Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -0,0 +1,205 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#####################################
|
2 |
+
## BitsAndBytes
|
3 |
+
#####################################
|
4 |
+
|
5 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
6 |
+
|
7 |
+
model_name = "bn22/Mistral-7B-Instruct-v0.1-sharded"
|
8 |
+
|
9 |
+
###### other models:
|
10 |
+
# "Trelis/Llama-2-7b-chat-hf-sharded-bf16"
|
11 |
+
# "bn22/Mistral-7B-Instruct-v0.1-sharded"
|
12 |
+
# "HuggingFaceH4/zephyr-7b-beta"
|
13 |
+
|
14 |
+
# function for loading 4-bit quantized model
|
15 |
+
def load_quantized_model(model_name: str):
|
16 |
+
"""
|
17 |
+
:param model_name: Name or path of the model to be loaded.
|
18 |
+
:return: Loaded quantized model.
|
19 |
+
"""
|
20 |
+
bnb_config = BitsAndBytesConfig(
|
21 |
+
load_in_4bit=True,
|
22 |
+
bnb_4bit_use_double_quant=True,
|
23 |
+
bnb_4bit_quant_type="nf4",
|
24 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
25 |
+
)
|
26 |
+
|
27 |
+
model = AutoModelForCausalLM.from_pretrained(
|
28 |
+
model_name,
|
29 |
+
load_in_4bit=True,
|
30 |
+
torch_dtype=torch.bfloat16,
|
31 |
+
quantization_config=bnb_config
|
32 |
+
)
|
33 |
+
return model
|
34 |
+
|
35 |
+
##################################################
|
36 |
+
## vs chat
|
37 |
+
##################################################
|
38 |
+
import torch
|
39 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
|
40 |
+
|
41 |
+
from langchain_core.messages import AIMessage, HumanMessage
|
42 |
+
from langchain_community.document_loaders import WebBaseLoader
|
43 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
44 |
+
from langchain_community.vectorstores import Chroma
|
45 |
+
|
46 |
+
#from langchain_openai import OpenAIEmbeddings, ChatOpenAI
|
47 |
+
from langchain.embeddings import HuggingFaceBgeEmbeddings
|
48 |
+
from langchain.vectorstores.faiss import FAISS
|
49 |
+
|
50 |
+
|
51 |
+
from dotenv import load_dotenv
|
52 |
+
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
|
53 |
+
from langchain.chains import create_history_aware_retriever, create_retrieval_chain
|
54 |
+
from langchain.chains.combine_documents import create_stuff_documents_chain
|
55 |
+
|
56 |
+
|
57 |
+
load_dotenv()
|
58 |
+
|
59 |
+
def get_vectorstore_from_url(url):
|
60 |
+
# get the text in document form
|
61 |
+
loader = WebBaseLoader(url)
|
62 |
+
document = loader.load()
|
63 |
+
|
64 |
+
# split the document into chunks
|
65 |
+
text_splitter = RecursiveCharacterTextSplitter()
|
66 |
+
document_chunks = text_splitter.split_documents(document)
|
67 |
+
#######
|
68 |
+
'''
|
69 |
+
FAISS
|
70 |
+
A FAISS vector store containing the embeddings of the text chunks.
|
71 |
+
'''
|
72 |
+
model = "BAAI/bge-base-en-v1.5"
|
73 |
+
encode_kwargs = {
|
74 |
+
"normalize_embeddings": True
|
75 |
+
} # set True to compute cosine similarity
|
76 |
+
embeddings = HuggingFaceBgeEmbeddings(
|
77 |
+
model_name=model, encode_kwargs=encode_kwargs, model_kwargs={"device": "cpu"}
|
78 |
+
)
|
79 |
+
# load from disk
|
80 |
+
vector_store = Chroma(persist_directory="./chroma_db", embedding_function=embeddings)
|
81 |
+
|
82 |
+
#vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
|
83 |
+
vector_store = Chroma.from_documents(document_chunks, embeddings, persist_directory="./chroma_db")
|
84 |
+
|
85 |
+
|
86 |
+
|
87 |
+
|
88 |
+
print("-----")
|
89 |
+
print(vector_store.similarity_search("What is ALiBi?"))
|
90 |
+
print("-----")
|
91 |
+
|
92 |
+
#######
|
93 |
+
# create a vectorstore from the chunks
|
94 |
+
|
95 |
+
return vector_store
|
96 |
+
|
97 |
+
|
98 |
+
|
99 |
+
|
100 |
+
|
101 |
+
def get_context_retriever_chain(vector_store):
|
102 |
+
|
103 |
+
# specify model huggingface mode name
|
104 |
+
model_name = "anakin87/zephyr-7b-alpha-sharded"
|
105 |
+
# model_name = "bn22/Mistral-7B-Instruct-v0.1-sharded"
|
106 |
+
|
107 |
+
###### other models:
|
108 |
+
# "Trelis/Llama-2-7b-chat-hf-sharded-bf16"
|
109 |
+
# "bn22/Mistral-7B-Instruct-v0.1-sharded"
|
110 |
+
# "HuggingFaceH4/zephyr-7b-beta"
|
111 |
+
|
112 |
+
# function for loading 4-bit quantized model
|
113 |
+
|
114 |
+
|
115 |
+
llm = load_quantized_model(model_name)
|
116 |
+
|
117 |
+
retriever = vector_store.as_retriever()
|
118 |
+
|
119 |
+
prompt = ChatPromptTemplate.from_messages([
|
120 |
+
MessagesPlaceholder(variable_name="chat_history"),
|
121 |
+
("user", "{input}"),
|
122 |
+
("user", "Given the above conversation, generate a search query to look up in order to get information relevant to the conversation")
|
123 |
+
])
|
124 |
+
|
125 |
+
retriever_chain = create_history_aware_retriever(llm, retriever, prompt)
|
126 |
+
|
127 |
+
return retriever_chain
|
128 |
+
|
129 |
+
def get_conversational_rag_chain(retriever_chain):
|
130 |
+
|
131 |
+
llm = load_quantized_model(model_name)
|
132 |
+
|
133 |
+
prompt = ChatPromptTemplate.from_messages([
|
134 |
+
("system", "Answer the user's questions based on the below context:\n\n{context}"),
|
135 |
+
MessagesPlaceholder(variable_name="chat_history"),
|
136 |
+
("user", "{input}"),
|
137 |
+
])
|
138 |
+
|
139 |
+
stuff_documents_chain = create_stuff_documents_chain(llm,prompt)
|
140 |
+
|
141 |
+
return create_retrieval_chain(retriever_chain, stuff_documents_chain)
|
142 |
+
|
143 |
+
def get_response(user_input):
|
144 |
+
retriever_chain = get_context_retriever_chain(st.session_state.vector_store)
|
145 |
+
conversation_rag_chain = get_conversational_rag_chain(retriever_chain)
|
146 |
+
|
147 |
+
response = conversation_rag_chain.invoke({
|
148 |
+
"chat_history": st.session_state.chat_history,
|
149 |
+
"input": user_query
|
150 |
+
})
|
151 |
+
|
152 |
+
return response['answer']
|
153 |
+
|
154 |
+
|
155 |
+
|
156 |
+
###################
|
157 |
+
|
158 |
+
###################
|
159 |
+
import gradio as gr
|
160 |
+
|
161 |
+
##from langchain_core.runnables.base import ChatPromptValue
|
162 |
+
#from torch import tensor
|
163 |
+
|
164 |
+
# Create Gradio interface
|
165 |
+
#vector_store = None # Set your vector store here
|
166 |
+
chat_history = [] # Set your chat history here
|
167 |
+
|
168 |
+
# Define your function here
|
169 |
+
def get_response(user_input):
|
170 |
+
|
171 |
+
# Define the prompt as a ChatPromptValue object
|
172 |
+
#user_input = ChatPromptValue(user_input)
|
173 |
+
|
174 |
+
# Convert the prompt to a tensor
|
175 |
+
#input_ids = user_input.tensor
|
176 |
+
|
177 |
+
|
178 |
+
#vs = get_vectorstore_from_url(user_url, all_domain)
|
179 |
+
vs = get_vectorstore_from_url("https://www.bofrost.de/shop/laenderkueche_5573/italienische-kueche_5576/linguine-mit-feinen-pilzen.html?position=1&clicked=")
|
180 |
+
print("------ here 22 " )
|
181 |
+
chat_history =[]
|
182 |
+
retriever_chain = get_context_retriever_chain(vs)
|
183 |
+
conversation_rag_chain = get_conversational_rag_chain(retriever_chain)
|
184 |
+
|
185 |
+
response = conversation_rag_chain.invoke({
|
186 |
+
"chat_history": chat_history,
|
187 |
+
"input": user_input
|
188 |
+
})
|
189 |
+
|
190 |
+
return response['answer']
|
191 |
+
|
192 |
+
def simple(text:str):
|
193 |
+
return text +" hhhmmm "
|
194 |
+
|
195 |
+
app = gr.Interface(
|
196 |
+
fn=get_response,
|
197 |
+
#fn=simple,
|
198 |
+
inputs=["text"],
|
199 |
+
outputs="text",
|
200 |
+
title="Chat with Websites",
|
201 |
+
description="Type your message and chat with websites.",
|
202 |
+
#allow_flagging=False
|
203 |
+
)
|
204 |
+
|
205 |
+
app.launch(debug=True, share=True)#wie registriere ich mich bei bofrost? Was kosten Linguine
|