Spaces:
Runtime error
Runtime error
File size: 9,371 Bytes
027bfbf |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 |
"""
Python Backend API to chat with private data
08/14/2023
D.M. Theekshana Samaradiwakara
"""
import os
import time
from dotenv import load_dotenv
from langchain.chains import RetrievalQA
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from langchain.llms import GPT4All
from langchain.llms import HuggingFaceHub
from langchain.chat_models import ChatOpenAI
# from langchain.retrievers._query.base import SelfQueryRetriever
# from langchain.chains.query_constructor.base import AttributeInfo
# from chromaDb import load_store
from faissDb import load_FAISS_store
from langchain.agents import ZeroShotAgent, Tool, AgentExecutor
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain, ConversationalRetrievalChain
from conversationBufferWindowMemory import ConversationBufferWindowMemory
from langchain.memory import ReadOnlySharedMemory
load_dotenv()
#gpt4 all model
gpt4all_model_path = os.environ.get('GPT4ALL_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))
openai_api_key = os.environ.get('OPENAI_API_KEY')
verbose = os.environ.get('VERBOSE')
# activate/deactivate the streaming StdOut callback for LLMs
callbacks = [StreamingStdOutCallbackHandler()]
memory = ConversationBufferWindowMemory(
memory_key="chat_history",
input_key="question",
return_messages=True,
k=3
)
readonlymemory = ReadOnlySharedMemory(memory=memory)
print("\n\n> Initializing QAPipeline:")
global llm_name
llm_name = 'None'
global llm
llm = 'None'
global dataset_name
dataset_name = 'None'
global vectorstore
vectorstore = 'None'
qa_chain = None
agent = None
def run(query, model, dataset):
if (llm_name != model) or (dataset_name != dataset) or (qa_chain == None):
set_model(model)
set_vectorstore(dataset)
set_qa_chain()
# Get the answer from the chain
start = time.time()
res = qa_chain(query)
# answer, docs = res['result'],res['source_documents']
end = time.time()
# Print the result
print("\n\n> Question:")
print(query)
print(f"\n> Answer (took {round(end - start, 2)} s.):")
print( res)
return res
def run_agent(query, model, dataset):
try:
if (llm_name != model) or (dataset_name != dataset) or (agent == None):
set_model(model)
set_vectorstore(dataset)
set_qa_chain_with_agent()
# Get the answer from the chain
start = time.time()
res = agent(query)
# answer, docs = res['result'],res['source_documents']
end = time.time()
# Print the result
print("\n\n> Question:")
print(query)
print(f"\n> Answer (took {round(end - start, 2)} s.):")
print( res)
return res["output"]
except Exception as e:
# logger.error(f"Answer retrieval failed with {e}")
print(f"> QAPipeline run_agent Error : {e}")#, icon=":books:")
return
def set_model(model_type):
if model_type != llm_name:
global llm
match model_type:
case "gpt4all":
# llm = GPT4All(model=gpt4all_model_path, n_ctx=model_n_ctx, backend='gptj', n_batch=model_n_batch, callbacks=callbacks, verbose=verbose)
llm = GPT4All(model=gpt4all_model_path, max_tokens=model_n_ctx, backend='gptj', n_batch=model_n_batch, callbacks=callbacks, verbose=verbose)
# llm = HuggingFaceHub(repo_id="nomic-ai/gpt4all-j", model_kwargs={"temperature":0.001, "max_length":1024})
case "google/flan-t5-xxl":
llm = HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":0.001, "max_length":1024})
case "tiiuae/falcon-7b-instruct":
llm = HuggingFaceHub(repo_id=model_type, model_kwargs={"temperature":0.001, "max_length":1024})
case "openai":
llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
case _default:
# raise exception if model_type is not supported
raise Exception(f"Model type {model_type} is not supported. Please choose a valid one")
# global llm_name
llm_name = model_type
def set_vectorstore( dataset):
if dataset != dataset_name:
# vectorstore = load_store(dataset)
global vectorstore
vectorstore = load_FAISS_store()
print("\n\n> vectorstore loaded:")
dataset_name = dataset
def set_qa_chain():
global qa_chain
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever = vectorstore.as_retriever(),
# retriever = vectorstore.as_retriever(search_kwargs={"k": target_source_chunks}
return_source_documents= True
)
def set_qa_chain_with_agent():
try:
# Define a custom prompt
general_qa_template = (
"""You can have a general conversation with the users like greetings.
Continue the conversation and only answer questions related to banking sector like financial and legal.
If you dont know the answer say you dont know, dont try to makeup answers.
Conversation: {chat_history}
Question: {question}
"""
)
general_qa_chain_prompt = PromptTemplate(input_variables=["question", "chat_history"], template=general_qa_template)
general_qa_chain = LLMChain(
llm=llm,
prompt=general_qa_chain_prompt,
verbose=True,
memory=readonlymemory, # use the read-only memory to prevent the tool from modifying the memory
)
general_qa_chain_tool = Tool(
name="general qa",
func= general_qa_chain.run,
description='''useful for when you need to have a general conversation with the users like greetings
or to answer general purpose questions related to banking sector like financial and legal.
Input should be a fully formed question.''',
return_direct=True,
)
# Define a custom prompt
retrieval_qa_template = (
"""
please answer the question based on the chat history and context with the latest information.
You have provided context information below related to central bank acts published in various years.
The content of a bank act can updated by a bank act from a latest year.
If you dont know the answer say you dont know, dont try to makeup answers.
Conversation: {chat_history}
Context: {context}
Question : {question}
"""
)
retrieval_qa_chain_prompt = PromptTemplate(
input_variables=["question", "context", "chat_history"],
template=retrieval_qa_template
)
bank_regulations_qa = ConversationalRetrievalChain.from_llm(
llm=llm,
chain_type="stuff",
retriever = vectorstore.as_retriever(),
# retriever = vectorstore.as_retriever(search_kwargs={"k": target_source_chunks}
return_source_documents= True,
get_chat_history=lambda h : h,
combine_docs_chain_kwargs={"prompt": retrieval_qa_chain_prompt},
verbose=True,
memory=readonlymemory, # use the read-only memory to prevent the tool from modifying the memory
)
bank_regulations_qa_tool = Tool(
name="bank regulations",
func= lambda question: bank_regulations_qa({"question": question}),
description='''useful for when you need to answer questions about
financial and legal information issued from central bank regarding banks and bank regulations.
Input should be a fully formed question.''',
return_direct=True,
)
tools = [
bank_regulations_qa_tool,
general_qa_chain_tool
]
prefix = """Have a conversation with a human, answering the following questions as best you can. You have access to the following tools:"""
suffix = """Begin!"
{chat_history}
Question: {question}
{agent_scratchpad}"""
agent_prompt = ZeroShotAgent.create_prompt(
tools,
prefix=prefix,
suffix=suffix,
input_variables=["question", "chat_history", "agent_scratchpad"],
)
llm_chain = LLMChain(llm=llm, prompt=agent_prompt)
zeroShotAgent = ZeroShotAgent(
llm_chain=llm_chain,
tools=tools,
verbose=True,
)
agent_chain = AgentExecutor.from_agent_and_tools(
agent=zeroShotAgent,
tools=tools,
verbose=True,
memory=memory,
handle_parsing_errors=True,
)
global agent
agent = agent_chain
print(f"\n> agent_chain created")
except Exception as e:
# logger.error(f"Answer retrieval failed with {e}")
print(f"> QAPipeline set_qa_chain_with_agent Error : {e}")#, icon=":books:")
return
|