Spaces:
Running
Running
Upload chain.py
Browse files
chain.py
ADDED
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""This module contains functions for loading a ConversationalRetrievalChain"""
|
2 |
+
|
3 |
+
import logging
|
4 |
+
|
5 |
+
import wandb
|
6 |
+
from langchain.chains import ConversationalRetrievalChain
|
7 |
+
from langchain.chat_models import ChatOpenAI
|
8 |
+
from langchain.embeddings import OpenAIEmbeddings
|
9 |
+
from langchain.vectorstores import Chroma
|
10 |
+
from prompts import load_chat_prompt
|
11 |
+
|
12 |
+
|
13 |
+
logger = logging.getLogger(__name__)
|
14 |
+
|
15 |
+
|
16 |
+
def load_vector_store(wandb_run: wandb.run, openai_api_key: str) -> Chroma:
|
17 |
+
"""Load a vector store from a Weights & Biases artifact
|
18 |
+
Args:
|
19 |
+
run (wandb.run): An active Weights & Biases run
|
20 |
+
openai_api_key (str): The OpenAI API key to use for embedding
|
21 |
+
Returns:
|
22 |
+
Chroma: A chroma vector store object
|
23 |
+
"""
|
24 |
+
# load vector store artifact
|
25 |
+
vector_store_artifact_dir = wandb_run.use_artifact(
|
26 |
+
wandb_run.config.vector_store_artifact, type="search_index"
|
27 |
+
).download()
|
28 |
+
embedding_fn = OpenAIEmbeddings(openai_api_key=openai_api_key)
|
29 |
+
# load vector store
|
30 |
+
vector_store = Chroma(
|
31 |
+
embedding_function=embedding_fn, persist_directory=vector_store_artifact_dir
|
32 |
+
)
|
33 |
+
|
34 |
+
return vector_store
|
35 |
+
|
36 |
+
|
37 |
+
def load_chain(wandb_run: wandb.run, vector_store: Chroma, openai_api_key: str):
|
38 |
+
"""Load a ConversationalQA chain from a config and a vector store
|
39 |
+
Args:
|
40 |
+
wandb_run (wandb.run): An active Weights & Biases run
|
41 |
+
vector_store (Chroma): A Chroma vector store object
|
42 |
+
openai_api_key (str): The OpenAI API key to use for embedding
|
43 |
+
Returns:
|
44 |
+
ConversationalRetrievalChain: A ConversationalRetrievalChain object
|
45 |
+
"""
|
46 |
+
retriever = vector_store.as_retriever()
|
47 |
+
llm = ChatOpenAI(
|
48 |
+
openai_api_key=openai_api_key,
|
49 |
+
model_name=wandb_run.config.model_name,
|
50 |
+
temperature=wandb_run.config.chat_temperature,
|
51 |
+
max_retries=wandb_run.config.max_fallback_retries,
|
52 |
+
)
|
53 |
+
chat_prompt_dir = wandb_run.use_artifact(
|
54 |
+
wandb_run.config.chat_prompt_artifact, type="prompt"
|
55 |
+
).download()
|
56 |
+
qa_prompt = load_chat_prompt(f"{chat_prompt_dir}/chat_prompt_massa.json")
|
57 |
+
|
58 |
+
print ( '\\n===================\\nqa_prompt = ', qa_prompt)
|
59 |
+
|
60 |
+
qa_chain = ConversationalRetrievalChain.from_llm(
|
61 |
+
llm=llm,
|
62 |
+
chain_type="stuff",
|
63 |
+
retriever=retriever,
|
64 |
+
combine_docs_chain_kwargs={"prompt": qa_prompt},
|
65 |
+
return_source_documents=True,
|
66 |
+
)
|
67 |
+
return qa_chain
|
68 |
+
|
69 |
+
|
70 |
+
def get_answer(
|
71 |
+
chain: ConversationalRetrievalChain,
|
72 |
+
question: str,
|
73 |
+
chat_history: list[tuple[str, str]],
|
74 |
+
):
|
75 |
+
"""Get an answer from a ConversationalRetrievalChain
|
76 |
+
Args:
|
77 |
+
chain (ConversationalRetrievalChain): A ConversationalRetrievalChain object
|
78 |
+
question (str): The question to ask
|
79 |
+
chat_history (list[tuple[str, str]]): A list of tuples of (question, answer)
|
80 |
+
Returns:
|
81 |
+
str: The answer to the question
|
82 |
+
"""
|
83 |
+
result = chain(
|
84 |
+
inputs={"question": question, "chat_history": chat_history},
|
85 |
+
return_only_outputs=True,
|
86 |
+
)
|
87 |
+
response = f"Answer:\t{result['answer']}"
|
88 |
+
return response
|