Upload 2 files

app.py

@@ -0,0 +1,225 @@
+import re
+import json
+import base64
+import requests
+import torch
+import nest_asyncio
+from fastapi import HTTPException
+from pydantic import BaseModel
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from sentence_transformers import SentenceTransformer, models
+import gradio as gr
+
+# Apply nest_asyncio to allow async operations in the notebook/Spaces
+nest_asyncio.apply()
+
+############################################
+# Configuration
+############################################
+
+# Replace with your actual tokens
+HF_TOKEN = "YOUR_HF_TOKEN"
+GITHUB_TOKEN = "YOUR_GITHUB_TOKEN"
+
+############################################
+# GitHub API Functions
+############################################
+
+def extract_repo_info(github_url: str):
+    pattern = r"github\.com/([^/]+)/([^/]+)"
+    match = re.search(pattern, github_url)
+    if match:
+        owner = match.group(1)
+        repo = match.group(2).replace('.git', '')
+        return owner, repo
+    else:
+        raise ValueError("Invalid GitHub URL provided.")
+
+def get_repo_metadata(owner: str, repo: str):
+    headers = {'Authorization': f'token {GITHUB_TOKEN}'}
+    repo_url = f"https://api.github.com/repos/{owner}/{repo}"
+    response = requests.get(repo_url, headers=headers)
+    return response.json()
+
+def get_repo_tree(owner: str, repo: str, branch: str):
+    headers = {'Authorization': f'token {GITHUB_TOKEN}'}
+    tree_url = f"https://api.github.com/repos/{owner}/{repo}/git/trees/{branch}?recursive=1"
+    response = requests.get(tree_url, headers=headers)
+    return response.json()
+
+def get_file_content(owner: str, repo: str, file_path: str):
+    headers = {'Authorization': f'token {GITHUB_TOKEN}'}
+    content_url = f"https://api.github.com/repos/{owner}/{repo}/contents/{file_path}"
+    response = requests.get(content_url, headers=headers)
+    data = response.json()
+    if 'content' in data:
+        return base64.b64decode(data['content']).decode('utf-8')
+    else:
+        return None
+
+############################################
+# Embedding Functions
+############################################
+
+def preprocess_text(text: str) -> str:
+    cleaned_text = text.strip()
+    cleaned_text = re.sub(r'\s+', ' ', cleaned_text)
+    return cleaned_text
+
+def load_embedding_model(model_name: str = 'huggingface/CodeBERTa-small-v1') -> SentenceTransformer:
+    transformer_model = models.Transformer(model_name)
+    pooling_model = models.Pooling(transformer_model.get_word_embedding_dimension(),
+                                   pooling_mode_mean_tokens=True)
+    model = SentenceTransformer(modules=[transformer_model, pooling_model])
+    return model
+
+def generate_embedding(text: str, model_name: str = 'huggingface/CodeBERTa-small-v1') -> list:
+    processed_text = preprocess_text(text)
+    model = load_embedding_model(model_name)
+    embedding = model.encode(processed_text)
+    return embedding
+
+############################################
+# LLM Integration Functions
+############################################
+
+def is_detailed_query(query: str) -> bool:
+    keywords = ["detail", "detailed", "thorough", "in depth", "comprehensive", "extensive"]
+    return any(keyword in query.lower() for keyword in keywords)
+
+def generate_prompt(query: str, context_snippets: list) -> str:
+    context = "\n\n".join(context_snippets)
+    if is_detailed_query(query):
+        instruction = "Provide an extremely detailed and thorough explanation of at least 500 words."
+    else:
+        instruction = "Answer concisely."
+    
+    prompt = (
+        f"Below is some context from a GitHub repository:\n\n"
+        f"{context}\n\n"
+        f"Based on the above, {instruction}\n{query}\n"
+        f"Answer:"
+    )
+    return prompt
+
+def get_llm_response(prompt: str, model_name: str = "meta-llama/Llama-2-7b-chat-hf", max_new_tokens: int = None) -> str:
+    if max_new_tokens is None:
+        max_new_tokens = 1024 if is_detailed_query(prompt) else 256
+
+    torch.cuda.empty_cache()
+    
+    # Load tokenizer and model with authentication using the 'token' parameter.
+    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, token=HF_TOKEN)
+    model = AutoModelForCausalLM.from_pretrained(
+        model_name,
+        device_map="auto",
+        use_safetensors=False,
+        trust_remote_code=True,
+        torch_dtype=torch.float16,
+        token=HF_TOKEN
+    )
+    
+    text_gen = pipeline("text-generation", model=model, tokenizer=tokenizer)
+    outputs = text_gen(prompt, max_new_tokens=max_new_tokens, do_sample=True, temperature=0.7)
+    full_response = outputs[0]['generated_text']
+    
+    marker = "Answer:"
+    if marker in full_response:
+        answer = full_response.split(marker, 1)[1].strip()
+    else:
+        answer = full_response.strip()
+    
+    return answer
+
+############################################
+# Gradio Interface Functions
+############################################
+
+def load_repo_contents(github_url: str):
+    try:
+        owner, repo = extract_repo_info(github_url)
+    except Exception as e:
+        return f"Error: {str(e)}"
+    repo_data = get_repo_metadata(owner, repo)
+    default_branch = repo_data.get("default_branch", "main")
+    tree_data = get_repo_tree(owner, repo, default_branch)
+    if "tree" not in tree_data:
+        return "Error: Could not fetch repository tree."
+    file_list = [item["path"] for item in tree_data["tree"] if item["type"] == "blob"]
+    return file_list
+
+def get_file_content_for_choice(github_url: str, file_choice: int):
+    try:
+        owner, repo = extract_repo_info(github_url)
+    except Exception as e:
+        return str(e)
+    repo_data = get_repo_metadata(owner, repo)
+    default_branch = repo_data.get("default_branch", "main")
+    tree_data = get_repo_tree(owner, repo, default_branch)
+    if "tree" not in tree_data:
+        return "Error: Could not fetch repository tree."
+    file_list = [item["path"] for item in tree_data["tree"] if item["type"] == "blob"]
+    if file_choice < 1 or file_choice > len(file_list):
+        return "Error: Invalid file choice."
+    selected_file = file_list[file_choice - 1]
+    content = get_file_content(owner, repo, selected_file)
+    return content, selected_file
+
+def chat_with_file(github_url: str, file_choice: int, user_query: str):
+    result = get_file_content_for_choice(github_url, file_choice)
+    if isinstance(result, str):
+        return result  # Error message
+    file_content, selected_file = result
+    preprocessed = preprocess_text(file_content)
+    context_snippet = preprocessed[:1000]  # use first 1000 characters as context
+    prompt = generate_prompt(user_query, [context_snippet])
+    llm_response = get_llm_response(prompt)
+    return f"File: {selected_file}\n\nLLM Response:\n{llm_response}"
+
+############################################
+# Gradio Interface Setup
+############################################
+
+with gr.Blocks() as demo:
+    gr.Markdown("# RepoChat - Chat with Repository Files")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            gr.Markdown("### Repository Information")
+            github_url_input = gr.Textbox(label="GitHub Repository URL", placeholder="https://github.com/username/repository")
+            load_repo_btn = gr.Button("Load Repository Contents")
+            file_dropdown = gr.Dropdown(label="Select a File", interactive=True)
+            repo_content_output = gr.Textbox(label="File Content", interactive=False, lines=10)
+        with gr.Column(scale=2):
+            gr.Markdown("### Chat Interface")
+            chat_query_input = gr.Textbox(label="Your Query", placeholder="Type your query here")
+            chat_output = gr.Textbox(label="Chatbot Response", interactive=False, lines=10)
+            chat_btn = gr.Button("Send Query")
+    
+    # When clicking "Load Repository Contents", update file dropdown
+    def update_file_dropdown(github_url):
+        files = load_repo_contents(github_url)
+        return files
+    
+    load_repo_btn.click(fn=update_file_dropdown, inputs=[github_url_input], outputs=[file_dropdown])
+    
+    # When file selection changes, update file content display
+    def update_repo_content(github_url, file_choice):
+        if not file_choice:
+            return "No file selected."
+        try:
+            file_index = int(file_choice)
+        except:
+            file_index = 1
+        content, _ = get_file_content_for_choice(github_url, file_index)
+        return content
+    
+    file_dropdown.change(fn=update_repo_content, inputs=[github_url_input, file_dropdown], outputs=[repo_content_output])
+    
+    # When sending a chat query, process it
+    def process_chat(github_url, file_choice, chat_query):
+        return chat_with_file(github_url, int(file_choice), chat_query)
+    
+    chat_btn.click(fn=process_chat, inputs=[github_url_input, file_dropdown, chat_query_input], outputs=[chat_output])
+    
+demo.launch()

requirements.txt

@@ -0,0 +1,8 @@
+fastapi
+uvicorn
+nest_asyncio
+requests
+torch
+transformers
+sentence-transformers
+gradio