Create handler.py
Browse files- handler.py +173 -0
handler.py
ADDED
@@ -0,0 +1,173 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import locale
|
3 |
+
import os
|
4 |
+
from typing import Dict, List, Any
|
5 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
6 |
+
from langchain.llms import HuggingFacePipeline
|
7 |
+
from langchain.retrievers.document_compressors import LLMChainExtractor
|
8 |
+
from langchain.retrievers import ContextualCompressionRetriever
|
9 |
+
from langchain.vectorstores import Chroma
|
10 |
+
from langchain import PromptTemplate, LLMChain
|
11 |
+
from langchain.chains import RetrievalQA, ConversationalRetrievalChain
|
12 |
+
from langchain.prompts import PromptTemplate
|
13 |
+
from langchain.prompts.prompt import PromptTemplate
|
14 |
+
from langchain.memory import ConversationBufferMemory
|
15 |
+
from langchain.embeddings import HuggingFaceBgeEmbeddings
|
16 |
+
from langchain.document_loaders import WebBaseLoader
|
17 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
18 |
+
from langchain.chains.qa_with_sources import load_qa_with_sources_chain
|
19 |
+
from langchain.chains.combine_documents import create_stuff_documents_chain
|
20 |
+
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
21 |
+
from langchain_core.messages import HumanMessage
|
22 |
+
from langchain_core.output_parsers import StrOutputParser
|
23 |
+
from langchain_core.runnables import RunnableLambda, RunnableBranch, RunnablePassthrough
|
24 |
+
from operator import itemgetter
|
25 |
+
from langchain.schema import format_document
|
26 |
+
from langchain.memory import ConversationBufferMemory
|
27 |
+
from langchain_core.messages import AIMessage, HumanMessage, get_buffer_string
|
28 |
+
|
29 |
+
class EndpointHandler():
|
30 |
+
def __init__(self, path=""):
|
31 |
+
|
32 |
+
# Config LangChain
|
33 |
+
# os.environ["LANGCHAIN_TRACING_V2"] = "true"
|
34 |
+
# os.environ["LANGCHAIN_API_KEY"] = "ls__9834e6b2ff094d43a28418c9ecea2fd5"
|
35 |
+
|
36 |
+
# Create LLM
|
37 |
+
model_id = path
|
38 |
+
|
39 |
+
model = AutoModelForCausalLM.from_pretrained(
|
40 |
+
model_id,
|
41 |
+
device_map='auto',
|
42 |
+
torch_dtype=torch.float16,
|
43 |
+
load_in_8bit=True
|
44 |
+
)
|
45 |
+
model.eval()
|
46 |
+
|
47 |
+
# model_kwargs = {
|
48 |
+
# "input_ids":input_ids,
|
49 |
+
# "max_new_tokens":1024,
|
50 |
+
# "do_sample":True,
|
51 |
+
# "top_k":50,
|
52 |
+
# "top_p":self.top_p,
|
53 |
+
# "temperature":self.temperature,
|
54 |
+
# "repetition_penalty":1.2,
|
55 |
+
# "eos_token_id":self.tokenizer.eos_token_id,
|
56 |
+
# "bos_token_id":self.tokenizer.bos_token_id,
|
57 |
+
# "pad_token_id":self.tokenizer.pad_token_id
|
58 |
+
# }
|
59 |
+
|
60 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
61 |
+
model_id,
|
62 |
+
)
|
63 |
+
tokenizer.pad_token = tokenizer.eos_token
|
64 |
+
|
65 |
+
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=1024)
|
66 |
+
chat = HuggingFacePipeline(pipeline=pipe)
|
67 |
+
|
68 |
+
# Create Text-Embedding Model
|
69 |
+
embedding_function = HuggingFaceBgeEmbeddings(
|
70 |
+
model_name="DMetaSoul/Dmeta-embedding",
|
71 |
+
model_kwargs={'device': 'cuda'},
|
72 |
+
encode_kwargs={'normalize_embeddings': True}
|
73 |
+
)
|
74 |
+
|
75 |
+
# Load Vector db
|
76 |
+
urls = [
|
77 |
+
"https://www.wenweipo.com/epaper/view/newsDetail/1582436861224292352.html",
|
78 |
+
"https://www.thinkhk.com/article/2023-03/24/59874.html"
|
79 |
+
]
|
80 |
+
|
81 |
+
loader = WebBaseLoader(urls)
|
82 |
+
data = loader.load()
|
83 |
+
|
84 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size = 1000, chunk_overlap = 16)
|
85 |
+
all_splits = text_splitter.split_documents(data)
|
86 |
+
|
87 |
+
vectorstore = Chroma.from_documents(documents=all_splits, embedding=embedding_function)
|
88 |
+
retriever = vectorstore.as_retriever()
|
89 |
+
|
90 |
+
compressor = LLMChainExtractor.from_llm(chat)
|
91 |
+
compression_retriever = ContextualCompressionRetriever(
|
92 |
+
base_compressor=compressor, base_retriever=retriever
|
93 |
+
)
|
94 |
+
|
95 |
+
_template = """[INST] Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question, in its original language.
|
96 |
+
Chat History:
|
97 |
+
{chat_history}
|
98 |
+
Follow Up Input: {question}
|
99 |
+
Standalone question: [/INST]"""
|
100 |
+
|
101 |
+
CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)
|
102 |
+
|
103 |
+
template = """[INST] Answer the question based only on the following context:
|
104 |
+
{context}
|
105 |
+
|
106 |
+
Question: {question} [/INST]
|
107 |
+
"""
|
108 |
+
|
109 |
+
ANSWER_PROMPT = ChatPromptTemplate.from_template(template)
|
110 |
+
|
111 |
+
self.memory = ConversationBufferMemory(
|
112 |
+
return_messages=True, output_key="answer", input_key="question"
|
113 |
+
)
|
114 |
+
|
115 |
+
# First we add a step to load memory
|
116 |
+
# This adds a "memory" key to the input object
|
117 |
+
loaded_memory = RunnablePassthrough.assign(
|
118 |
+
chat_history=RunnableLambda(self.memory.load_memory_variables) | itemgetter("history"),
|
119 |
+
)
|
120 |
+
# Now we calculate the standalone question
|
121 |
+
standalone_question = {
|
122 |
+
"standalone_question": {
|
123 |
+
"question": lambda x: x["question"],
|
124 |
+
"chat_history": lambda x: get_buffer_string(x["chat_history"]),
|
125 |
+
}
|
126 |
+
| CONDENSE_QUESTION_PROMPT
|
127 |
+
| chat
|
128 |
+
| StrOutputParser(),
|
129 |
+
}
|
130 |
+
|
131 |
+
DEFAULT_DOCUMENT_PROMPT = PromptTemplate.from_template(template="{page_content}")
|
132 |
+
|
133 |
+
def _combine_documents(
|
134 |
+
docs, document_prompt=DEFAULT_DOCUMENT_PROMPT, document_separator="\n\n"
|
135 |
+
):
|
136 |
+
doc_strings = [format_document(doc, document_prompt) for doc in docs]
|
137 |
+
return document_separator.join(doc_strings)
|
138 |
+
|
139 |
+
# Now we retrieve the documents
|
140 |
+
retrieved_documents = {
|
141 |
+
"docs": itemgetter("standalone_question") | retriever,
|
142 |
+
"question": lambda x: x["standalone_question"],
|
143 |
+
}
|
144 |
+
# Now we construct the inputs for the final prompt
|
145 |
+
final_inputs = {
|
146 |
+
"context": lambda x: _combine_documents(x["docs"]),
|
147 |
+
"question": itemgetter("question"),
|
148 |
+
}
|
149 |
+
# And finally, we do the part that returns the answers
|
150 |
+
answer = {
|
151 |
+
"answer": final_inputs | ANSWER_PROMPT | chat,
|
152 |
+
"docs": itemgetter("docs"),
|
153 |
+
}
|
154 |
+
# And now we put it all together!
|
155 |
+
self.final_chain = loaded_memory | standalone_question | retrieved_documents | answer
|
156 |
+
|
157 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
158 |
+
# get inputs
|
159 |
+
inputs = data.pop("inputs",data)
|
160 |
+
date = data.pop("date", None)
|
161 |
+
|
162 |
+
result = self.final_chain.invoke({"question": inputs})
|
163 |
+
|
164 |
+
answer = result['answer']
|
165 |
+
|
166 |
+
# Note that the memory does not save automatically
|
167 |
+
# This will be improved in the future
|
168 |
+
# For now you need to save it yourself
|
169 |
+
# self.memory.save_context(inputs, {"answer": answer})
|
170 |
+
self.memory.load_memory_variables({})
|
171 |
+
|
172 |
+
return answer
|
173 |
+
|