Spaces:
Sleeping
Sleeping
Asaad Almutareb
commited on
Commit
·
8a93d38
1
Parent(s):
7d8f356
added stuff
Browse files- app.py +171 -13
- lc-embeddings.py +0 -0
app.py
CHANGED
@@ -3,9 +3,162 @@ from langchain.chains import RetrievalQAWithSourcesChain
|
|
3 |
|
4 |
# gradio
|
5 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
global qa
|
8 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
|
11 |
#####
|
@@ -21,24 +174,29 @@ def create_gradio_interface(qa:RetrievalQAWithSourcesChain):
|
|
21 |
history = history + [(text, None)]
|
22 |
return history, ""
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
def infer(question, history):
|
35 |
query = question
|
36 |
result = qa({"query": query, "history": history, "question": question})
|
37 |
return result
|
38 |
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
|
43 |
title = """
|
44 |
<div style="text-align: center;max-width: 1920px;">
|
|
|
3 |
|
4 |
# gradio
|
5 |
import gradio as gr
|
6 |
+
#import random
|
7 |
+
import time
|
8 |
+
#boto3 for S3 access
|
9 |
+
import boto3
|
10 |
+
from botocore import UNSIGNED
|
11 |
+
from botocore.client import Config
|
12 |
+
# access .env file
|
13 |
+
import os
|
14 |
+
from dotenv import load_dotenv
|
15 |
+
#from bs4 import BeautifulSoup
|
16 |
+
# HF libraries
|
17 |
+
from langchain.llms import HuggingFaceHub
|
18 |
+
from langchain.embeddings import HuggingFaceHubEmbeddings
|
19 |
+
# vectorestore
|
20 |
+
from langchain.vectorstores import Chroma
|
21 |
+
#from langchain.vectorstores import FAISS
|
22 |
+
# retrieval chain
|
23 |
+
from langchain.chains import RetrievalQA
|
24 |
+
# from langchain.chains import RetrievalQAWithSourcesChain
|
25 |
+
# prompt template
|
26 |
+
from langchain.prompts import PromptTemplate
|
27 |
+
from langchain.memory import ConversationBufferMemory
|
28 |
+
# logging
|
29 |
+
import logging
|
30 |
+
#import zipfile
|
31 |
+
# improve results with retriever
|
32 |
+
# from langchain.retrievers import ContextualCompressionRetriever
|
33 |
+
# from langchain.retrievers.document_compressors import LLMChainExtractor
|
34 |
+
# from langchain.retrievers.document_compressors import EmbeddingsFilter
|
35 |
+
# from langchain.retrievers.multi_query import MultiQueryRetriever
|
36 |
+
# from langchain.retrievers import BM25Retriever, EnsembleRetriever
|
37 |
+
# reorder retrived documents
|
38 |
+
# from langchain.document_transformers import LongContextReorder
|
39 |
+
# github issues
|
40 |
+
#from langchain.document_loaders import GitHubIssuesLoader
|
41 |
+
# debugging
|
42 |
+
from langchain.globals import set_verbose
|
43 |
+
# caching
|
44 |
+
from langchain.globals import set_llm_cache
|
45 |
+
#from langchain.cache import InMemoryCache
|
46 |
+
# We can do the same thing with a SQLite cache
|
47 |
+
from langchain.cache import SQLiteCache
|
48 |
+
#set_llm_cache(InMemoryCache())
|
49 |
+
|
50 |
+
set_verbose(True)
|
51 |
+
|
52 |
+
# load .env variables
|
53 |
+
config = load_dotenv(".env")
|
54 |
+
HUGGINGFACEHUB_API_TOKEN=os.getenv('HUGGINGFACEHUB_API_TOKEN')
|
55 |
+
AWS_S3_LOCATION=os.getenv('AWS_S3_LOCATION')
|
56 |
+
AWS_S3_FILE=os.getenv('AWS_S3_FILE')
|
57 |
+
VS_DESTINATION=os.getenv('VS_DESTINATION')
|
58 |
+
|
59 |
+
# initialize Model config
|
60 |
+
# HuggingFaceH4/zephyr-7b-beta
|
61 |
+
# mistralai/Mistral-7B-Instruct-v0.1
|
62 |
+
model_id = HuggingFaceHub(repo_id="HuggingFaceH4/zephyr-7b-beta", model_kwargs={
|
63 |
+
"temperature":0.1,
|
64 |
+
"max_new_tokens":1024,
|
65 |
+
"repetition_penalty":1.2,
|
66 |
+
# "streaming": True,
|
67 |
+
# "return_full_text":True
|
68 |
+
})
|
69 |
+
|
70 |
+
#model_name = "sentence-transformers/multi-qa-mpnet-base-dot-v1"
|
71 |
+
model_name = "sentence-transformers/all-mpnet-base-v2"
|
72 |
+
embeddings = HuggingFaceHubEmbeddings(repo_id=model_name)
|
73 |
+
|
74 |
+
# remove old vectorstore
|
75 |
+
if os.path.exists(VS_DESTINATION):
|
76 |
+
os.remove(VS_DESTINATION)
|
77 |
+
|
78 |
+
# remove old sqlite cache
|
79 |
+
if os.path.exists('.langchain.sqlite'):
|
80 |
+
os.remove('.langchain.sqlite')
|
81 |
+
|
82 |
+
set_llm_cache(SQLiteCache(database_path=".langchain.sqlite"))
|
83 |
+
|
84 |
+
# retrieve vectorsrore
|
85 |
+
s3 = boto3.client('s3', config=Config(signature_version=UNSIGNED))
|
86 |
+
|
87 |
+
## Chroma DB
|
88 |
+
s3.download_file(AWS_S3_LOCATION, AWS_S3_FILE, VS_DESTINATION)
|
89 |
+
# use the cached embeddings instead of embeddings to speed up re-retrival
|
90 |
+
db = Chroma(persist_directory="./vectorstore", embedding_function=embeddings)
|
91 |
+
db.get()
|
92 |
+
|
93 |
+
## FAISS DB
|
94 |
+
# s3.download_file('rad-rag-demos', 'vectorstores/faiss_db_ray.zip', './chroma_db/faiss_db_ray.zip')
|
95 |
+
# with zipfile.ZipFile('./chroma_db/faiss_db_ray.zip', 'r') as zip_ref:
|
96 |
+
# zip_ref.extractall('./chroma_db/')
|
97 |
+
|
98 |
+
# FAISS_INDEX_PATH='./chroma_db/faiss_db_ray'
|
99 |
+
# db = FAISS.load_local(FAISS_INDEX_PATH, embeddings)
|
100 |
+
|
101 |
+
# initialize the bm25 retriever and chroma/faiss retriever
|
102 |
+
# bm25_retriever = BM25Retriever.
|
103 |
+
# bm25_retriever.k = 2
|
104 |
+
|
105 |
+
# Retrieve more documents with higher diversity useful if your dataset has many similar documents
|
106 |
+
retriever = db.as_retriever(search_type="mmr")#, search_kwargs={'k': 3, 'lambda_mult': 0.25})
|
107 |
+
# Above a certain threshold
|
108 |
+
# retriever = db.as_retriever(
|
109 |
+
# search_type="similarity_score_threshold",
|
110 |
+
# search_kwargs={'score_threshold': 0.6}
|
111 |
+
# )
|
112 |
+
|
113 |
+
# # asks LLM to create 3 alternatives baed on user query
|
114 |
+
# multi_retriever = MultiQueryRetriever.from_llm(retriever=retriever, llm=model_id)
|
115 |
+
|
116 |
+
# # asks LLM to extract relevant parts from retrieved documents
|
117 |
+
# compressor = LLMChainExtractor.from_llm(model_id)
|
118 |
+
# compression_retriever = ContextualCompressionRetriever(base_compressor=compressor, base_retriever=multi_retriever)
|
119 |
|
120 |
global qa
|
121 |
+
template = """
|
122 |
+
You are the friendly documentation AI buddy Arti, who helps the Human in using RAY, the open-source unified framework for scaling AI and Python applications.
|
123 |
+
Use the following context (delimited by <ctx></ctx>) and the chat history (delimited by <hs></hs>) to answer the question :
|
124 |
+
------
|
125 |
+
<ctx>
|
126 |
+
{context}
|
127 |
+
</ctx>
|
128 |
+
------
|
129 |
+
<hs>
|
130 |
+
{history}
|
131 |
+
</hs>
|
132 |
+
------
|
133 |
+
{question}
|
134 |
+
Answer:
|
135 |
+
"""
|
136 |
+
prompt = PromptTemplate(
|
137 |
+
input_variables=["history", "context", "question"],
|
138 |
+
template=template,
|
139 |
+
)
|
140 |
+
memory = ConversationBufferMemory(memory_key="history", input_key="question")
|
141 |
+
|
142 |
+
# logging for the chain
|
143 |
+
logging.basicConfig()
|
144 |
+
logging.getLogger("langchain.retrievers").setLevel(logging.INFO)
|
145 |
+
logging.getLogger("langchain.chains").setLevel(logging.INFO)
|
146 |
+
|
147 |
+
|
148 |
+
|
149 |
+
qa = RetrievalQA.from_chain_type(llm=model_id, retriever=retriever, return_source_documents=True, verbose=True, chain_type_kwargs={
|
150 |
+
"verbose": True,
|
151 |
+
"memory": memory,
|
152 |
+
"prompt": prompt
|
153 |
+
}
|
154 |
+
)
|
155 |
+
# qa = RetrievalQAWithSourcesChain.from_chain_type(llm=model_id, retriever=retriever, return_source_documents=True, verbose=True, chain_type_kwargs={
|
156 |
+
# "verbose": True,
|
157 |
+
# "memory": memory,
|
158 |
+
# "prompt": prompt,
|
159 |
+
# "document_variable_name": "context"
|
160 |
+
# }
|
161 |
+
# )
|
162 |
|
163 |
|
164 |
#####
|
|
|
174 |
history = history + [(text, None)]
|
175 |
return history, ""
|
176 |
|
177 |
+
def bot(history):
|
178 |
+
response = infer(history[-1][0], history)
|
179 |
+
sources = [doc.metadata.get("source") for doc in response['source_documents']]
|
180 |
+
src_list = '\n'.join(sources)
|
181 |
+
print_this = response['result'] + "\n\n\n Sources: \n\n\n" + src_list
|
182 |
+
|
183 |
+
history[-1][1] = ""
|
184 |
+
for character in response['result']: #print_this: #
|
185 |
+
#print_this:
|
186 |
+
history[-1][1] += character
|
187 |
+
time.sleep(0.01)
|
188 |
+
yield history
|
189 |
+
# history[-1][1] = print_this #response['answer']
|
190 |
+
# return history
|
191 |
|
192 |
def infer(question, history):
|
193 |
query = question
|
194 |
result = qa({"query": query, "history": history, "question": question})
|
195 |
return result
|
196 |
|
197 |
+
css="""
|
198 |
+
#col-container {min-width: 800px; max-width: 1920px; margin-left: auto; margin-right: auto;}
|
199 |
+
"""
|
200 |
|
201 |
title = """
|
202 |
<div style="text-align: center;max-width: 1920px;">
|
lc-embeddings.py
ADDED
File without changes
|