updates llama_index to 0.10.15

README.md

@@ -3,6 +3,7 @@ title: Mlk8s
 emoji: 😻
 colorFrom: pink
 colorTo: pink
+python_version: 3.9
 sdk: streamlit
 sdk_version: 1.25.0
 app_file: app.py
@@ -11,3 +12,4 @@ license: openrail
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+

app.py

@@ -29,18 +29,19 @@ hf_api_key = os.environ['HF_TOKEN']
 ch_api_key = os.environ['COHERE_TOKEN']
 bs_api_key = os.environ['BASETEN_TOKEN']
 
-index_model = "Writer/camel-5b-hf"
+#index_model = "Writer/camel-5b-hf"
+index_model = "Arylwen/instruct-palmyra-20b-gptq-8"
 INDEX_NAME = f"{index_model.replace('/', '-')}-default-no-coref"
 persist_path = f"storage/{INDEX_NAME}"
 MAX_LENGTH = 1024
 MAX_NEW_TOKENS = 250
 
-import baseten
-@st.cache_resource
-def set_baseten_key(bs_api_key):
-    baseten.login(bs_api_key)
+#import baseten
+#@st.cache_resource
+#def set_baseten_key(bs_api_key):
+#    baseten.login(bs_api_key)
 
-set_baseten_key(bs_api_key)
+#set_baseten_key(bs_api_key)
 
 def autoplay_video(video_path):
     with open(video_path, "rb") as f:
@@ -70,16 +71,16 @@ f'''
     st.caption('''###### corpus by [@[email protected]](https://sigmoid.social/@ArxivHealthcareNLP)''')
     st.caption('''###### KG Questions by [arylwen](https://github.com/arylwen/mlk8s)''')
     
-from llama_index import StorageContext
-from llama_index import ServiceContext
-from llama_index import load_index_from_storage 
-from llama_index.langchain_helpers.text_splitter import SentenceSplitter
-from llama_index.node_parser import SimpleNodeParser
-from llama_index import LLMPredictor
+from llama_index.core import StorageContext, ServiceContext, load_index_from_storage
+#from llama_index import ServiceContext
+# from llama_index import load_index_from_storage 
+from llama_index.core.node_parser import SentenceSplitter
+#from llama_index.node_parser import SimpleNodeParser
+from llama_index.core.service_context_elements.llm_predictor import LLMPredictor
 
 from langchain import HuggingFaceHub
 from langchain.llms.cohere import Cohere
-from langchain.llms import Baseten
+#from langchain.llms import Baseten
 
 import tiktoken
 
@@ -87,11 +88,11 @@ import openai
 #extensions to llama_index to support openai compatible endpoints, e.g. llama-api
 from kron.llm_predictor.KronOpenAILLM import KronOpenAI
 #baseten deployment expects a specific request format
-from kron.llm_predictor.KronBasetenCamelLLM import KronBasetenCamelLLM
+#from kron.llm_predictor.KronBasetenCamelLLM import KronBasetenCamelLLM
 from kron.llm_predictor.KronLLMPredictor import KronLLMPredictor
 
 #writer/camel uses endoftext 
-from llama_index.utils import globals_helper
+from llama_index.core.utils import globals_helper
 enc = tiktoken.get_encoding("gpt2")
 tokenizer = lambda text: enc.encode(text, allowed_special={"<|endoftext|>"})
 globals_helper._tokenizer = tokenizer
@@ -129,15 +130,15 @@ def get_cohere_predictor(query_model):
     llm_predictor = LLMPredictor(llm)
     return llm_predictor
 
-def get_baseten_predictor(query_model):
-    # no embeddings for now
-    set_openai_local()
-    llm=KronBasetenCamelLLM(model='3yd1ke3', temperature = 0.01,
+#def get_baseten_predictor(query_model):
+#    # no embeddings for now
+#    set_openai_local()
+#    llm=KronBasetenCamelLLM(model='3yd1ke3', temperature = 0.01,
 #                       model_kwargs={"temperature": 0.01, "max_length": MAX_LENGTH, 'repetition_penalty':1.07}, 
-                       model_kwargs={"temperature": 0.01, "max_length": MAX_LENGTH, 'frequency_penalty':1}, 
-                       cohere_api_key=ch_api_key)
-    llm_predictor = LLMPredictor(llm)
-    return llm_predictor
+#                       model_kwargs={"temperature": 0.01, "max_length": MAX_LENGTH, 'frequency_penalty':1}, 
+#                       cohere_api_key=ch_api_key)
+#    llm_predictor = LLMPredictor(llm)
+#    return llm_predictor
 
 def get_kron_openai_predictor(query_model): 
     # define LLM
@@ -150,7 +151,8 @@ def get_servce_context(llm_predictor):
     # define TextSplitter
     text_splitter = SentenceSplitter(chunk_size=192, chunk_overlap=48, paragraph_separator='\n')
     #define NodeParser
-    node_parser = SimpleNodeParser(text_splitter=text_splitter)
+    #node_parser = SimpleNodeParser(text_splitter=text_splitter)
+    node_parser = text_splitter
     #define ServiceContext
     service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, node_parser=node_parser)
     return service_context
@@ -219,11 +221,11 @@ def build_cohere_query_engine(query_model, persist_path):
     query_engine = load_query_engine(llm_predictor, persist_path)
     return query_engine
 
-@st.cache_resource
-def build_baseten_query_engine(query_model, persist_path):
-    llm_predictor = get_baseten_predictor(query_model)    
-    query_engine = load_query_engine(llm_predictor, persist_path)
-    return query_engine
+#@st.cache_resource
+#def build_baseten_query_engine(query_model, persist_path):
+#    llm_predictor = get_baseten_predictor(query_model)    
+#    query_engine = load_query_engine(llm_predictor, persist_path)
+#    return query_engine
 
 def format_response(answer):
     # Replace any eventual --
@@ -257,18 +259,18 @@ if __spaces__ :
     with query:
         answer_model = st.radio(
             "Choose the model used for inference:",
-            ('hf/tiiuae/falcon-7b-instruct', 'cohere/command', 'baseten/Camel-5b', 'openai/text-davinci-003') #TODO start hf inference container on demand
+            ('hf/tiiuae/falcon-7b-instruct', 'cohere/command', 'openai/gpt-3.5-turbo-instruct') #TODO start hf inference container on demand
         )
 else :    
     with query:
         answer_model = st.radio(
             "Choose the model used for inference:",
-            ('Writer/camel-5b-hf', 'mosaicml/mpt-7b-instruct', 'hf/tiiuae/falcon-7b-instruct', 'cohere/command', 'baseten/Camel-5b', 'openai/text-davinci-003')
+            ('Writer/camel-5b-hf', 'mosaicml/mpt-7b-instruct', 'hf/tiiuae/falcon-7b-instruct', 'cohere/command', 'baseten/Camel-5b', 'openai/gpt-3.5-turbo-instruct')
         )
 
-if answer_model == 'openai/text-davinci-003':
+if answer_model == 'openai/gpt-3.5-turbo-instruct':
     print(answer_model)
-    query_model = 'text-davinci-003'
+    query_model = 'gpt-3.5-turbo-instruct'
     clear_question(query_model)
     set_openai()
     query_engine = build_kron_query_engine(query_model, persist_path)

explainable.py

@@ -7,7 +7,7 @@ def explain(answer):
             #all_reference_texts = ''
     for nodewithscore in answer.source_nodes:  
         node = nodewithscore.node
-        from llama_index.schema import NodeRelationship
+        from llama_index.core.schema import NodeRelationship
         if NodeRelationship.SOURCE in node.relationships:
             node_id = node.relationships[NodeRelationship.SOURCE].node_id
             node_id = node_id.split('/')[-1]

graph.html

@@ -88,8 +88,8 @@
                   
 
                   // parsing and collecting nodes and edges from the python
-                  nodes = new vis.DataSet([]);
-                  edges = new vis.DataSet([]);
+                  nodes = new vis.DataSet([{"color": "#97c2fc", "id": "input-output pairs", "label": "input-output pairs", "shape": "dot", "title": "input-output pairs"}, {"color": "#97c2fc", "id": "input", "label": "input", "shape": "dot", "title": "input"}, {"color": "#97c2fc", "id": "is converted into", "label": "is converted into", "shape": "dot", "title": "is converted into"}, {"color": "#97c2fc", "id": "of", "label": "of", "shape": "dot", "title": "of"}, {"color": "#97c2fc", "id": "and", "label": "and", "shape": "dot", "title": "and"}, {"color": "#97c2fc", "id": "includes", "label": "includes", "shape": "dot", "title": "includes"}, {"color": "#97c2fc", "id": "examples", "label": "examples", "shape": "dot", "title": "examples"}, {"color": "#97c2fc", "id": "text-to-text-transfer-transformer", "label": "text-to-text-transfer-transformer", "shape": "dot", "title": "text-to-text-transfer-transformer"}, {"color": "#97c2fc", "id": "by", "label": "by", "shape": "dot", "title": "by"}, {"color": "#97c2fc", "id": "is strong baseline", "label": "is strong baseline", "shape": "dot", "title": "is strong baseline"}, {"color": "#97c2fc", "id": "combined with cross-encoder re-ranker", "label": "combined with cross-encoder re-ranker", "shape": "dot", "title": "combined with cross-encoder re-ranker"}, {"color": "#97c2fc", "id": "is", "label": "is", "shape": "dot", "title": "is"}, {"color": "#97c2fc", "id": "for", "label": "for", "shape": "dot", "title": "for"}, {"color": "#97c2fc", "id": "mimic-cxr", "label": "mimic-cxr", "shape": "dot", "title": "mimic-cxr"}, {"color": "#97c2fc", "id": "output", "label": "output", "shape": "dot", "title": "output"}, {"color": "#97c2fc", "id": "mimic", "label": "mimic", "shape": "dot", "title": "mimic"}, {"color": "#97c2fc", "id": "cxr", "label": "cxr", "shape": "dot", "title": "cxr"}, {"color": "#97c2fc", "id": "generate", "label": "generate", "shape": "dot", "title": "generate"}, {"color": "#97c2fc", "id": "consumer-oriented replies", "label": "consumer-oriented replies", "shape": "dot", "title": "consumer-oriented replies"}, {"color": "#97c2fc", "id": "pairs", "label": "pairs", "shape": "dot", "title": "pairs"}, {"color": "#97c2fc", "id": "de-identified", "label": "de-identified", "shape": "dot", "title": "de-identified"}, {"color": "#97c2fc", "id": "both", "label": "both", "shape": "dot", "title": "both"}]);
+                  edges = new vis.DataSet([{"from": "input", "title": "input", "to": "is converted into"}, {"from": "input", "title": "probability", "to": "of"}, {"from": "input", "title": "input", "to": "and"}, {"from": "input", "title": "input", "to": "includes"}, {"from": "input", "title": "input", "to": "examples"}, {"from": "input", "title": "input", "to": "text-to-text-transfer-transformer"}, {"from": "input", "title": "input", "to": "by"}, {"from": "input", "title": "BM25", "to": "is strong baseline"}, {"from": "input", "title": "BM25", "to": "combined with cross-encoder re-ranker"}, {"from": "input", "title": "BM25", "to": "is"}, {"from": "input", "title": "input", "to": "for"}, {"from": "generate", "title": "generate", "to": "consumer-oriented replies"}, {"from": "pairs", "title": "pairs", "to": "de-identified"}, {"from": "pairs", "title": "summaries", "to": "both"}]);
 
                   nodeColors = {};
                   allNodes = nodes.get({ returnType: "Object" });

kron/llm_predictor/KronLLMPredictor.py

@@ -1,9 +1,9 @@
 
 from typing import Any, Generator, Optional, Protocol, Tuple, runtime_checkable
 
-from llama_index import  LLMPredictor
-from llama_index.llms.utils import LLMType
-from llama_index.callbacks.base import CallbackManager
+from llama_index.core.service_context_elements.llm_predictor import  LLMPredictor
+from llama_index.core.llms.utils import LLMType
+from llama_index.core.callbacks.base import CallbackManager
 
 from kron.llm_predictor.utils import kron_resolve_llm
 
@@ -32,7 +32,7 @@ class KronLLMPredictor(LLMPredictor):
      ) -> None:
         """Initialize params."""
         self._llm = kron_resolve_llm(llm)
-        self.callback_manager = callback_manager or CallbackManager([])
+        self._llm.callback_manager = callback_manager or CallbackManager([])
 
 
 

kron/llm_predictor/KronLangChainLLM.py

@@ -1,8 +1,8 @@
-from llama_index.bridge.langchain import BaseLanguageModel, BaseChatModel
+from llama_index.core.bridge.langchain import BaseLanguageModel, BaseChatModel
 from llama_index.llms.langchain import LangChainLLM
-from llama_index.bridge.langchain import OpenAI, ChatOpenAI
+from llama_index.core.bridge.langchain import OpenAI, ChatOpenAI
 
-from llama_index.llms.base import LLMMetadata
+from llama_index.core.llms import LLMMetadata
 
 from kron.llm_predictor.openai_utils import kron_openai_modelname_to_contextsize
 

kron/llm_predictor/KronOpenAILLM.py

@@ -1,11 +1,11 @@
 from typing import Any, Awaitable, Callable, Dict, Optional, Sequence
 
-from llama_index.bridge.langchain import BaseLanguageModel, BaseChatModel
-from llama_index.llms.langchain import LangChainLLM
+from llama_index.core.bridge.langchain import BaseLanguageModel, BaseChatModel
+#from llama_index.core.llms.langchain import LangChainLLM
 from llama_index.llms.openai import OpenAI
 
-from llama_index.llms.base import (
-    LLM,
+from llama_index.core.base.llms.types import (
+    #LLM,
     ChatMessage,
     ChatResponse,
     ChatResponseAsyncGen,
@@ -16,6 +16,15 @@ from llama_index.llms.base import (
     LLMMetadata,
 )
 
+from llama_index.llms.openai.utils import (
+    from_openai_message,
+    is_chat_model,
+    is_function_calling_model,
+    openai_modelname_to_contextsize,
+    resolve_openai_credentials,
+    to_openai_message_dicts,
+)
+
 from kron.llm_predictor.openai_utils import kron_openai_modelname_to_contextsize
 
 class KronOpenAI(OpenAI):
@@ -25,7 +34,9 @@ class KronOpenAI(OpenAI):
         return LLMMetadata(
             context_window=kron_openai_modelname_to_contextsize(self.model),
             num_output=self.max_tokens or -1,
-            is_chat_model=self._is_chat_model,
+            is_chat_model=is_chat_model(model=self._get_model_name()),
+            is_function_calling_model=is_function_calling_model(model=self._get_model_name()),
+            model_name=self.model,
         )
     
     def complete(self, prompt: str, **kwargs: Any) -> CompletionResponse:

kron/llm_predictor/openai_utils.py

@@ -37,12 +37,14 @@ TURBO_MODELS = {
     "gpt-3.5-turbo-16k-0613": 16384,
     # 0301 models
     "gpt-3.5-turbo-0301": 4096,
+    # turbo models
+    "gpt-3.5-turbo-instruct": 4096
 }
 
-GPT3_5_MODELS = {
-    "text-davinci-003": 4097,
-    "text-davinci-002": 4097,
-}
+#GPT3_5_MODELS = {
+#    "text-davinci-003": 4097,
+#    "text-davinci-002": 4097,
+#}
 
 GPT3_MODELS = {
     "text-ada-001": 2049,
@@ -57,7 +59,7 @@ GPT3_MODELS = {
 ALL_AVAILABLE_MODELS = {
     **GPT4_MODELS,
     **TURBO_MODELS,
-    **GPT3_5_MODELS,
+#    **GPT3_5_MODELS,
     **GPT3_MODELS,
     **LOCAL_MODELS,
 }

kron/llm_predictor/utils.py

@@ -1,11 +1,11 @@
 from typing import Optional, Union
-from llama_index.llms.base import LLM
+from llama_index.core.llms.llm import LLM
 from langchain.base_language import BaseLanguageModel
 
 from kron.llm_predictor.KronLangChainLLM import KronLangChainLLM
 from llama_index.llms.openai import OpenAI
 
-from llama_index.llms.utils import LLMType
+from llama_index.core.llms.utils import LLMType
 
 
 def kron_resolve_llm(llm: Optional[LLMType] = None) -> LLM:

measurable.py

@@ -18,7 +18,7 @@ def display_wordcloud(answer, answer_str):
         all_reference_texts = ''
         for nodewithscore in answer.source_nodes:  
             node = nodewithscore.node
-            from llama_index.schema import NodeRelationship
+            from llama_index.core.schema import NodeRelationship
                     #if NodeRelationship.SOURCE in node.relationships:
             all_reference_texts = all_reference_texts + '\n' + node.text
         wordcloud_r = wordcloud.generate(all_reference_texts)

requirements.txt

@@ -7,13 +7,15 @@
 #streamlit run appname.py
 
 torch
-transformers
+transformers>=4.34
 llama_index
+llama-index-llms-langchain
+langchain
 pyvis
 nltk
 python-dotenv
 cohere
-baseten
+#baseten
 st-star-rating
 wordcloud
 gensim
@@ -21,3 +23,5 @@ amazon-dax-client>=1.1.7
 boto3>=1.26.79
 pytest>=7.2.1
 requests>=2.28.2
+packaging>20.0
+langchain-core>=0.1

storage/Arylwen-instruct-palmyra-20b-gptq-8-default-no-coref/default__vector_store.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d17ed74c1649a438e518a8dc56a7772913dfe1ea7a7605bce069c63872431455
+size 72

storage/Arylwen-instruct-palmyra-20b-gptq-8-default-no-coref/docstore.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:362322a55d70bbd5957c4a46426ba99b249b776b1ddeb41f1744bc389892a2be
+size 38549003

storage/Arylwen-instruct-palmyra-20b-gptq-8-default-no-coref/graph_store.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36417c5a46c4748fd020428edcedfd267f5298e752d3bef5ce750c3fc301d209
+size 2419238

storage/Arylwen-instruct-palmyra-20b-gptq-8-default-no-coref/index_store.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce19d43152283102b16ec8d7f1eb0216482f6134f6acba8ab098ef3fd2e45071
+size 4573376