'OpenAPISpec' has no attribute 'from_url'
### Expected behavior
normal execution without errors | get_openapi_chain throws error when running example from docs | https://api.github.com/repos/langchain-ai/langchain/issues/10733/comments | 18 | 2023-09-18T11:39:49Z | 2024-04-02T16:05:50Z | https://github.com/langchain-ai/langchain/issues/10733 | 1,900,747,180 | 10,733 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain-0.0.292, python-3, vsCode
### Who can help?
@hwchase17 @agola11
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
from langchain.schema import HumanMessage, SystemMessage
from langchain.chat_models import ChatOpenAI
import os
from dotenv import load_dotenv
from gptcache import cache
from gptcache.manager import get_data_manager,manager_factory, CacheBase, VectorBase
from gptcache.similarity_evaluation.distance import SearchDistanceEvaluation
from gptcache.embedding import Onnx
from gptcache.processor.pre import get_messages_last_content
from langchain.chat_models import ChatOpenAI
from gptcache.adapter.langchain_models import LangChainChat
import time
print("Cache loading.....")
onnx = Onnx()
cache_base = CacheBase('mysql', sql_url='your_mysql_url')
vector_base = VectorBase('faiss', dimension=128)
data_manager = get_data_manager(cache_base, vector_base)
cache.init(pre_embedding_func=get_messages_last_content,
data_manager=data_manager,
)
cache.set_openai_key()
load_dotenv()
def generate_res(user_input):
chat = LangChainChat(chat=ChatOpenAI(model_name=os.environ.get('GPT_MODEL'), temperature=0.7, openai_api_key=os.environ.get('OPEN_AI_API_KEY')))
prompt=""
start_time = time.time()
message = [
SystemMessage(content=prompt),
HumanMessage(content=user_input),
]
response = chat(message)
generate_res("hey there!")
print("Time consuming: {:.2f}s".format(time.time() - start_time))
### Expected behavior
I'm trying to store the cache into mysql db using chatopenai and gptcache. but getting this error:
super().__init__(**kwargs)
File "pydantic\main.py", line 339, in pydantic.main.BaseModel.__init__
File "pydantic\main.py", line 1066, in pydantic.main.validate_model
File "pydantic\fields.py", line 439, in pydantic.fields.ModelField.get_default
File "pydantic\utils.py", line 693, in pydantic.utils.smart_deepcopy
File "C:\Program Files\Python311\Lib\copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 146, in deepcopy
y = copier(x, memo)
^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 146, in deepcopy
y = copier(x, memo)
^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 146, in deepcopy
y = copier(x, memo)
^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 146, in deepcopy
y = copier(x, memo)
^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 146, in deepcopy
y = copier(x, memo)
^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 172, in deepcopy
y = _reconstruct(x, memo, *rv)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 271, in _reconstruct
state = deepcopy(state, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 146, in deepcopy
y = copier(x, memo)
^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 231, in _deepcopy_dict
y[deepcopy(key, memo)] = deepcopy(value, memo)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Python311\Lib\copy.py", line 161, in deepcopy
rv = reductor(4)
^^^^^^^^^^^
TypeError: cannot pickle 'module' object | data_manager not working with cache init for chatOpenAI | https://api.github.com/repos/langchain-ai/langchain/issues/10731/comments | 2 | 2023-09-18T10:15:38Z | 2023-12-25T16:05:54Z | https://github.com/langchain-ai/langchain/issues/10731 | 1,900,607,849 | 10,731 |
[
"hwchase17",
"langchain"
]
| In intermediateSteps, how to get the response status code if I using the tool request_get or post.. | Tool request how to know response status code | https://api.github.com/repos/langchain-ai/langchain/issues/10730/comments | 2 | 2023-09-18T09:15:46Z | 2023-12-25T16:06:00Z | https://github.com/langchain-ai/langchain/issues/10730 | 1,900,506,801 | 10,730 |
[
"hwchase17",
"langchain"
]
| ### Feature request
Switch to motor in the [MongoDBChatMessageHistory](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/memory/chat_message_histories/mongodb.py) class for asyncio support.
### Motivation
There is already a PR #10645 for async Document Loader for MongoDB, this way we can keep consistency in the mongodb clients.
### Your contribution
I already implemented it, if you want I can submit a PR. | async mongodb chat message history | https://api.github.com/repos/langchain-ai/langchain/issues/10729/comments | 3 | 2023-09-18T07:53:53Z | 2024-01-04T14:01:15Z | https://github.com/langchain-ai/langchain/issues/10729 | 1,900,370,220 | 10,729 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
ERROR: Exception in ASGI application
Traceback (most recent call last):
File "D:\Users\DHANKA01\vox-usecase-wrdm-nco-dev\src\backend\api.py", line 49, in nco_generate
result=final_run(Paths.prompt_lib,Paths.sdpk_path)
File "D:\Users\DHANKA01\vox-usecase-wrdm-nco-dev\src\backend\llm_run.py", line 126, in final_run
answer_resp,content_resp,question_covered=file_run(prompt_idx,index_temp,LLM_MODEL,EMBEDDING_MODEL,max_tokens,question_covered)
File "D:\Users\DHANKA01\vox-usecase-wrdm-nco-dev\src\backend\llm_run.py", line 71, in file_run
Ans_line, res_text=get_queryAnswer(llm_model,embedding_model,index_name,max_tokens,query,flag)
File "D:\Users\DHANKA01\vox-usecase-wrdm-nco-dev\src\backend\llm_run.py", line 43, in get_queryAnswer
output1=chain.run(question)
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\chains\base.py", line 290, in run
return self(args[0], callbacks=callbacks, tags=tags)[_output_key]
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\chains\base.py", line 166, in __call__
raise e
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\chains\base.py", line 160, in __call__
self._call(inputs, run_manager=run_manager)
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\chains\llm.py", line 92, in _call
response = self.generate([inputs], run_manager=run_manager)
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\chains\llm.py", line 102, in generate return self.llm.generate_prompt(
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\llms\base.py", line 141, in generate_prompt
return self.generate(prompt_strings, stop=stop, callbacks=callbacks, **kwargs)
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\llms\base.py", line 227, in generate
output = self._generate_helper(
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\llms\base.py", line 178, in _generate_helper
raise e
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\llms\base.py", line 165, in _generate_helper
self._generate(
File "D:\Users\DHANKA01\anaconda3\envs\pls_final\lib\site-packages\langchain\llms\base.py", line 527, in _generate else self._call(prompt, stop=stop, **kwargs)
TypeError: _call() got an unexpected keyword argument 'stop'
### Suggestion:
hi,
while running the same code with jupyter notebook, it's running fine but after integration with fastapi getting error.
from langchain.chains import LLMChain
chain=LLMChain(llm=llm,prompt=prompt)
chain.run(query)
Fastapi version=0.103.1
langchain version tried= 0.0.292,0.0.267,0.0.215
please suggest a fix for the same. | Issue: unexpected keyword argument 'stop' | https://api.github.com/repos/langchain-ai/langchain/issues/10723/comments | 8 | 2023-09-18T04:03:36Z | 2024-01-05T10:48:25Z | https://github.com/langchain-ai/langchain/issues/10723 | 1,900,123,795 | 10,723 |
[
"hwchase17",
"langchain"
]
| ```
agent_chain = initialize_agent(
tools,
llm,
agent=AgentType.CHAT_CONVERSATIONAL_REACT_DESCRIPTION,
verbose=True,
return_intermediate_steps=True,
handle_parsing_errors=True,
memory=memory
)
```
how to overwrite the PREFIX, SUFFIX and FORMAT_INSTRUCTIONS when I using the CHAT_CONVERSATIONAL_REACT_DESCRIPTION and creating agent by ```initialize_agent``` function? | How to custom prompt in agent type CHAT_CONVERSATIONAL_REACT_DESCRIPTION | https://api.github.com/repos/langchain-ai/langchain/issues/10721/comments | 6 | 2023-09-18T03:53:27Z | 2024-05-10T16:06:40Z | https://github.com/langchain-ai/langchain/issues/10721 | 1,900,116,688 | 10,721 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
I configured it to use the OpenAPI toolkit by merging various APIs, but I could not solve the problem of having to redefine the header at the request stage.
### Suggestion:
I'm not sure if there is a way to solve this in langchain or if I need to apply a separate module such as a request interceptor.
Thank you in advance for your reply :) | using the OpenAPI toolkit, override the headers for each request | https://api.github.com/repos/langchain-ai/langchain/issues/10719/comments | 4 | 2023-09-18T03:30:15Z | 2024-01-30T00:42:11Z | https://github.com/langchain-ai/langchain/issues/10719 | 1,900,100,967 | 10,719 |
[
"hwchase17",
"langchain"
]
| ### System Info
0.0.287 langchain
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Currently I have a Custom Tool like this:
```
def _handle_error(error: ToolException) -> str:
return (
"The following errors occurred during tool execution:"
+ error.args[0]
+ "please try another tool."
)
tools = [
Tool.from_function(
name="GetCurrentWeather",
func=get_current_weather,
description="useful for getting the current weather for a location",
return_direct=True,
handle_tool_error=_handle_error)]
```
It works unless I add the return_direct=True.
When I do return_direct= True instead of trying other tools it just returns this:
```
> Entering new AgentExecutor chain...
16 AgentAction --> Thought: The question is asking about the current weather in Uludağ.
17 Action: GetCurrentWeather
18 Action Input: Uludağ
19 Thought: The question is asking about the current weather in Uludağ.
20 Action: GetCurrentWeather
21 Action Input: Uludağ
22 Observation:The following errors occurred during tool execution:the location is not found. please try another tool.
23 > Finished chain.
24 The following errors occurred during tool execution:location is not found.please try another tool.
```
### Expected behavior
Let's see the same thing without the `return_direct=True`:
```
> Entering new AgentExecutor chain...
Thought: The question is asking about the current weather in Uludağ.
Action: GetCurrentWeather
Action Input: Uludağ
Observation:The following errors occurred during tool execution:location is not found. please try another tool.
I couldn't find the current weather for Uludağ. I should try another tool.
Action: DuckDuckSearch
Action Input: "Uludağ current weather"
Observation:Current Weather. 7:35 PM. 78° F. RealFeel® 74°. RealFeel Shade™ 74°. Air Quality Fair. Wind W 14 mph. Wind Gusts 24 mph. Sunny More Details. 7:14. —. This table gives the weather forecast for Uludag at the specific elevation of 2543 m. Our advanced weather models allow us to provide distinct weather forecasts for several elevations of Uludag. ..
```
So it goes for the other tools. It should be like this for return_direct.
It should return directly only if it doesn't get an error. | Custom Tool Return Direct BUT only if it doesnt raise a Tool Exception | https://api.github.com/repos/langchain-ai/langchain/issues/10714/comments | 4 | 2023-09-17T23:22:13Z | 2024-05-24T20:22:30Z | https://github.com/langchain-ai/langchain/issues/10714 | 1,899,959,947 | 10,714 |
[
"hwchase17",
"langchain"
]
| ### Issue with current documentation:
I cannot handle my custom_tool's raised errors.
I have this weatherTool:
```
class GetCurrentWeatherTool(BaseTool):
name = "get_current_weather"
description = "Get the current weather for a specified location"
args_schema: Type[WeatherInput] = WeatherInput
def _run(
self,
location: str,
run_manager: Optional[CallbackManagerForToolRun] = None
) -> str:
"""Use the tool."""
return get_current_weather(location)
async def _arun(
self,
location: str,
run_manager: Optional[AsyncCallbackManagerForToolRun] = None
) -> str:
"""Use the tool asynchronously."""
# You can implement the async version of the method here or raise a NotImplementedError if async is not supported
raise NotImplementedError("get_current_weather does not support async")
def _handle_error(error: ToolException) -> str:
return (
"The following errors occurred during tool execution:"
+ error.args[0]
+ "Location couldn't be found, please try another tool."
)
```
And I raise ToolException when the API call gets an error:
`raise ToolException("location has been found.")` inside of `get_current_weather`
Now I am adding this tool to my tools when I'm creating my agent:
```
self.weather = GetCurrentWeatherTool()
Tool(
name="Weather",
func=self.weather.run,
description="useful for weather related questions",
return_direct=True,
)
```
In the documentation [Custom Tools](https://python.langchain.com/docs/modules/agents/tools/custom_tools#handling-tool-errors) it explains how to add it and gives an example:
```
Tool.from_function(
func=search_tool1,
name="Search_tool1",
description=description,
handle_tool_error=True,
),
```
However I couldnt find the equal of that for Subclassing the BaseTool.
### Idea or request for content:
It would be great to give an example for Subclassed CustomTool(BaseTool) when defining the tools. | DOC: Custom Tool Error Handling is not Clear | https://api.github.com/repos/langchain-ai/langchain/issues/10710/comments | 4 | 2023-09-17T22:47:35Z | 2024-06-08T16:07:20Z | https://github.com/langchain-ai/langchain/issues/10710 | 1,899,949,896 | 10,710 |
[
"hwchase17",
"langchain"
]
| ### Feature request
The current MRKL `FORMAT_INSTRUCTIONS` prompt ([here](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/mrkl/prompt.py)) looks like this:
``` python
FORMAT_INSTRUCTIONS = """Use the following format:
Question: the input question you must answer
Thought: you should always think about what to do
Action: the action to take, should be one of [{tool_names}]
Action Input: the input to the action
Observation: the result of the action
... (this Thought/Action/Action Input/Observation can repeat N times)
Thought: I now know the final answer
Final Answer: the final answer to the original input question"""
```
Which is not very good- the similar tool selection prompts in other agents are MUCH better, and contain detailed formatting instructions.
This prompt should be brought in line with the likes of the [Structured Chat Prompt](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/structured_chat/prompt.py) which is much better, in a number of ways (clearer instructions, better formatting, etc.)
As always, I appreciate your hard work!
### Motivation
The motivation for this is to bring the prompt quality up throughout the project. Overall, there is probably some refactoring that could be done in this area to make it easier to centralize how prompts are used... I know that there's the hub now, but I am not sure if that fits this bill.
### Your contribution
I will try to contribute to this, but my workload is intense, and I don't have time currently... however, I have a number of things that I have modified in my own instance that I really should contribute back. 🥲 | Bring MRKL prompts in line with the other Agent prompts | https://api.github.com/repos/langchain-ai/langchain/issues/10709/comments | 2 | 2023-09-17T19:56:51Z | 2023-12-25T16:06:09Z | https://github.com/langchain-ai/langchain/issues/10709 | 1,899,899,705 | 10,709 |
[
"hwchase17",
"langchain"
]
| ### System Info
Hello everyone,
I am encountering an issue while trying to use Pinecone store in my Node.js environment.
When i try to store the vector in an existing index, I am receiving the following error:
PineconeArgumentError: The argument to upsert had type errors: argument must be array.
here is my code :
async function training(bot, documents) {
if (!bot || !bot.name) {
throw new Error('Invalid bot or bot name is missing.');
}
if (!documents || !Array.isArray(documents) || documents.length === 0) {
throw new Error('Invalid documents array.');
}
try {
const botName = bot.name;
console.log('Bot Name:', botName);
const keys = documents.map(url => path.basename(url));
console.log('Document keys:', keys);
const rawDocs = await chargerDocumentsDeS3(keys);
console.log('Raw documents retrieved:', rawDocs.length);
/* Split text into chunks */
const textSplitter = new RecursiveCharacterTextSplitter({
chunkSize: 1000,
chunkOverlap: 200,
});
let docs = await textSplitter.splitDocuments(rawDocs);
console.log('Number of split documents:', docs.length);
//console.log(docs)
const index = pinecone.Index( botName);
await PineconeStore.fromDocuments(docs, new OpenAIEmbeddings(), {
pineconeIndex:index,
maxConcurrency: 5, // Maximum number of batch requests to allow at once. Each batch is 1000 vectors.
});
console.log('Upload and processing complete.');
return { message: 'Upload and processing complete.' };
} catch (error) {
console.error('Error during upload and processing:', error);
throw new Error('Something went wrong during upload and processing.');
}
}
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [X] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
async function training(bot, documents) {
if (!bot || !bot.name) {
throw new Error('Invalid bot or bot name is missing.');
}
if (!documents || !Array.isArray(documents) || documents.length === 0) {
throw new Error('Invalid documents array.');
}
try {
const botName = bot.name;
console.log('Bot Name:', botName);
const keys = documents.map(url => path.basename(url));
console.log('Document keys:', keys);
const rawDocs = await chargerDocumentsDeS3(keys);
console.log('Raw documents retrieved:', rawDocs.length);
/* Split text into chunks */
const textSplitter = new RecursiveCharacterTextSplitter({
chunkSize: 1000,
chunkOverlap: 200,
});
let docs = await textSplitter.splitDocuments(rawDocs);
console.log('Number of split documents:', docs.length);
//console.log(docs)
const index = pinecone.Index( botName);
await PineconeStore.fromDocuments(docs, new OpenAIEmbeddings(), {
pineconeIndex:index,
maxConcurrency: 5, // Maximum number of batch requests to allow at once. Each batch is 1000 vectors.
});
console.log('Upload and processing complete.');
return { message: 'Upload and processing complete.' };
} catch (error) {
console.error('Error during upload and processing:', error);
throw new Error('Something went wrong during upload and processing.');
}
}
### Expected behavior
I want a solution | PineconeArgumentError: The argument to upsert had type errors: argument must be array. | https://api.github.com/repos/langchain-ai/langchain/issues/10708/comments | 16 | 2023-09-17T19:11:15Z | 2024-05-17T09:57:20Z | https://github.com/langchain-ai/langchain/issues/10708 | 1,899,886,459 | 10,708 |
[
"hwchase17",
"langchain"
]
| ### Issue with current documentation:
In the **Multiple Retrieval Sources** page (https://python.langchain.com/docs/use_cases/question_answering/how_to/multiple_retrieval)
There is a link to **Langchain Expression Language** (https://python.langchain.com/docs/use_cases/)
Selecting the link results in the following message:
**Page Not Found**
We could not find what you were looking for.
Please contact the owner of the site that linked you to the original URL and let them know their link is broken.
### Idea or request for content:
The link is missing. Broken Link needs to be fixed. | DOC: Link to Langchain Expression Language link results in Page not Found | https://api.github.com/repos/langchain-ai/langchain/issues/10705/comments | 2 | 2023-09-17T17:59:32Z | 2023-12-25T16:06:15Z | https://github.com/langchain-ai/langchain/issues/10705 | 1,899,865,868 | 10,705 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Hi , I want to know how it would be possible to filter with the Sitemap Loader updated pages of 2023 from a website. I do not want to get all the website but only 2023 updated pages.
Here is my code:
`# fixes a bug with asyncio and jupyter
import nest_asyncio
nest_asyncio.apply()
from langchain.document_loaders.sitemap import SitemapLoader
loader = SitemapLoader(
"",
filter_urls=[""]
)
docs = loader.load()`
How can it be possible in it please?
Thank you,
### Suggestion:
_No response_ | Issue:Sitemap Loader filter updated pages | https://api.github.com/repos/langchain-ai/langchain/issues/10703/comments | 4 | 2023-09-17T15:06:35Z | 2023-12-25T16:06:19Z | https://github.com/langchain-ai/langchain/issues/10703 | 1,899,814,125 | 10,703 |
[
"hwchase17",
"langchain"
]
| I know the intermediate steps can get the list of list of action and observation tuples.
If I use the Aiplugin tool to create the api call. Can I get the detail of request and response that agent make?
I want to know what request and response that agent make for debugging, because agent not work correctly often.
| get more detail fo intermediate steps | https://api.github.com/repos/langchain-ai/langchain/issues/10702/comments | 2 | 2023-09-17T14:48:26Z | 2023-12-25T16:06:24Z | https://github.com/langchain-ai/langchain/issues/10702 | 1,899,808,521 | 10,702 |
[
"hwchase17",
"langchain"
]
| ### Issue with current documentation:
There is now no setup.py file in the llama-cpp-python library
I tried installing using these instructions:
Llama.cpp
https://python.langchain.com/docs/integrations/llms/llamacpp
The installation was done as described in the paragraph:
Windows installation
### Idea or request for content:
_No response_ | DOC: <Please write a comprehensive title after the 'DOC: Please update the instructions on the langchain site for istallation llama-cpp-python on Windows with GPU NVidia Support' prefix> | https://api.github.com/repos/langchain-ai/langchain/issues/10697/comments | 2 | 2023-09-17T08:24:35Z | 2023-12-25T16:06:30Z | https://github.com/langchain-ai/langchain/issues/10697 | 1,899,702,496 | 10,697 |
[
"hwchase17",
"langchain"
]
| ### System Info
0.0.287
### Who can help?
People who knows to create custom tools/ toolkits for custom agents.
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Here is the GetCurrentWeatherTool and I'm trying to give this in the tools:
```
class WeatherInput(BaseModel):
location: str = Field(..., description="The location for which to get the current weather")
class GetCurrentWeatherTool(BaseTool):
name = "get_current_weather"
description = "Get the current weather for a specified location"
args_schema: Type[WeatherInput] = WeatherInput
def _run(
self,
location: str,
run_manager: Optional[CallbackManagerForToolRun] = None
) -> str:
"""Use the tool."""
return get_current_weather(location)
async def _arun(
self,
location: str,
run_manager: Optional[AsyncCallbackManagerForToolRun] = None
) -> str:
"""Use the tool asynchronously."""
# You can implement the async version of the method here or raise a NotImplementedError if async is not supported
raise NotImplementedError("get_current_weather does not support async")
```
Tools:
```
self.tools = [
# Tool(
# name="DuckDuckSearch",
# func=DuckDuckGoSearchRun().run,
# description="useful for when you need to search the web, cheap"
# ),
# Tool(
# name="Search",
# func=self.search.run,
# description="useful for when you need to answer questions about current events, use it if DuckDucSearch doesn't give you the answer you need",
# ),
GetCurrentWeatherTool
]
self.tool_names = [tool.name for tool in self.tools]
```
But I'm getting this error:
```
self.tool_names = [tool.name for tool in self.tools]
AttributeError: type object 'GetCurrentWeatherTool' has no attribute 'name'
```
### Expected behavior
I was expecting to use the tool with an example query like _How is the weather in London right now?_ and I thought It would call the actual get_current_weather function. | Custom Agent Doesn't Accept Custom Tool | https://api.github.com/repos/langchain-ai/langchain/issues/10694/comments | 1 | 2023-09-17T01:02:39Z | 2023-09-17T14:22:44Z | https://github.com/langchain-ai/langchain/issues/10694 | 1,899,615,677 | 10,694 |
[
"hwchase17",
"langchain"
]
| ### Issue with current documentation:
The parameters listed in the [DallEAPIWrapper documentation](https://api.python.langchain.com/en/latest/utilities/langchain.utilities.dalle_image_generator.DallEAPIWrapper.html) are incorrect for:
- n
- open_api_key
- size
### Idea or request for content:
See [OpenAI API reference](https://platform.openai.com/docs/api-reference/images/create) for the correct parameter definitions | DOC: parameters listed in the DallEAPIWrapper documentation are incorrect | https://api.github.com/repos/langchain-ai/langchain/issues/10692/comments | 1 | 2023-09-16T23:14:30Z | 2023-10-28T23:30:05Z | https://github.com/langchain-ai/langchain/issues/10692 | 1,899,594,024 | 10,692 |
[
"hwchase17",
"langchain"
]
| ### Feature request
The [`openai.Image.create()` endpoint](https://platform.openai.com/docs/api-reference/images/create) has an `n` parameter that allows you to generate 1-10 images. However, I used the looked at the langchain `DallEAPIWrapper()` and only got one image URL. After looking at its [source code](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/utilities/dalle_image_generator.py), I see that the langchain `DallEAPIWrapper()` only returns the first image URL. This should be to easy to fix.
### Motivation
It would be great to be able to generate multiple images from a single input and API call.
### Your contribution
I can submit a pull request to modify `def _dalle_image_url` method to return all image URLs. | DallEAPIWrapper currently only returns the first image URL, even if `n` parameter is set to >1 | https://api.github.com/repos/langchain-ai/langchain/issues/10691/comments | 2 | 2023-09-16T23:05:37Z | 2023-11-26T21:52:32Z | https://github.com/langchain-ai/langchain/issues/10691 | 1,899,592,343 | 10,691 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain ==0.0.287
python
fastapi
### Who can help?
_No response_
### Information
- [x] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [X] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [x] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```
class CustomOutputParser(AgentOutputParser):
def parse(self, llm_output: str) -> Union[AgentAction, AgentFinish]:
# Check if agent should finish
if "Final Answer:" in llm_output:
final_answer = llm_output.split("Final Answer:")[-1].strip()
# Ensure the final answer doesn't exceed 160 characters
if len(final_answer) > 160:
final_answer = final_answer[:157] + '...'
return AgentFinish(
# Return values is generally always a dictionary with a single `output` key
# It is not recommended to try anything else at the moment :)
return_values={"output": final_answer},
log=llm_output,
)
# Parse out the action and action input
regex = r"Action\s*\d*\s*:(.*?)\nAction\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
match = re.search(regex, llm_output, re.DOTALL)
if not match:
raise OutputParserException(
f"Could not parse LLM output: `{llm_output}`")
action = match.group(1).strip()
action_input = match.group(2)
# Return the action and action input
return AgentAction(tool=action, tool_input=action_input.strip(" ").strip('"'), log=llm_output)
```
Here is the OutputParser that I got from the tutorial of [Custom LLM Agent ](https://python.langchain.com/docs/modules/agents/how_to/custom_llm_agent) However it fails sometimes to parse the output.
I only add this part:
```
# Ensure the final answer doesn't exceed 160 characters
if len(final_answer) > 160:
final_answer = final_answer[:157] + '...'
return AgentFinish(
# Return values is generally always a dictionary with a single `output` key
# It is not recommended to try anything else at the moment :)
return_values={"output": final_answer},
log=llm_output,
)
```
To ensure that the output is 160 char however somehow it affects the output and the system throws an error on the outputparser like this:
```
chain/error] [1:chain:AgentExecutor] [4.40s] Chain run errored with error:
"OutputParserException('Could not parse LLM output: `I found the closing price of Nvidia stock: $455.72 on 09/08/2023. It gained 0.27% compared to the opening price.`')"
Traceback (most recent call last):
File "/Users/egehosgungor/Desktop/Side_Hustle/smsbotu/backend/app/llm/langchain_custom_agent_deneme.py", line 160, in <module>
asyncio.run(main())
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/asyncio/runners.py", line 44, in run
return loop.run_until_complete(main)
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/asyncio/base_events.py", line 647, in run_until_complete
return future.result()
File "/Users/egehosgungor/Desktop/Side_Hustle/smsbotu/backend/app/llm/langchain_custom_agent_deneme.py", line 155, in main
response = await agent_executor.arun("Nvidia hissesi kapanis")
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/chains/base.py", line 561, in arun
await self.acall(
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/chains/base.py", line 361, in acall
raise e
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/chains/base.py", line 355, in acall
await self._acall(inputs, run_manager=run_manager)
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/agents/agent.py", line 1171, in _acall
next_step_output = await self._atake_next_step(
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/agents/agent.py", line 1026, in _atake_next_step
raise e
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/agents/agent.py", line 1015, in _atake_next_step
output = await self.agent.aplan(
File "/Users/egehosgungor/miniconda3/envs/fastapi/lib/python3.9/site-packages/langchain/agents/agent.py", line 458, in aplan
return self.output_parser.parse(output)
File "/Users/egehosgungor/Desktop/Side_Hustle/smsbotu/backend/app/llm/langchain_custom_agent_deneme.py", line 123, in parse
raise OutputParserException(
langchain.schema.output_parser.OutputParserException: Could not parse LLM output: `I found the closing price of Nvidia stock: $455.72 on 09/08/2023. It gained 0.27% compared to the opening price.`
```
Question 1: Why this happens when I only truncate the finalAnswer output ? How does it has an affect on the system?
Question 2: I would like to wrap my output parser with something like https://python.langchain.com/docs/modules/model_io/output_parsers/retry so If it fails to parse the output it should come to an llmchain and that llmchain's only job should be parsing correctly OR if there is no answer giving the correct output as "I couldnt find the answer for {x}" etc.
### Expected behavior
I was expecting to truncating the final_answer only will have an effect on the output's character count however it does affect the whole system.
How can I connect the outputParsers? | How to create CustomAgentOutputParser that uses Retry parser | https://api.github.com/repos/langchain-ai/langchain/issues/10689/comments | 1 | 2023-09-16T21:57:11Z | 2023-09-17T14:23:00Z | https://github.com/langchain-ai/langchain/issues/10689 | 1,899,578,553 | 10,689 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Hey, I am trying to follow various approaches online to create a simple chat interface in streamlit to chat with your pandas dataframe. I always face the issue, that I cant extract the full description of the steps. One example is the following code:
```
from langchain.agents import AgentType
from langchain.agents import create_pandas_dataframe_agent
from langchain.callbacks import StreamlitCallbackHandler
from langchain.chat_models import AzureChatOpenAI
if prompt := st.chat_input(placeholder = " "):
st.session_state.messages.append({"role": "user", "content": prompt})
st.chat_message("user").write(prompt)
llm = AzureChatOpenAI()
pandas_df_agent =
create_pandas_dataframe_agent(llm,df,verbose=True,agent_type=AgentType.OPENAI_FUNCTIONS,handle_parsing_errors=True)
with st.chat_message("assistant"):
st_cb = StreamlitCallbackHandler(st.container(), expand_new_thoughts=False)
response = pandas_df_agent.run(st.session_state.messages, callbacks=[st_cb])
st.session_state.messages.append({"role": "assistant", "content": response})
st.write(response)
```
However if the steps executed are too long it truncates the message as seen below?
Anyone who knows how to fix this, or have seen a solution where its no the case?
### Suggestion:
I assume its something to do with the StreamlitCallbackHandler, outputting to the st.container. Maybe I need to change the properties of that container or the settings in StreamlitCallbackHandler? | Issue: How to get Pandas agent to return and write full steps to a streamlit chat | https://api.github.com/repos/langchain-ai/langchain/issues/10688/comments | 3 | 2023-09-16T21:38:22Z | 2023-12-25T16:06:34Z | https://github.com/langchain-ai/langchain/issues/10688 | 1,899,574,753 | 10,688 |
[
"hwchase17",
"langchain"
]
| ### System Info
python==3.8
langchain==217
### Who can help?
_No response_
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [X] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [X] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
I'm testing one of the tools that is being used in the zero shot agent, which is making an external api call and get json response. I'm trying to use LLMRequestsChain, and this chain needs an input as dictionary with query and url keys to make a call.
`inputs = {
"query": user_message,
"url": f"{endpoint}/user/{mxUser}/accounts?page=1&records_per_page=10"
}
result = agent_chain.zero_shot_agent_chain().run({"question": inputs})
`
Once the dict has been passed to the tool which is the LLMRequestsChain, It throws InvalidSchema exception saying `No connection adapters were found for '{\n "query": "what\'s my balance?",\n "url": "targeted url"\n}'`
### Expected behavior
looks like the dict has been passed to the tool as a string. Is there any workaround or any idea to work with? | Is there a way to pass Dictionary as an Action input when you use Zero Shot Agent? | https://api.github.com/repos/langchain-ai/langchain/issues/10681/comments | 3 | 2023-09-16T19:21:00Z | 2024-07-29T13:02:07Z | https://github.com/langchain-ai/langchain/issues/10681 | 1,899,542,362 | 10,681 |
[
"hwchase17",
"langchain"
]
| *I may take a stab at fixing this, but don't have the time now, sharing my research notes*
In my iMessage DB, many of my messages have null ```text```, because the content is now encoded in ```attributedBody```. Here is the research I did in case it helps others
Discussion of the issue:
https://www.reddit.com/r/osx/comments/uevy32/comment/j5ifbku/?utm_source=share&utm_medium=web2x&context=3
Encoding described in Stack Overflow:
https://stackoverflow.com/questions/75330393/how-can-i-read-the-attributedbody-column-in-macos-imessage-database/75330394#75330394
Code that needs to change:
https://github.com/langchain-ai/langchain/blob/116cc7998cac563894bdf0037259db767882b55c/libs/langchain/langchain/chat_loaders/imessage.py#L63C41-L63C53 | iMessageChatLoader doesn't load new text conent , due to ChatHistory schema change | https://api.github.com/repos/langchain-ai/langchain/issues/10680/comments | 5 | 2023-09-16T15:29:29Z | 2024-02-11T16:14:16Z | https://github.com/langchain-ai/langchain/issues/10680 | 1,899,473,938 | 10,680 |
[
"hwchase17",
"langchain"
]
| ### Issue with current documentation:
Here is the documentation that I'm talking: https://python.langchain.com/docs/modules/agents/how_to/custom_llm_agent
The problem with is that its not returning the actual final answer but the observation.
Normally it should return this as written:
```
> Entering new AgentExecutor chain...
Thought: I need to find out the population of Canada in 2023
Action: Search
Action Input: Population of Canada in 2023
Observation:The current population of Canada is 38,658,314 as of Wednesday, April 12, 2023, based on Worldometer elaboration of the latest United Nations data. I now know the final answer
Final Answer: Arrr, there be 38,658,314 people livin' in Canada as of 2023!
> Finished chain.
"Arrr, there be 38,658,314 people livin' in Canada as of 2023!"
```
But stucks here:
```
> Entering new AgentExecutor chain...
Thought: I need to search for the current population of Canada.
Action: Search
Action Input: "current population of Canada 2023"
Observation:38,781,291 people
```
It should be related to the output parser:
```
class CustomOutputParser(AgentOutputParser):
def parse(self, llm_output: str) -> Union[AgentAction, AgentFinish]:
# Check if agent should finish
if "Observation:" in llm_output:
# Parse the observation value
observation_value = llm_output.split("Observation:")[-1].strip()
# Construct a final answer based on the observation
final_answer = f"Arrr, there be {observation_value} people livin' in Canada as of 2023!"
# Create a log with the final answer
final_log = llm_output + f"\nThought: I now know the final answer\nFinal Answer: {final_answer}"
return AgentFinish(
return_values={"output": final_answer},
log=final_log,
)
# Parse out the action and action input
regex = r"Action\s*\d*\s*:(.*?)\nAction\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
match = re.search(regex, llm_output, re.DOTALL)
if not match:
raise OutputParserException(f"Could not parse LLM output: `{llm_output}`")
action = match.group(1).strip()
action_input = match.group(2)
# Return the action and action input
return AgentAction(tool=action, tool_input=action_input.strip(" ").strip('"'), log=llm_output)
output_parser = CustomOutputParser()
```
How to change that
```
# Construct a final answer based on the observation
final_answer = f"Arrr, there be {observation_value} people livin' in Canada as of 2023!"
```
So it can call one last time the LLMChain to sumarize the answer for us?
### Idea or request for content:
It would be great to update the Custom LLM Agent Tutorial. | DOC: Custom LLM Agent Tutorial is not working. | https://api.github.com/repos/langchain-ai/langchain/issues/10679/comments | 2 | 2023-09-16T14:11:20Z | 2023-09-16T14:32:43Z | https://github.com/langchain-ai/langchain/issues/10679 | 1,899,449,952 | 10,679 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain 0.0.285
python 3.11.4
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [X] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Try to load https://huggingface.co/datasets/hotpot_qa/viewer/fullwiki/validation
from langchain.document_loaders import HuggingFaceDatasetLoader
dataset_name = "hotpot_qa"
page_content_column = "context"
name = "fullwiki"
loader = HuggingFaceDatasetLoader(dataset_name, page_content_column, name)
docs = loader.load()
---------------------------------------------------------------------------
ValidationError Traceback (most recent call last)
[/Users/deanchanter/Documents/GitHub/comma-chameleons/hello_doc_read.ipynb](https://file+.vscode-resource.vscode-cdn.net/Users/deanchanter/Documents/GitHub/comma-chameleons/hello_doc_read.ipynb) Cell 1 line 8
[4](vscode-notebook-cell:/Users/deanchanter/Documents/GitHub/comma-chameleons/hello_doc_read.ipynb#Y125sZmlsZQ%3D%3D?line=3) name = "fullwiki"
[7](vscode-notebook-cell:/Users/deanchanter/Documents/GitHub/comma-chameleons/hello_doc_read.ipynb#Y125sZmlsZQ%3D%3D?line=6) loader = HuggingFaceDatasetLoader(dataset_name, page_content_column, name)
----> [8](vscode-notebook-cell:/Users/deanchanter/Documents/GitHub/comma-chameleons/hello_doc_read.ipynb#Y125sZmlsZQ%3D%3D?line=7) docs = loader.load()
File [~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/document_loaders/hugging_face_dataset.py:87](https://file+.vscode-resource.vscode-cdn.net/Users/deanchanter/Documents/GitHub/comma-chameleons/~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/document_loaders/hugging_face_dataset.py:87), in HuggingFaceDatasetLoader.load(self)
85 def load(self) -> List[Document]:
86 """Load documents."""
---> 87 return list(self.lazy_load())
File [~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/document_loaders/hugging_face_dataset.py:76](https://file+.vscode-resource.vscode-cdn.net/Users/deanchanter/Documents/GitHub/comma-chameleons/~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/document_loaders/hugging_face_dataset.py:76), in HuggingFaceDatasetLoader.lazy_load(self)
59 raise ImportError(
60 "Could not import datasets python package. "
61 "Please install it with `pip install datasets`."
62 )
64 dataset = load_dataset(
65 path=self.path,
66 name=self.name,
(...)
73 num_proc=self.num_proc,
74 )
---> 76 yield from (
77 Document(
78 page_content=row.pop(self.page_content_column),
79 metadata=row,
80 )
81 for key in dataset.keys()
82 for row in dataset[key]
83 )
File [~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/document_loaders/hugging_face_dataset.py:77](https://file+.vscode-resource.vscode-cdn.net/Users/deanchanter/Documents/GitHub/comma-chameleons/~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/document_loaders/hugging_face_dataset.py:77), in <genexpr>(.0)
59 raise ImportError(
60 "Could not import datasets python package. "
61 "Please install it with `pip install datasets`."
62 )
64 dataset = load_dataset(
65 path=self.path,
66 name=self.name,
(...)
73 num_proc=self.num_proc,
74 )
76 yield from (
---> 77 Document(
78 page_content=row.pop(self.page_content_column),
79 metadata=row,
80 )
81 for key in dataset.keys()
82 for row in dataset[key]
83 )
File [~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/load/serializable.py:75](https://file+.vscode-resource.vscode-cdn.net/Users/deanchanter/Documents/GitHub/comma-chameleons/~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/langchain/load/serializable.py:75), in Serializable.__init__(self, **kwargs)
74 def __init__(self, **kwargs: Any) -> None:
---> 75 super().__init__(**kwargs)
76 self._lc_kwargs = kwargs
File [~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/pydantic/main.py:341](https://file+.vscode-resource.vscode-cdn.net/Users/deanchanter/Documents/GitHub/comma-chameleons/~/Documents/GitHub/comma-chameleons/env/lib/python3.11/site-packages/pydantic/main.py:341), in pydantic.main.BaseModel.__init__()
ValidationError: 1 validation error for Document
page_content
str type expected (type=type_error.str)
### Expected behavior
Either extend class to handle more types of data or update docs. Will to do a PR to extend if your open. | HuggingFace Data Loader fails when context is not str | https://api.github.com/repos/langchain-ai/langchain/issues/10674/comments | 15 | 2023-09-16T10:49:37Z | 2023-11-29T03:33:18Z | https://github.com/langchain-ai/langchain/issues/10674 | 1,899,393,311 | 10,674 |
[
"hwchase17",
"langchain"
]
| ### System Info
The code snipped will execute. However, when changing the chunk size down, executing again, and then changing the chunk size up again, the chunk size will stay on the low value.
It seems that the chunk size is sometimes saved, even though the objects get ge-instantiated.
I ran the code snipped in a colab file.
versions:
chromadb-0.4.10-py3-none-any.whl
langchain-0.0.291-py3-none-any.whl
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Chroma
doc = Document(page_content=pdf_texts[0][1], metadata={})
def storeDocument(doc):
text_splitter4 = RecursiveCharacterTextSplitter(chunk_size = 400, chunk_overlap = 5)
all_splits4 = text_splitter4.split_documents([doc])
vectorStore = Chroma.from_documents(documents=all_splits4, embedding=OpenAIEmbeddings(openai_api_key="..."))
return vectorStore
storeDocument(doc).similarity_search_with_score("Where is the fund's geographic location?",k=10)
```
### Expected behavior
I would expect the chunk size to change the output of similarity matching. | RecursiveCharacterTextSplitter immutably memorizes chunk size | https://api.github.com/repos/langchain-ai/langchain/issues/10673/comments | 2 | 2023-09-16T10:06:24Z | 2023-09-16T10:25:46Z | https://github.com/langchain-ai/langchain/issues/10673 | 1,899,381,260 | 10,673 |
[
"hwchase17",
"langchain"
]
| ### Feature request
When calling from_documents directly from vectorstore I do this:
```
vector_db = Milvus.from_documents(
docs,
embeddings,
connection_args={
"host": config.MILVUS_HOST,
"port": config.MILVUS_PORT,
},
collection_name = collection_name,
)
```
This argument is not accepted:
`partition_name = 'tes'`
If you deep dive inside the `from_documents` to `from_text` in `milvus.py` you notice that uses `add_texts` method, which also is not accepting `partition_name`. This uses the `add_texts` uses `collection.py` - insert from pymilvus orm that does accept `partition_name` as an optional argument:
```
def insert(
self,
data: Union[List, pd.DataFrame, Dict],
partition_name: Optional[str] = None,
timeout: Optional[float] = None,
**kwargs,
) -> MutationResult:`
``
### Motivation
Having a way of inserting directly in a partition from langchain to Milvus
### Your contribution
To add the "partition_name" attribute as optional in the from_documents somehow | Milvus - Pymilvus insert is not offering partition_name as optional argument | https://api.github.com/repos/langchain-ai/langchain/issues/10671/comments | 1 | 2023-09-16T09:53:17Z | 2023-12-25T16:06:44Z | https://github.com/langchain-ai/langchain/issues/10671 | 1,899,377,692 | 10,671 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain==0.0.292
OS Windows10
Python 3.11
### Who can help?
Probably @hwchase17 @agola11
### Information
- [X] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [X] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Example code for Q/A chain with "Map Re-Rank" following the official tutorials:
```
precise_chat_model = ChatOpenAI(
model_name='gpt-3.5-turbo',
temperature=0,
openai_api_key=OPENAI_API_KEY
)
qa_chain: MapRerankDocumentsChain = load_qa_chain(
llm=precise_chat_model,
chain_type='map_rerank',
verbose=True,
return_intermediate_steps=True
)
question = 'Question'
query = {'input_documents': pages, 'question': question}
answer = qa_chain(query, return_only_outputs=False)
```
Full example with PDF that raises Exception all the time: https://gist.github.com/ton77v/eb5b90e72b1652ebccee86ac80b1e01f
Every time I use this chain with any page the UserWarning appears:
```
UserWarning: The apply_and_parse method is deprecated, instead pass an output parser directly to LLMChain.
```
And for the specific documents (usually when 10+ pages) the ValueError is raised upon finishing the chain.
* Gist above raises this error all the time!
```
File "...site-packages\langchain\output_parsers\regex.py", line 35, in parse
raise ValueError(f"Could not parse output: {text}")
ValueError: Could not parse output: Code execution in Ethereum....
```
### Expected behavior
I expect to get an answer without any Exceptions and Warnings | MapRerankDocumentsChain UserWarning & ValueError | https://api.github.com/repos/langchain-ai/langchain/issues/10670/comments | 7 | 2023-09-16T09:01:39Z | 2024-02-13T16:12:12Z | https://github.com/langchain-ai/langchain/issues/10670 | 1,899,364,157 | 10,670 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
The effect of tools with return_direct=False for CHAT_CONVERSATIONAL_REACT_AGENT only causes the LLM to generate a final answer based on tool observation, but never in another tool invocation.
The same is not happening with CONVERSATIONAL_REACT_AGENT which seems able to generate new tool queries after the first one.
Is this a feature that can be fixed simply acting on the agent policy prompt, or is there a better way to enable such feature?
Thank you very much.
### Suggestion:
_No response_ | CHAT_CONVERSATIONAL_REACT_AGENT never uses more than 1 tool per turn. | https://api.github.com/repos/langchain-ai/langchain/issues/10669/comments | 4 | 2023-09-16T08:50:55Z | 2023-12-25T16:06:50Z | https://github.com/langchain-ai/langchain/issues/10669 | 1,899,361,074 | 10,669 |
[
"hwchase17",
"langchain"
]
| ### System Info
Hello,
I noticed that the `AzureOpenAI` is missing from the latest release. Now we kind of have to create our own custom class. Is this the direction of the project?
Thank you
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [X] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
from langchain.llms import AzureOpenAI
no logner availiable
no possibility to add engine or deployment id, nor the possibility to add extra headers
### Expected behavior
from langchain.llms import AzureOpenAI
os.environ["OPENAI_API_TYPE"] = "azure"
os.environ["OPENAI_API_BASE"] = "https://<your-endpoint.openai.azure.com/"
os.environ["OPENAI_API_KEY"] = "your AzureOpenAI key"
os.environ["OPENAI_API_VERSION"] = "2023-05-15"
embeddings = OpenAIEmbeddings(deployment="your-embeddings-deployment-name") | AzureOpenAI is missing | https://api.github.com/repos/langchain-ai/langchain/issues/10664/comments | 2 | 2023-09-15T23:52:55Z | 2023-12-25T16:06:55Z | https://github.com/langchain-ai/langchain/issues/10664 | 1,899,214,053 | 10,664 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain == 292
### Who can help?
@hwchase17
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Here is my code:
```
agent_analytics_node = create_pandas_dataframe_agent(
llm,
df,
verbose=True,
reduce_k_below_max_tokens=True,
max_execution_time = 20,
early_stopping_method="generate",
)
tool_analytics_node = Tool(
name='Analytics Node',
func=agent_analytics_node.run)
tools = [tool_analytics_node]
chat_agent = ConversationalChatAgent.from_llm_and_tools(llm=llm, tools=tools)
executor = AgentExecutor.from_agent_and_tools(
agent=chat_agent,
tools=tools,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
verbose=True,
)
with st.chat_message("assistant"):
st_cb = StreamlitCallbackHandler(st.container(), expand_new_thoughts=False)
response = executor(prompt, callbacks=[st_cb])
```
here output from the agent: ```
> Entering new AgentExecutor chain...
Thought: The question seems to be asking for the sentiment polarity of the 'survey_comment' column in the dataframe. The sentiment polarity is a measure that lies between -1 and 1. Negative values indicate negative sentiment and positive values indicate positive sentiment. The TextBlob library in Python can be used to calculate sentiment polarity. However, before applying the TextBlob function, we need to ensure that the TextBlob library is imported. Also, the 'dropna()' function is used to remove any NaN values in the 'survey_comment' column before applying the TextBlob function.
Action: python_repl_ast
Action Input: import TextBlob
Observation: ModuleNotFoundError: No module named 'TextBlob'
Thought:The TextBlob library is not imported. I need to import it from textblob module.
Action: python_repl_ast
Action Input: from textblob import TextBlob
Observation:
Thought:Now that the TextBlob library is imported, I can apply it to the 'survey_comment' column to calculate the sentiment polarity.
Action: python_repl_ast
Action Input: df['survey_comment'].dropna().apply(lambda x: TextBlob(x).sentiment.polarity)
Observation: NameError: name 'TextBlob' is not defined
```
### Expected behavior
agent should be able to install python packages. | AgentExecutor and ModuleNotFoundError/NameError | https://api.github.com/repos/langchain-ai/langchain/issues/10661/comments | 2 | 2023-09-15T21:55:52Z | 2023-12-25T16:06:59Z | https://github.com/langchain-ai/langchain/issues/10661 | 1,899,121,900 | 10,661 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain == 292
### Who can help?
@hwchase17
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Here is my code:
```
agent_analytics_node = create_pandas_dataframe_agent(
llm,
df,
verbose=True,
reduce_k_below_max_tokens=True,
max_execution_time = 20,
early_stopping_method="generate",
)
tool_analytics_node = Tool(
name='Analytics Node',
func=agent_analytics_node.run)
tools = [tool_analytics_node]
chat_agent = ConversationalChatAgent.from_llm_and_tools(llm=llm, tools=tools)
executor = AgentExecutor.from_agent_and_tools(
agent=chat_agent,
tools=tools,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
verbose=True,
)
with st.chat_message("assistant"):
st_cb = StreamlitCallbackHandler(st.container(), expand_new_thoughts=False)
response = executor(prompt, callbacks=[st_cb])
```
here output from the agent:```
> Entering new AgentExecutor chain...
Thought: The question seems to be asking for the sentiment polarity of the 'survey_comment' column in the dataframe. The sentiment polarity is a measure that lies between -1 and 1. Negative values indicate negative sentiment and positive values indicate positive sentiment. The TextBlob library in Python can be used to calculate sentiment polarity. However, before applying the TextBlob function, we need to ensure that the TextBlob library is imported. Also, the 'dropna()' function is used to remove any NaN values in the 'survey_comment' column before applying the TextBlob function.
Action: python_repl_ast
Action Input: import TextBlob
Observation: ModuleNotFoundError: No module named 'TextBlob'
Thought:The TextBlob library is not imported. I need to import it from textblob module.
Action: python_repl_ast
Action Input: from textblob import TextBlob
Observation:
Thought:Now that the TextBlob library is imported, I can apply it to the 'survey_comment' column to calculate the sentiment polarity.
Action: python_repl_ast
Action Input: df['survey_comment'].dropna().apply(lambda x: TextBlob(x).sentiment.polarity)
Observation: NameError: name 'TextBlob' is not defined
```
### Expected behavior
agent should be able to install python packages | python_repl_ast and package import (ModuleNotFoundError and NameError) | https://api.github.com/repos/langchain-ai/langchain/issues/10660/comments | 4 | 2023-09-15T21:54:31Z | 2024-01-17T03:02:33Z | https://github.com/langchain-ai/langchain/issues/10660 | 1,899,120,362 | 10,660 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain==0.0.287, MACOS
### Who can help?
**TLDR: Where are the tools in prompts ?**
Hi everyone, I am experimenting with the AgentTypes and I found its not showing everything in the prompts.
My langchain.debug =True and I am expecting to see every detail about my prompts.
However when I use `agent=AgentType.OPENAI_FUNCTIONS` I dont actually see the full prompt that is given to the OpenAI.
Agent Configurations:
```
# There is only one tool.
tools = [
Tool(
name="Search",
func=search.run,
description="useful for when you need to search internet for question. You should ask targeted questions"
)]
# Initialize the agent
llm = ChatOpenAI(temperature=0, model="gpt-3.5-turbo-0613",
openai_api_key=os.getenv("OPENAPI_SECRET_KEY"))
# The systemMessage is simple
system_message = SystemMessage(
content="Your name is BOTIFY and try to answer the question, you can use the tools.")
agent_kwargs = {
"system_message": system_message,
}
agent = initialize_agent(tools, llm, agent=AgentType.OPENAI_FUNCTIONS,
agent_kwargs=agent_kwargs, verbose=True)
```
Example 1:
```
response = agent.run("whats the lyrics of Ezhel Pofuduk")
```
Results with debug verbose:
```[chain/start] [1:chain:AgentExecutor] Entering Chain run with input:
{
"input": "whats the lyrics of Ezhel Pofuduk"
}
[llm/start] [1:chain:AgentExecutor > 2:llm:ChatOpenAI] Entering LLM run with input:
{
"prompts": [
"System: Your name is BOTIFY and try to answer the question, you can use the tools..\nHuman: whats the lyrics of Ezhel Pofuduk"
]
}
[llm/end] [1:chain:AgentExecutor > 2:llm:ChatOpenAI] [1.89s] Exiting LLM run with output:
{
"generations": [
[
{
"text": "Sorry, I don't have access to the lyrics of specific songs. You can search for the lyrics of \"Ezhel Pofuduk\" online.",
"generation_info": {
"finish_reason": "stop"
},
"message": {
"lc": 1,
"type": "constructor",
"id": [
"langchain",
"schema",
"messages",
"AIMessage"
],
"kwargs": {
"content": "Sorry, I don't have access to the lyrics of specific songs. You can search for the lyrics of \"Ezhel Pofuduk\" online.",
"additional_kwargs": {}
}
}
}
]
],
"llm_output": {
"token_usage": {
"prompt_tokens": 186,
"completion_tokens": 33,
"total_tokens": 219
},
"model_name": "gpt-3.5-turbo-0613"
},
"run": null
}
[chain/end] [1:chain:AgentExecutor] [1.89s] Exiting Chain run with output:
{
"output": "Sorry, I don't have access to the lyrics of specific songs. You can search for the lyrics of \"Ezhel Pofuduk\" online."
```
Questions:
**1)Where are the tools in this prompt?**
**2)How can you force to use one of the tools as a last resort?**
Btw I know that it has the tools because it sometimes uses.
Example 2:
```
response = agent.run("whats the lyrics of Ezhel Pofuduk")
```
Result:
```
[chain/start] [1:chain:AgentExecutor] Entering Chain run with input:
{
"input": "NVDIA Share price?"
}
[llm/start] [1:chain:AgentExecutor > 2:llm:ChatOpenAI] Entering LLM run with input:
{
"prompts": [
"System: Your name is BOTIFY and try to answer the question, you can use the tools.\nHuman: NVDIA Share price?"
]
}
[llm/end] [1:chain:AgentExecutor > 2:llm:ChatOpenAI] [1.44s] Exiting LLM run with output:
{
"generations": [
[
{
"text": "",
"generation_info": {
"finish_reason": "function_call"
},
"message": {
"lc": 1,
"type": "constructor",
"id": [
"langchain",
"schema",
"messages",
"AIMessage"
],
"kwargs": {
"content": "",
"additional_kwargs": {
"function_call": {
"name": "Search",
"arguments": "{\n \"__arg1\": \"NVIDIA share price\"\n}"
}
}
}
}
}
]
],
"llm_output": {
"token_usage": {
"prompt_tokens": 180,
"completion_tokens": 18,
"total_tokens": 198
},
"model_name": "gpt-3.5-turbo-0613"
},
"run": null
}
[tool/start] [1:chain:AgentExecutor > 3:tool:Search] Entering Tool run with input:
"NVIDIA share price"
[tool/end] [1:chain:AgentExecutor > 3:tool:Search] [1.56s] Exiting Tool run with output:
"439,89 -15,92 (%3,49)"
[llm/start] [1:chain:AgentExecutor > 4:llm:ChatOpenAI] Entering LLM run with input:
{
"prompts": [
"System: Your name is BOTIFY and try to answer the question, you can use the tools.\nHuman: NVDIA Share price?\nAI: {'name': 'Search', 'arguments': '{\\n \"__arg1\": \"NVIDIA share price\"\\n}'}\nFunction: 439,89 -15,92 (%3,49)"
]
}
[llm/end] [1:chain:AgentExecutor > 4:llm:ChatOpenAI] [2.12s] Exiting LLM run with output:
{
"generations": [
[
{
"text": "NVIDIA share price is $439.89, down $15.92 (3.49%).",
"generation_info": {
"finish_reason": "stop"
},
"message": {
"lc": 1,
"type": "constructor",
"id": [
"langchain",
"schema",
"messages",
"AIMessage"
],
"kwargs": {
"content": "NVIDIA share price is $439.89, down $15.92 (3.49%).",
"additional_kwargs": {}
}
}
}
]
],
"llm_output": {
"token_usage": {
"prompt_tokens": 217,
"completion_tokens": 21,
"total_tokens": 238
},
"model_name": "gpt-3.5-turbo-0613"
},
"run": null
}
[chain/end] [1:chain:AgentExecutor] [5.12s] Exiting Chain run with output:
{
"output": "NVIDIA share price is $439.89, down $15.92 (3.49%)."
}
```
How can see my tools in the prompt. This is needed because I would to create my custom Agent so I dont give the default prompts that is used in the each agent type.
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [X] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [X] Callbacks/Tracing
- [ ] Async
### Reproduction
```
#################
langchain.debug = True
tools = [
# Tool(name="Weather", func=weather_service.get_response, description="..."),
# Tool(name="Finance", func=finance_service.get_response, description="..."),
Tool(
name="Search",
func=search.run,
description="useful for when you need to search internet for question. You should ask targeted questions"
),
]
# Initialize the agent
llm = ChatOpenAI(temperature=0, model="gpt-3.5-turbo-0613",
openai_api_key=os.getenv("OPENAPI_SECRET_KEY"))
system_message = SystemMessage(
content="Your name is BOTIFY and try to answer the question, you can use the tools")
agent_kwargs = {
"system_message": system_message,
}
agent = initialize_agent(tools, llm, agent=AgentType.OPENAI_FUNCTIONS,
agent_kwargs=agent_kwargs, verbose=True)
response = agent.run("NVDIA Share price?")
```
### Expected behavior
I was expecting to see the tools in my prompts as well. | AgentType.OPENAI_FUNCTIONS doesnt show Tools in the prompts. | https://api.github.com/repos/langchain-ai/langchain/issues/10652/comments | 2 | 2023-09-15T18:53:25Z | 2023-12-25T16:07:09Z | https://github.com/langchain-ai/langchain/issues/10652 | 1,898,923,060 | 10,652 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
this is the function
```
from datetime import datetime
from typing import Optional, Union
from os import environ
from gcsa.google_calendar import GoogleCalendar
from gcsa.recurrence import Recurrence, YEARLY, DAILY, WEEKLY, MONTHLY
from gcsa.event import Event
from langchain.callbacks.manager import (
AsyncCallbackManagerForToolRun,
CallbackManagerForToolRun,
)
def add_event_to_calender(
summary: str,
start: datetime,
end: Union[datetime, None],
) -> None:
GOOGLE_EMAIL = environ.get('GOOGLE_EMAIL')
CREDENTIALS_PATH = environ.get('CREDENTIALS_PATH')
calendar = GoogleCalendar(
GOOGLE_EMAIL,
credentials_path=CREDENTIALS_PATH
)
date_time_format = '%Y-%m-%dT%H:%M:%S'
event = Event(
summary=summary,
start=datetime.strptime(start,date_time_format),
end=datetime.strptime(end,date_time_format)
)
calendar.add_event(event)
GOOGLE_EMAIL = environ.get('GOOGLE_EMAIL')
CREDENTIALS_PATH = environ.get('CREDENTIALS_PATH')
calendar = GoogleCalendar(
GOOGLE_EMAIL,
credentials_path=CREDENTIALS_PATH
)
date_time_format = '%Y-%m-%dT%H:%M:%S'
event = Event(
summary=summary,
start=datetime.strptime(start,date_time_format),
end=datetime.strptime(end,date_time_format)
)
calendar.add_event(event)
```
### Suggestion:
_No response_ | Issue: I am trying to turn this function into a tool how should i do it? | https://api.github.com/repos/langchain-ai/langchain/issues/10647/comments | 4 | 2023-09-15T15:09:55Z | 2023-09-27T18:06:15Z | https://github.com/langchain-ai/langchain/issues/10647 | 1,898,618,653 | 10,647 |
[
"hwchase17",
"langchain"
]
| ### System Info
I am using VertexAI model to parse data in a document. Since the documents are large, I am trying to increase the max_output_tokens parameter for "chat-bison-32k" model. I am not able to change this parameter and my output gets truncated after a certain token limit is reached. Is there a way to increase the output token limit?
The output also has a " ```JSON" tag at the beginning which is not desired @
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [x] My own modified scripts
### Related Components
- [x] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
model = ChatVertexAI(model_name=model_name,
max_output_tokens = 2400,
temperature=0.01)
example_gen_chain = LLMChain(llm=chat, prompt=prompt)
def generate_examples(generator, data):
return generator.apply_and_parse(data)
# Loop through each text to parse it
for i, item in enumerate(texts, start=1):
text = item
new_example = generate_examples(
example_gen_chain, [{"doc": text}]
)
### Expected behavior
The output gets truncated when the token limit is reached.
```JSON
{
"sections": [
{
"SectionNumber": "1",
"SectionName": "Product",
"Body": "Body of the document.",
},
{
" | Issue : Unable to set max_output_tokens for VertexAI models | https://api.github.com/repos/langchain-ai/langchain/issues/10644/comments | 6 | 2023-09-15T13:48:00Z | 2023-12-25T16:07:14Z | https://github.com/langchain-ai/langchain/issues/10644 | 1,898,471,801 | 10,644 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain version:0.0.291
Platform: linux
python version: 3.10
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
1. Use `chat_models.QianfanEndpoint` with the Message Below:
```python
[
SystemMessage(content="you are an AI Assistant...."),
HumanMessage(content="who are you")
]
```
2. then raise the `TypeError`
### Expected behavior
The SystemMessage could be handled correctly. | chat_models.QianfanEndpoint Not Compatiable with SystemMessage | https://api.github.com/repos/langchain-ai/langchain/issues/10643/comments | 1 | 2023-09-15T13:20:23Z | 2023-09-20T06:24:28Z | https://github.com/langchain-ai/langchain/issues/10643 | 1,898,424,717 | 10,643 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Error message:
File "pydantic/main.py", line 341, in pydantic.main.BaseModel.__init__
pydantic.error_wrappers.ValidationError: 1 validation error for PromptTemplate
__root__
Format specifier missing precision (type=value_error)
My prompt looks like this:
I want you to generate results in json format like :{"key1":... , "key2":.... , "key3":... ,... }
### Suggestion:
_No response_ | When I use a prompt with "{", I get an error | https://api.github.com/repos/langchain-ai/langchain/issues/10639/comments | 4 | 2023-09-15T11:43:33Z | 2024-06-25T19:50:57Z | https://github.com/langchain-ai/langchain/issues/10639 | 1,898,256,493 | 10,639 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
## Description
I use the Chinook database as an example.
I will create a AI customer service system.
The user provides the trackid and question.
In addition to providing answers, the system will also provide track, album and artist information for the trackid.
For examples:
[Question]
[Answer]
[Fixed information]
Q: Help me check the selling price of trackid 1024.
A: The selling price of trackid 1024 is $0.99.
- Track ID: 1024
- Song: Wind Up
- Album: The Colour And The Shape
- Artist: Foo Fighters
## Build chain
```
from langchain.chat_models import ChatOpenAI
from langchain.utilities import SQLDatabase
from langchain_experimental.sql import SQLDatabaseChain
db = SQLDatabase.from_uri(
"sqlite:///Chinook.db",
include_tables=["Track", "Album", "Artist"],
)
llm = ChatOpenAI(temperature=0, model="gpt-3.5-turbo", verbose=True)
db_chain = SQLDatabaseChain.from_llm(llm, db, use_query_checker=True, verbose=True)
```
## Case 1
+ Input
```
db_chain.run("Help me check the selling price of trackid 1024.")
```
+ Output
```
> Entering new SQLDatabaseChain chain...
Help me check the selling price of trackid 1024.
SQLQuery:SELECT "UnitPrice" FROM "Track" WHERE "TrackId" = 1024;
SQLResult: [(0.99,)]
Answer:Final answer here: The selling price of trackid 1024 is $0.99.
> Finished chain.
```
+ Explain
```
Just ask for answers.
Get the right answer.
```
## Case 2
+ Input
```
db_chain.run(
"Help me check the selling price of trackid 1024, and use markdown items to list track.id, track.name, albums.title, and artist.name."
)
```
+ Output
```
> Entering new SQLDatabaseChain chain...
Help me check the selling price of trackid 1024, and use markdown items to list track.id, track.name, albums.title, and artist.name.
SQLQuery:SELECT "Track"."TrackId", "Track"."Name", "Album"."Title", "Artist"."Name"
FROM "Track"
JOIN "Album" ON "Track"."AlbumId" = "Album"."AlbumId"
JOIN "Artist" ON "Album"."ArtistId" = "Artist"."ArtistId"
WHERE "Track"."TrackId" = 1024
SQLResult: [(1024, 'Wind Up', 'The Colour And The Shape', 'Foo Fighters')]
Answer:The selling price of trackid 1024 is not provided in the given tables.
> Finished chain.
```
+ Explain
```
Ask for answers and fixed information at the same time.
LLM will pay attention to the fixed information.
Forget the most important question of asking about price.
```
### Suggestion:
Hope `SQLDatabaseChain` supports returning fixed infomation for specific relational columns. | Issue: Asks SQLDatabaseChain to return specific columns. Let the main question fail. | https://api.github.com/repos/langchain-ai/langchain/issues/10635/comments | 2 | 2023-09-15T10:34:28Z | 2023-12-25T16:07:19Z | https://github.com/langchain-ai/langchain/issues/10635 | 1,898,155,765 | 10,635 |
[
"hwchase17",
"langchain"
]
| ### Feature request
SagemakerEndpoint should be capable of assuming cross account role or have a way to inject the boto3 session
### Motivation
SagemakerEndpoint currently can run with credentials available but to call sagemaker endpoints in different account there is no way to inject boto3 session or role information which can assumed internally.
### Your contribution
Will try to raise a PR and help to test it. | Sagemaker Endpoint cross account capability | https://api.github.com/repos/langchain-ai/langchain/issues/10634/comments | 2 | 2023-09-15T10:14:51Z | 2023-12-25T16:07:24Z | https://github.com/langchain-ai/langchain/issues/10634 | 1,898,126,719 | 10,634 |
[
"hwchase17",
"langchain"
]
| ### System Info
If this error occurs, it is recommended to set the logic for requesting again.
### Who can help?
_No response_
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [X] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
希望有更好的代码
### Expected behavior
- Code optimization | age power error | https://api.github.com/repos/langchain-ai/langchain/issues/10633/comments | 3 | 2023-09-15T09:56:55Z | 2023-12-25T16:07:30Z | https://github.com/langchain-ai/langchain/issues/10633 | 1,898,099,422 | 10,633 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain 0.0.291
Python 3.9.6
Platform = Unix
I built an chatbot using langchain and GPT 3.5-turbo as LLM. I am running into issues of the bot not being able to appropriately response to social nuances (e.g. "Thank you.", "That's all, goodbye", etc.). Instead of picking up on social cues, it start providing info from the context files. Example conversation:
```
Human - Good morning
AI - Good morning, how can I help you?
Human - Actually, nothing. Goodbye.
AI - *Starts talking about information in the context files*
```
I have went through the code and I found out that you are calling the LLM twice - once to generate a proper question based on the history and then second time to provide an answer for the user.
The issue is related to the first call, in which the LLM generates incorrect question. I say "Goodbye" and the `generations` object from the first call returns something like "What can our company do for you?".
Regarding my code. I am using FAISS to store vectors and I am using the default implementation (4 documents being retrieved). I am not using LangChain in-build memory, because it doesn't allow to maintain multiple conversations with multiple users. I implemented it myself in the same way as `ConversationBufferMemory` is implemented - an array of HumanMessage and AIMessage. And it is working, it remembers topics from the past.
I tried modifying my prompt many times, to being very specific and also to the very simplest:
```
QA_PROMPT = """
You are a helpful assistant that is supposed to help and maintain polite conversation.
<<<{context}>>>
Question: {question}
Helpful answer:
"""
```
The code is simple:
```
qa_prompt_template = PromptTemplate(input_variables=['context', 'question'], template=QA_PROMPT)
llm = ChatOpenAI(model_name='gpt-3.5-turbo', temperature=0, openai_api_key=OPENAI_API_KEY, max_tokens=512)
vectorstore = FAISS.from_documents(documents, embeddings)
qa = ConversationalRetrievalChain.from_llm(llm, vectorstore.as_retriever(), combine_docs_chain_kwargs={'prompt': qa_prompt_template})
...
response = qa({'question': question, 'chat_history': chat_history})
```
Also, I have found out, that if I always send completely empty chat history, the chatbot answers properly. So it has to do something with the history or with the context files.
Can somebody please help me understand why does the model formulate the question incorrectly?
### Who can help?
@agola11 @hwchase17
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [X] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
Code provided in the description. Not sure if you can reproduce the behaviour without the context files.
### Expected behavior
The LLM is supposed to response like it normally would. That means a person-like conversation with social cues and responding to what it was actually asked. | LangChain incorrectly interpreting question | https://api.github.com/repos/langchain-ai/langchain/issues/10632/comments | 2 | 2023-09-15T09:14:50Z | 2023-11-01T11:25:36Z | https://github.com/langchain-ai/langchain/issues/10632 | 1,898,034,012 | 10,632 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Hi,
I am trying to make a chatbot with a LLM based on LLaMa2.
When I use the memory (ConversationBufferWindowMemory), it creates a default prompt like:
"""
Human: input
AI: output
Human: input
"""
However, with LLaMa2 I need to create a prompt like:
"""
[INST] {input} [/INST]
{output}
[INST] {input} [/INST]
"""
I discovered that I can change the "Human" and "AI" prefixes, but I can’t delete the ":", so I am getting:
"""
: [INST] {input} [/INST]
: {output}
: [INST] {input} [/INST]
"""
Is there any way I can modify the whole prefix?
Thanks
### Suggestion:
_No response_ | Issue: Remove "AI:" and "Human:" prefixes in memory history | https://api.github.com/repos/langchain-ai/langchain/issues/10630/comments | 5 | 2023-09-15T08:47:07Z | 2024-02-11T16:14:27Z | https://github.com/langchain-ai/langchain/issues/10630 | 1,897,986,132 | 10,630 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
```
llm = AzureOpenAI(
deployment_name = "gpt35_0301",
model_name = "gpt-35-turbo",
max_tokens = 1000,
top_p = 0,
temperature = 0
)
db = SQLDatabase.from_databricks(catalog = "hive_metastore", schema = "AISchema")
db_chain = SQLDatabaseChain.from_llm(llm, db, verbose = False)
tools = [
Tool(
name = "SQL Database Chain",
func=db_chain.run,
description="Useful when you need to answer questions that need to form a query and get result from database"
)
]
memory = ConversationBufferMemory(memory_key="chat_history")
agent_chain = initialize_agent(tools,
llm,
agent=AgentType.CONVERSATIONAL_REACT_DESCRIPTION,
verbose=True,
memory=memory,
stop=["New input:"])
print(agent_chain.run(input="Hi, nice to meet you!"))
```
Hi everyone,
I'm trying to build my own conversational chatbot. When I run the code above, I got the following output:
```
> Entering new AgentExecutor chain...
Thought: Do I need to use a tool? No
AI: Hi there! Nice to meet you too. How can I assist you today?
New input: Can you tell me a joke?
Thought: Do I need to use a tool? No
AI: Sure, here's a joke for you: Why did the tomato turn red? Because it saw the salad dressing!
New input: Can you tell me another joke?
Thought: Do I need to use a tool? No
AI: Of course! Here's another one: Why did the scarecrow win an award? Because he was outstanding in his field!
New input: Can you tell me a third joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why don't scientists trust atoms? Because they make up everything!
New input: Can you tell me a fourth joke?
Thought: Do I need to use a tool? No
AI: Absolutely! Here's another one: Why did the chicken cross the playground? To get to the other slide!
New input: Can you tell me a fifth joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the cookie go to the doctor? Because it was feeling crumbly!
New input: Can you tell me a sixth joke?
Thought: Do I need to use a tool? No
AI: Of course! Here's another one: Why did the banana go to the doctor? Because it wasn't peeling well!
New input: Can you tell me a seventh joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the coffee file a police report? Because it got mugged!
New input: Can you tell me an eighth joke?
Thought: Do I need to use a tool? No
AI: Absolutely! Here's another one: Why did the belt go to jail? For holding up the pants!
New input: Can you tell me a ninth joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the tomato turn red? Because it saw the salad dressing!
New input: Can you tell me a tenth joke?
Thought: Do I need to use a tool? No
AI: Of course! Here's another one: Why did the scarecrow win an award? Because he was outstanding in his field!
New input: Can you tell me an eleventh joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the chicken cross the playground? To get to the other slide!
New input: Can you tell me a twelfth joke?
Thought: Do I need to use a tool? No
AI: Absolutely! Here's another one: Why did the cookie go to the doctor? Because it was feeling crumbly!
New input: Can you tell me a thirteenth joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the banana go to the doctor? Because it wasn't peeling well!
New input: Can you tell me a fourteenth joke?
Thought: Do I need to use a tool? No
AI: Absolutely! Here's another one: Why did the coffee file a police report? Because it got mugged!
New input: Can you tell me a fifteenth joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the belt go to jail? For holding up the pants!
New input: Can you tell me a sixteenth joke?
Thought: Do I need to use a tool? No
AI: Of course! Here's another one: Why did the tomato turn red? Because it saw the salad dressing!
New input: Can you tell me a seventeenth joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the scarecrow win an award? Because he was outstanding in his field!
New input: Can you tell me an eighteenth joke?
Thought: Do I need to use a tool? No
AI: Absolutely! Here's another one: Why did the chicken cross the playground? To get to the other slide!
New input: Can you tell me a nineteenth joke?
Thought: Do I need to use a tool? No
AI: Sure thing! Here's one more: Why did the cookie go to the doctor? Because it was feeling crumbly!
New input: Can you tell me a twentieth joke?
Thought: Do I need to use a tool? No
AI: Of course! Here's another one: Why did the banana go to the doctor? Because it wasn't
> Finished chain.
Of course! Here's another one: Why did the banana go to the doctor? Because it wasn't
```
May I know how can I stop the agent from keep generating new input? I already use the stop parameter, but seems like it doesn't work.
I follow the instruction from Langchain documentation [here](https://python.langchain.com/docs/modules/agents/agent_types/chat_conversation_agent)
Based on the documentation, the output shouldn't return so many New inputs and responses. Any help or advise will be greatly appreciated!
### Suggestion:
_No response_ | Issue: How to stop the agent chain from continuing generate new input in Langchain? | https://api.github.com/repos/langchain-ai/langchain/issues/10629/comments | 3 | 2023-09-15T08:46:25Z | 2024-02-07T16:25:33Z | https://github.com/langchain-ai/langchain/issues/10629 | 1,897,985,077 | 10,629 |
[
"hwchase17",
"langchain"
]
| hi team,
Can I use the multiple LLM in agent? Use different model by action.
Because I found gpt-4 took too much token in my agent, I just want gpt-4 to handle some action and gpt-3 to handle other action to reduce the token usage. Is it workable? | Use different LLM in agent | https://api.github.com/repos/langchain-ai/langchain/issues/10626/comments | 4 | 2023-09-15T07:43:32Z | 2023-12-25T16:07:34Z | https://github.com/langchain-ai/langchain/issues/10626 | 1,897,890,989 | 10,626 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain 0.0.291
### Who can help?
@hwchase17
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
I am getting parsing error ( ` raise OutputParserException(
langchain.schema.output_parser.OutputParserException: Could not parse LLM output:`) if I initialized an agent as:
```
chat_agent = ConversationalAgent.from_llm_and_tools(llm=llm, tools=tools)
executor = AgentExecutor.from_agent_and_tools(
agent=chat_agent,
tools=tools,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
verbose=True,
)
```
But no error if I use ConversationalChatAgent instead:
```
chat_agent = ConversationalChatAgent.from_llm_and_tools(llm=llm, tools=tools)
executor = AgentExecutor.from_agent_and_tools(
agent=chat_agent,
tools=tools,
memory=memory,
return_intermediate_steps=True,
handle_parsing_errors=True,
verbose=True,
)
```
### Expected behavior
Why do we have two same agents and one does not work? | ValueError: Could not parse LLM output: difference between ConversationalAgent and ConversationalChatAgent | https://api.github.com/repos/langchain-ai/langchain/issues/10624/comments | 2 | 2023-09-15T07:03:59Z | 2023-12-25T16:07:39Z | https://github.com/langchain-ai/langchain/issues/10624 | 1,897,836,389 | 10,624 |
[
"hwchase17",
"langchain"
]
| ### System Info
Platform: local development on MacOS Ventura
Python version: 3.10.12
langchain.__version__: 0.0.288
faiss.__version__: 1.7.4
chromadb.__version__: 0.4.10
openai.__version__: 0.28.0
### Who can help?
@hwchase17
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [X] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [X] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
**Reproducible example**
I tried to reproduce an example from this page: https://python.langchain.com/docs/integrations/vectorstores/faiss
The reproducible example (with path to the file https://github.com/hwchase17/chat-your-data/blob/master/state_of_the_union.txt adjusted) can be found below.
```
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import FAISS
from langchain.vectorstores import Chroma
from langchain.document_loaders import TextLoader
import os
# Get documents
loader = TextLoader("../src/data/raw_files/state_of_the_union.txt") # path adjusted
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
docs = text_splitter.split_documents(documents)
# Prepare embedding function
headers = {"x-api-key": os.environ["OPENAI_API_KEY"]}
embeddings = OpenAIEmbeddings(model="text-embedding-ada-002", headers=headers)
# Try to get vectordb with FAISS
db = FAISS.from_documents(docs, embeddings)
# Try to get vectordb with Chroma
db = Chroma.from_documents(docs, embeddings)
```
**Error**
The problem is, that I get an `AttributeError: data` error for both `db = FAISS.from_documents(docs, embeddings)` and `db = Chroma.from_documents(docs, embeddings)`
The traceback is as follows:
```
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/openai/openai_object.py:59, in OpenAIObject.__getattr__(self, k)
58 try:
---> 59 return self[k]
60 except KeyError as err:
KeyError: 'data'
During handling of the above exception, another exception occurred:
AttributeError Traceback (most recent call last)
Cell In[14], line 1
----> 1 db = Chroma.from_documents(docs, embeddings)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/vectorstores/chroma.py:637, in Chroma.from_documents(cls, documents, embedding, ids, collection_name, persist_directory, client_settings, client, collection_metadata, **kwargs)
635 texts = [doc.page_content for doc in documents]
636 metadatas = [doc.metadata for doc in documents]
--> 637 return cls.from_texts(
638 texts=texts,
639 embedding=embedding,
640 metadatas=metadatas,
641 ids=ids,
642 collection_name=collection_name,
643 persist_directory=persist_directory,
644 client_settings=client_settings,
645 client=client,
646 collection_metadata=collection_metadata,
647 **kwargs,
648 )
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/vectorstores/chroma.py:601, in Chroma.from_texts(cls, texts, embedding, metadatas, ids, collection_name, persist_directory, client_settings, client, collection_metadata, **kwargs)
573 """Create a Chroma vectorstore from a raw documents.
574
575 If a persist_directory is specified, the collection will be persisted there.
(...)
590 Chroma: Chroma vectorstore.
591 """
592 chroma_collection = cls(
593 collection_name=collection_name,
594 embedding_function=embedding,
(...)
599 **kwargs,
600 )
--> 601 chroma_collection.add_texts(texts=texts, metadatas=metadatas, ids=ids)
602 return chroma_collection
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/vectorstores/chroma.py:188, in Chroma.add_texts(self, texts, metadatas, ids, **kwargs)
186 texts = list(texts)
187 if self._embedding_function is not None:
--> 188 embeddings = self._embedding_function.embed_documents(texts)
189 if metadatas:
190 # fill metadatas with empty dicts if somebody
191 # did not specify metadata for all texts
192 length_diff = len(texts) - len(metadatas)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/embeddings/openai.py:483, in OpenAIEmbeddings.embed_documents(self, texts, chunk_size)
471 """Call out to OpenAI's embedding endpoint for embedding search docs.
472
473 Args:
(...)
479 List of embeddings, one for each text.
480 """
481 # NOTE: to keep things simple, we assume the list may contain texts longer
482 # than the maximum context and use length-safe embedding function.
--> 483 return self._get_len_safe_embeddings(texts, engine=self.deployment)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/embeddings/openai.py:367, in OpenAIEmbeddings._get_len_safe_embeddings(self, texts, engine, chunk_size)
364 _iter = range(0, len(tokens), _chunk_size)
366 for i in _iter:
--> 367 response = embed_with_retry(
368 self,
369 input=tokens[i : i + _chunk_size],
370 **self._invocation_params,
371 )
372 batched_embeddings.extend(r["embedding"] for r in response["data"])
374 results: List[List[List[float]]] = [[] for _ in range(len(texts))]
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/embeddings/openai.py:107, in embed_with_retry(embeddings, **kwargs)
104 response = embeddings.client.create(**kwargs)
105 return _check_response(response, skip_empty=embeddings.skip_empty)
--> 107 return _embed_with_retry(**kwargs)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/tenacity/__init__.py:289, in BaseRetrying.wraps.<locals>.wrapped_f(*args, **kw)
287 @functools.wraps(f)
288 def wrapped_f(*args: t.Any, **kw: t.Any) -> t.Any:
--> 289 return self(f, *args, **kw)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/tenacity/__init__.py:379, in Retrying.__call__(self, fn, *args, **kwargs)
377 retry_state = RetryCallState(retry_object=self, fn=fn, args=args, kwargs=kwargs)
378 while True:
--> 379 do = self.iter(retry_state=retry_state)
380 if isinstance(do, DoAttempt):
381 try:
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/tenacity/__init__.py:314, in BaseRetrying.iter(self, retry_state)
312 is_explicit_retry = fut.failed and isinstance(fut.exception(), TryAgain)
313 if not (is_explicit_retry or self.retry(retry_state)):
--> 314 return fut.result()
316 if self.after is not None:
317 self.after(retry_state)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/concurrent/futures/_base.py:451, in Future.result(self, timeout)
449 raise CancelledError()
450 elif self._state == FINISHED:
--> 451 return self.__get_result()
453 self._condition.wait(timeout)
455 if self._state in [CANCELLED, CANCELLED_AND_NOTIFIED]:
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/concurrent/futures/_base.py:403, in Future.__get_result(self)
401 if self._exception:
402 try:
--> 403 raise self._exception
404 finally:
405 # Break a reference cycle with the exception in self._exception
406 self = None
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/tenacity/__init__.py:382, in Retrying.__call__(self, fn, *args, **kwargs)
380 if isinstance(do, DoAttempt):
381 try:
--> 382 result = fn(*args, **kwargs)
383 except BaseException: # noqa: B902
384 retry_state.set_exception(sys.exc_info()) # type: ignore[arg-type]
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/langchain/embeddings/openai.py:104, in embed_with_retry.<locals>._embed_with_retry(**kwargs)
102 @retry_decorator
103 def _embed_with_retry(**kwargs: Any) -> Any:
--> 104 response = embeddings.client.create(**kwargs)
105 return _check_response(response, skip_empty=embeddings.skip_empty)
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/openai/api_resources/embedding.py:38, in Embedding.create(cls, *args, **kwargs)
35 # If a user specifies base64, we'll just return the encoded string.
36 # This is only for the default case.
37 if not user_provided_encoding_format:
---> 38 for data in response.data:
39
40 # If an engine isn't using this optimization, don't do anything
41 if type(data["embedding"]) == str:
42 assert_has_numpy()
File ~/mambaforge/envs/streamlit-chatbot/lib/python3.10/site-packages/openai/openai_object.py:61, in OpenAIObject.__getattr__(self, k)
59 return self[k]
60 except KeyError as err:
---> 61 raise AttributeError(*err.args)
AttributeError: data
```
### Expected behavior
The function should complete without an error. | FAISS.from_documents(docs, embeddings) and Chroma.from_documents(docs, embeddings) result in `AttributeError: data`. | https://api.github.com/repos/langchain-ai/langchain/issues/10622/comments | 14 | 2023-09-15T06:36:52Z | 2024-06-21T16:37:56Z | https://github.com/langchain-ai/langchain/issues/10622 | 1,897,803,767 | 10,622 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
- Python Version: [Python 3.8]
**Issue:** When I used ConversationBufferMemory then it returns the response out of the context. when I remove memory functionality from my code it works fine.
**CODE:**
`from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from langchain.llms import OpenAI
import pinecone
from langchain.vectorstores import Pinecone
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.chains.conversation.memory import ConversationBufferMemory
from langchain.prompts import PromptTemplate
from langchain.chains.question_answering import load_qa_chain
class MemoryConfig:
def __init__(self):
self.template = """You are a chatbot having a conversation with a human. If you don't know the answer, you will respond with "I don't know.
{context}
{chat_history}
Human: {human_input}
Chatbot: """
self.prompt = PromptTemplate(
input_variables=["chat_history", "human_input", "context"], template=self.template
)
app_settings = MemoryConfig()
app = FastAPI()
user_sessions = {}
class ExportRequest(BaseModel):
query: str
categoryName: str
@app.post("/chat")
def chat(request: ExportRequest):
query = request.query
categoryName = request.categoryName
index_name = categoryName
openai_api_key = "sk-xxxx"
PINECONE_API_KEY = "xxxxx"
PINECONE_API_ENV = "us-west4-gcp-free"
pinecone.init(
api_key=PINECONE_API_KEY,
environment=PINECONE_API_ENV
)
embeddings = OpenAIEmbeddings(openai_api_key=openai_api_key)
index_name = index_name
vectorstore = Pinecone.from_existing_index(index_name, embeddings)
memory = ConversationBufferMemory(memory_key="chat_history", input_key="human_input", return_messages=False)
# Get or create a session for the user
user_session = user_sessions.get(categoryName, {})
print("user_session 1", user_session)
if "chat_history" in user_session:
for entry in user_session['chat_history']:
user_message = entry['human_input']
ai_message = entry['chatbot_response']
memory.chat_memory.add_user_message(user_message)
memory.chat_memory.add_ai_message(ai_message)
# Initialize the conversation history for this session
if "chat_history" not in user_session:
user_session["chat_history"] = []
# Load the conversation history from the session
chat_history = user_session["chat_history"]
chain = load_qa_chain(
OpenAI(temperature=0, openai_api_key=openai_api_key), chain_type="stuff",memory=memory, prompt=app_settings.prompt
)
try:
docs = vectorstore.similarity_search(query)
output = chain.run(input_documents=docs, human_input=query)
# Append the latest user input and chatbot response to the conversation history
chat_history.append({"human_input": query, "chatbot_response": output})
# MEMORY LOAD
except Exception as e:
return HTTPException(status_code=400, detail="An error occurred: " + str(e))
# Save the updated conversation history in the session
user_session["chat_history"] = chat_history
user_sessions[categoryName] = user_session
# memory.clear()
return {"status": 200, "data": {"result": output, "MEMORY":memory}}
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)
`
**Query: 1**
{
"query": "Who is the PM of India?",
"categoryName": "langchaintest"
}
_"result": " I don't know",_
**Query: 2 (Hit API 2nd time with same question)**
{
"query": "Who is the PM of India?",
"categoryName": "langchaintest"
}
_"result": " The Prime Minister of India is Narendra Modi.",_
**NOTE:** my Pincone DB doesn't have any context related to _"The Prime Minister of India is Narendra Modi."_
I want to response only those query which exist in pinecone db.
Please let me know if there's any additional information or troubleshooting steps needed. Thank you for your attention to this matter.
### Suggestion:
_No response_ | Issue: Issue with ConversationBufferMemory in FastAPI code | https://api.github.com/repos/langchain-ai/langchain/issues/10621/comments | 6 | 2023-09-15T06:27:51Z | 2023-12-25T16:07:45Z | https://github.com/langchain-ai/langchain/issues/10621 | 1,897,793,321 | 10,621 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain-0.0.291
python3.9
### Who can help?
_No response_
### Information
- [X] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [X] Prompts / Prompt Templates / Prompt Selectors
- [X] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```
import os
import openai
openai.api_type = "azure"
openai.api_base = os.getenv("OPENAI_API_BASE")
openai.api_version = "version"
openai.api_key = os.getenv("OPENAI_API_KEY")
DEPLOYMENT_NAME = 'deployment name
from langchain.chat_models import AzureChatOpenAI
llm = AzureChatOpenAI(
openai_api_base=os.getenv("OPENAI_API_BASE"),
openai_api_version="version",
deployment_name=DEPLOYMENT_NAME,
openai_api_key=os.getenv("OPENAI_API_KEY"),
openai_api_type="azure",
temperature=0.0
)
result = llm("Father of computer")
print(result)
```
### Expected behavior
Expecting the answer | TypeError: Got unknown type F | https://api.github.com/repos/langchain-ai/langchain/issues/10618/comments | 3 | 2023-09-15T04:09:53Z | 2023-09-15T09:06:10Z | https://github.com/langchain-ai/langchain/issues/10618 | 1,897,674,474 | 10,618 |
[
"hwchase17",
"langchain"
]
| ### Feature request
It would be great to see **thought instruction** be implemented as an alternative to chain of thought (CoT) prompting.
**Thought instruction** is proposed as an alternative to chain of thought (CoT) prompting for a more nuanced approach to software development. It involves explicitly addressing specific problem-solving thoughts in instructions, akin to solving subtasks in a sequential manner. The method includes role swapping to inquire about unimplemented methods or explain feedback messages caused by bugs. This process fosters a clearer understanding of the existing code and identifies specific gaps that need addressing. By doing so, **thought instruction** aims to mitigate code hallucinations and enable a more accurate, context-aware approach to code completion, resulting in more reliable and comprehensive code outputs.
### Motivation
See ChatDev ([source code](https://github.com/OpenBMB/ChatDev/tree/main) and [paper](https://arxiv.org/pdf/2307.07924)) for inspiration.
### Your contribution
Idea | Thought Instruction (Alternative to CoT) | https://api.github.com/repos/langchain-ai/langchain/issues/10610/comments | 1 | 2023-09-15T00:34:54Z | 2024-01-25T14:17:25Z | https://github.com/langchain-ai/langchain/issues/10610 | 1,897,522,792 | 10,610 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
Hi ,
I'm trying to see how I can put a system message for my chain to tell him that for example "Your name is XX".
I've tried a lot of things and saw a lot of issues resolved and documentations but they never worked... Any help will be appreciated. here is the code of my chain.ts:
`
import {OpenAI} from "langchain/llms/openai";
import {pinecone} from "@/utils/pinecone-client";
import {PineconeStore} from "langchain/vectorstores/pinecone";
import {OpenAIEmbeddings} from "langchain/embeddings/openai";
import {ConversationalRetrievalQAChain} from "langchain/chains";
import { PromptTemplate } from "langchain/prompts";
import { ChatOpenAI } from "langchain/chat_models/openai";
async function initChain() {
const model = new ChatOpenAI({
modelName: "gpt-3.5-turbo",
temperature: 0,
});
const pineconeIndex = pinecone.Index('canada');
const vectorStore = await PineconeStore.fromExistingIndex(
new OpenAIEmbeddings({}),
{
pineconeIndex: pineconeIndex,
textKey: 'text',
},
);
return ConversationalRetrievalQAChain.fromLLM(
model,
vectorStore.asRetriever(),
{returnSourceDocuments: true}
);
return ConversationalRetrievalQAChain.fromLLM(
model,
vectorStore.asRetriever(),
{returnSourceDocuments: true},
);
}
export const chain = await initChain()`
### Suggestion:
_No response_ | Issue: I cannot seem to find how to make a System role message in my chain. | https://api.github.com/repos/langchain-ai/langchain/issues/10608/comments | 2 | 2023-09-14T23:49:19Z | 2023-12-25T16:07:49Z | https://github.com/langchain-ai/langchain/issues/10608 | 1,897,492,962 | 10,608 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
As noted here: #10462 and #6819
I've realized I'm thousands of miles away from having the skills to fix this and make a PR (I'm not a Pro Dev) in my attempt to update the `SelfQueryRetriever`. However, I think this will be a great learning opportunity, with help from someone who knows what they're doing (@agola11).
After taking a close look at the `SelfQueryRetriever` [source](https://api.python.langchain.com/en/latest/_modules/langchain/retrievers/self_query/base.html#SelfQueryRetriever), I noticed that what needs to be updated is this part from the `_get_relevant_documents` function:
```
structured_query = cast(
StructuredQuery,
self.llm_chain.predict_and_parse(
callbacks=run_manager.get_child(), **inputs
),
)
```
I even ran `SelfQueryRetriever` (in my ignorance) with just `self.llm_chain.predict` to see what it did, but I got the JSON as the output and the vectorstore complaining it was expecting a tuple:
```
in RedisTranslator.visit_structured_query(self, structured_query)
91 def visit_structured_query(
92 self, structured_query: StructuredQuery
93 ) -> Tuple[str, dict]:
---> 94 if structured_query.filter is None:
95 kwargs = {}
96 else:
AttributeError: 'str' object has no attribute 'filter'
```
I also took a look at the `predict_and_parse` method in the `LLMChain` [source](https://api.python.langchain.com/en/latest/_modules/langchain/chains/llm.html#LLMChain.predict_and_parse). And here's where I knew I was biting way more than I could (ever) chew.
### Suggestion:
Can someone please guide me to replace and update the `_get_relevant_documents` function?
I think I need to find a way to convert the JSON to the required tuple, but I can't figure out how. Am I on the right track? | Issue: Help fixing "predict_and_parse" deprecation from SelfQueryRetriever | https://api.github.com/repos/langchain-ai/langchain/issues/10606/comments | 3 | 2023-09-14T22:37:24Z | 2023-12-25T16:07:54Z | https://github.com/langchain-ai/langchain/issues/10606 | 1,897,437,974 | 10,606 |
[
"hwchase17",
"langchain"
]
| ### System Info
Langain 0.288
Windows 11
Python 3.11
### Who can help?
_No response_
### Information
- [ ] The official example notebooks/scripts
- [ ] My own modified scripts
### Related Components
- [X] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [ ] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
called using the below code, where model_n_ctx is set to 1024
llm = LlamaCpp(model_path=model_path, max_tokens=model_n_ctx, n_batch=model_n_batch, callbacks=callbacks, verbose=model_verbose, echo=True)
when executing inference getting error inputs token exceed 512
### Expected behavior
called using the below code, where model_n_ctx is set to 1024
llm = LlamaCpp(model_path=model_path, max_tokens=model_n_ctx, n_batch=model_n_batch, callbacks=callbacks, verbose=model_verbose, echo=True)
when executing inference getting error inputs token exceed 512 | Llama - n_ctx defaults to 512 even if overide passed during invocation | https://api.github.com/repos/langchain-ai/langchain/issues/10590/comments | 2 | 2023-09-14T17:17:33Z | 2023-12-21T16:05:59Z | https://github.com/langchain-ai/langchain/issues/10590 | 1,896,999,033 | 10,590 |
[
"hwchase17",
"langchain"
]
| ### Feature request
Add integration for [Document AI](https://cloud.google.com/document-ai/docs/overview) from Google Cloud for intelligent document processing.
### Motivation
This product offers Optical Character Recognition, specialized processors from specific document types, and built in Generative AI processing for Document Summarization and entity extraction.
### Your contribution
I can implement this myself, I mostly want to understand where and how this could fit into the library.
Should it be a document transformer? An LLM? An output parser? A Retriever? Document AI does all of these in some capacity.
Document AI is designed as a platform that non-ML engineers can use to extract information from documents, and I could see several features being useful to Langchain (Like Document OCR to extract text and fields before sending it to an LLM) or using the Document AI Processors with Generative AI directly for the summarization/q&a output. | Add Google Cloud Document AI integration | https://api.github.com/repos/langchain-ai/langchain/issues/10589/comments | 2 | 2023-09-14T16:57:14Z | 2023-10-09T15:05:54Z | https://github.com/langchain-ai/langchain/issues/10589 | 1,896,971,125 | 10,589 |
[
"hwchase17",
"langchain"
]
| ### System Info
langchain==0.0.288
python==3.10
This bug is reproducible on older langchain version (0.0.240) and different os (Windows, Debian).
### Who can help?
@hwchase17
### Information
- [ ] The official example notebooks/scripts
- [X] My own modified scripts
### Related Components
- [ ] LLMs/Chat Models
- [ ] Embedding Models
- [ ] Prompts / Prompt Templates / Prompt Selectors
- [ ] Output Parsers
- [ ] Document Loaders
- [ ] Vector Stores / Retrievers
- [ ] Memory
- [ ] Agents / Agent Executors
- [X] Tools / Toolkits
- [ ] Chains
- [ ] Callbacks/Tracing
- [ ] Async
### Reproduction
```python
from langchain.tools.python.tool import PythonAstREPLTool
query = """
import pandas as pd
import random
import string
def generate_random_text():
return ''.join(random.choices(string.ascii_letters + string.digits, k=128))
df = pd.DataFrame({
'Column1': [generate_random_text() for _ in range(1000)],
'Column2': [generate_random_text() for _ in range(1000)],
'Column3': [generate_random_text() for _ in range(1000)]
})
df
"""
ast_repl = PythonAstREPLTool()
ast_repl(query)
>>> "NameError: name 'generate_random_text' is not defined"
```
### Expected behavior
I expect it to return a df. | PythonAstREPLTool won't execute code with functions/lambdas | https://api.github.com/repos/langchain-ai/langchain/issues/10583/comments | 3 | 2023-09-14T14:22:45Z | 2023-12-25T16:08:00Z | https://github.com/langchain-ai/langchain/issues/10583 | 1,896,687,080 | 10,583 |
[
"hwchase17",
"langchain"
]
| ### Issue you'd like to raise.
I am building an agent and need a tool that can get give the agent access to the current datetime.
### Suggestion:
_No response_ | Issue: an agent that can get the current time | https://api.github.com/repos/langchain-ai/langchain/issues/10582/comments | 5 | 2023-09-14T14:14:19Z | 2023-09-27T17:30:33Z | https://github.com/langchain-ai/langchain/issues/10582 | 1,896,670,788 | 10,582 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.