Spaces:
Sleeping
Sleeping
Update util.py
Browse files
util.py
CHANGED
@@ -1,48 +1,17 @@
|
|
1 |
import os
|
2 |
import streamlit as st
|
3 |
-
from langchain_community.embeddings import
|
4 |
-
from sentence_transformers import SentenceTransformer
|
5 |
from langchain_community.vectorstores import Chroma
|
6 |
-
|
7 |
-
|
8 |
-
from langchain.chains.question_answering import load_qa_chain
|
9 |
-
from langchain.prompts import PromptTemplate
|
10 |
|
11 |
-
|
12 |
-
import google.generativeai as genai
|
13 |
|
14 |
-
|
15 |
|
16 |
-
|
17 |
|
18 |
-
|
19 |
-
# load_in_4bit=True,
|
20 |
-
# bnb_4bit_compute_dtype=torch.bfloat16
|
21 |
-
# )
|
22 |
-
|
23 |
-
|
24 |
-
model_kwargs = {'device': 'cpu'}
|
25 |
-
embeddings = HuggingFaceEmbeddings(model_name="michaelfeil/ct2fast-e5-small",model_kwargs=model_kwargs, )
|
26 |
-
# embeddings = SentenceTransformer(model_name_or_path="All-MiniLM-L6-v2")
|
27 |
-
|
28 |
-
# tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
|
29 |
-
# model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2", device_map='auto', quantization_config = quantization_config)
|
30 |
-
|
31 |
-
# pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens = 1000)
|
32 |
-
# llm = HuggingFacePipeline(pipeline=pipe)
|
33 |
-
|
34 |
-
|
35 |
-
# def clone_repo(repo):
|
36 |
-
# if os.path.exists("githubCode") and os.path.isdir("githubCode"):
|
37 |
-
# print("File already exists!!")
|
38 |
-
# pass
|
39 |
-
# else:
|
40 |
-
# print("Cloning repo!!")
|
41 |
-
# git.Repo.clone_from(repo,"githubCode")
|
42 |
-
|
43 |
-
# git.Repo.clone_from("https://github.com/Divyansh3021/Github_code_assistant.git", "githubCode")
|
44 |
-
|
45 |
-
llm = genai.GenerativeModel('gemini-pro')
|
46 |
|
47 |
def get_folder_paths(directory = "githubCode"):
|
48 |
folder_paths = []
|
@@ -71,40 +40,26 @@ with open("Code.txt", "w", encoding='utf-8') as output:
|
|
71 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
72 |
from langchain_community.document_loaders import TextLoader
|
73 |
|
74 |
-
# for filename in os.listdir(directory_path):
|
75 |
-
# if filename.endswith(".txt"): # Only process PD files
|
76 |
-
# file_path = os.path.join(directory_path, filename)
|
77 |
loader = TextLoader("Code.txt", encoding="utf-8")
|
78 |
pages = loader.load_and_split()
|
79 |
|
80 |
# Split data into chunks
|
81 |
-
text_splitter = RecursiveCharacterTextSplitter(
|
82 |
-
|
83 |
-
|
84 |
-
add_start_index = True,
|
85 |
-
)
|
86 |
-
chunks = text_splitter.split_documents(pages)
|
87 |
|
88 |
-
|
89 |
-
db=Chroma.from_documents(chunks,embedding=embeddings,persist_directory="test_index")
|
90 |
-
db.persist()
|
91 |
|
92 |
-
|
93 |
-
|
|
|
|
|
94 |
|
95 |
-
|
96 |
-
retriever = vectordb.as_retriever(search_kwargs = {"k": 3})
|
97 |
|
98 |
# Function to generate assistant's response using ask function
|
99 |
def generate_assistant_response(question):
|
100 |
-
|
101 |
-
|
102 |
-
Context: ```
|
103 |
-
{context}
|
104 |
-
```
|
105 |
-
### Question: {question} [/INST]"""
|
106 |
-
print("Context: ", context)
|
107 |
-
answer = llm.generate_content(qna_prompt_template).text
|
108 |
-
return answer
|
109 |
|
110 |
# print(generate_assistant_response("Tell me about the instructor_embeddings function."))
|
|
|
1 |
import os
|
2 |
import streamlit as st
|
3 |
+
from langchain_community.embeddings import HuggingFaceHubEmbeddings
|
|
|
4 |
from langchain_community.vectorstores import Chroma
|
5 |
+
from langchain.chains import RetrievalQA
|
6 |
+
from langchain_google_genai import ChatGoogleGenerativeAI, GoogleGenerativeAIEmbeddings
|
|
|
|
|
7 |
|
8 |
+
import git
|
|
|
9 |
|
10 |
+
from chromadb.utils import embedding_functions
|
11 |
|
12 |
+
embeddings = GoogleGenerativeAIEmbeddings(model="models/embedding-001", google_api_key=os.environ['GOOGLE_API_KEY'], task_type="retrieval_query")
|
13 |
|
14 |
+
model = ChatGoogleGenerativeAI(model="gemini-pro",google_api_key=os.environ['GOOGLE_API_KEY'],temperature=0.2,convert_system_message_to_human=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
def get_folder_paths(directory = "githubCode"):
|
17 |
folder_paths = []
|
|
|
40 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
41 |
from langchain_community.document_loaders import TextLoader
|
42 |
|
|
|
|
|
|
|
43 |
loader = TextLoader("Code.txt", encoding="utf-8")
|
44 |
pages = loader.load_and_split()
|
45 |
|
46 |
# Split data into chunks
|
47 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=2000, chunk_overlap=200)
|
48 |
+
context = "\n\n".join(str(p.page_content) for p in pages)
|
49 |
+
texts = text_splitter.split_text(context)
|
|
|
|
|
|
|
50 |
|
51 |
+
vector_index = Chroma.from_texts(texts, embeddings).as_retriever(search_kwargs={"k":5})
|
|
|
|
|
52 |
|
53 |
+
qa_chain = RetrievalQA.from_chain_type(
|
54 |
+
model,
|
55 |
+
retriever=vector_index,
|
56 |
+
return_source_documents=True
|
57 |
|
58 |
+
)
|
|
|
59 |
|
60 |
# Function to generate assistant's response using ask function
|
61 |
def generate_assistant_response(question):
|
62 |
+
answer = qa_chain({"query": question})
|
63 |
+
return answer['result']
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
# print(generate_assistant_response("Tell me about the instructor_embeddings function."))
|