changes

app.py

@@ -1,64 +1,162 @@
+import os
+import re
+import json
+import torch
+import spaces
+import pymupdf
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+from qdrant_client import QdrantClient
+from utils import download_pdf_from_gdrive, merge_strings_with_prefix
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
+def rag_query(query: str):
+    """
+    Allows searching the vector database which contains
+    information for a man named Suvaditya for a given query
+    by performing semantic search. Returns results by
+    looking at his resume, which contains a plethora of
+    information about him.
+
+    Args:
+        query: The query against which the search will be run,
+               in the form a single string phrase no more than
+               10 words.
+
+    Returns:
+        search_results: A list of results that come closest
+                        to the given query semantically,
+                        determined by Cosine Similarity.
+    """
+    return client.query(
+        collection_name="resume",
+        query_text=query
+    )
+
+def generate_answer(chat_history):
+    # Generate result
+    tool_prompt = tokenizer.apply_chat_template(
+        chat_history,
+        tools=[rag_query],
+        return_tensors="pt",
+        return_dict=True,
+        add_generation_prompt=True,
+    )
+    tool_prompt = tool_prompt.to(model.device)
+    out = model.generate(**tool_prompt, max_new_tokens=512)
+    generated_text = out[0, tool_prompt['input_ids'].shape[1]:]
+    generated_text = tokenizer.decode(generated_text)
+    return generated_text
+
+def parse_tool_request(tool_call, top_k=5):
+    pattern = r"<tool_call>(.*?)</tool_call>"
+    match_result = re.search(pattern, tool_call, re.DOTALL)
+    if match_result:
+        result = match_result.group(1).strip()
+    else:
+        return None, None
+
+    query = json.loads(result)["arguments"]["query"]
+    query_results = [
+        query_piece.metadata["document"] for query_piece in rag_query(query)
+    ]
+
+    return query_results[:top_k], query
+
+def update_chat_history(chat_history, tool_query, query_results):
+    assistant_tool_message = {
+        "role": "assistant",
+        "tool_calls": [{
+            "type": "function",
+            "function": {
+                "name": "rag_query",
+                "arguments": {"query": f"{tool_query}"}
+            }
+        }]
+    }
+    result_tool_message = {
+        "role": "tool",
+        "name": "rag_query",
+        "content": "\n".join(query_results)
+    }
+
+    chat_history.append(assistant_tool_message)
+    chat_history.append(result_tool_message)
+
+    return chat_history
 
 if __name__ == "__main__":
-    demo.launch()
+    RESUME_PATH = os.path.join(os.getcwd(), "Resume.pdf")
+    RESUME_URL = "https://drive.google.com/file/d/1YMF9NNTG5gubwJ7ipI5JfxAJKhlD9h2v/"
+
+    # Download file
+    download_pdf_from_gdrive(RESUME_URL, RESUME_PATH)
+
+    doc = pymupdf.open(RESUME_PATH)
+    fulltext = doc[0].get_text().split("\n")
+    fulltext = merge_strings_with_prefix(fulltext)
+
+    # Embed the sentences
+    client = QdrantClient(":memory:")
+
+    client.set_model("sentence-transformers/all-MiniLM-L6-v2")
+
+    if not client.collection_exists(collection_name="resume"):
+        client.create_collection(
+            collection_name="resume",
+            vectors_config=client.get_fastembed_vector_params(),
+        )
+
+    _ = client.add(
+        collection_name="resume",
+        documents=fulltext,
+        ids=range(len(fulltext)),
+        batch_size=100,
+        parallel=0,
+    )
+
+    # FOR QWEN, THIS IS WORKING
+
+    model_name = "Qwen/Qwen2.5-3B-Instruct"
+
+    @spaces.GPU
+    def rag_process(message, chat_history):
+        # Append current user message to chat history
+        current_message = {
+            "role": "user",
+            "content": message
+        }
+        chat_history.append(current_message)
+
+        # Generate LLM answer
+        generated_text = generate_answer(chat_history)
+
+        # Detect if tool call is requested by LLM. If yes, then
+        # execute tool and use else return None
+        query_results, tool_query = parse_tool_request(generated_text)
+
+        # If tool call was requested
+        if query_results is not None and tool_query is not None:
+            print("Inside")
+            # Update chat history with result of tool call
+            chat_history = update_chat_history(
+                chat_history, tool_query, query_results
+            )
+            # Generate result from the
+            generated_text = generate_answer(chat_history)
+
+        return generated_text[:-10]
+
+    model = AutoModelForCausalLM.from_pretrained(
+        model_name,
+        torch_dtype=torch.bfloat16,
+        device_map="auto",
+        load_in_4bit=True,
+    ).to_bettertransformer().to('cuda')
+
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+    demo = gr.ChatInterface(
+        fn=rag_process,
+        type="messages",
+    )
+    demo.launch()

requirements.txt

@@ -1 +1,11 @@
-huggingface_hub==0.25.2
+huggingface_hub==0.25.2
+qdrant_client
+pymupdf
+gdown
+fastembed
+transformers
+torch
+torchvision
+torchaudio
+accelerate
+bitsandbytes