Update app.py

app.py

@@ -1,64 +1,140 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from huggingface_hub import login
+from datasets import load_dataset
+import chromadb
+import torch
+from sentence_transformers import SentenceTransformer
+import os
 
 """
 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")
 
+# ChromaDB Setup (Persistent Client)
+CHROMA_DB_PATH = "new_hadith_rag_source" # Directory to store ChromaDB data
+client = chromadb.PersistentClient(path=CHROMA_DB_PATH)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+COLLECTION_NAME = "hadiths_new_complete"
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# Function to load or create the ChromaDB collection
+def load_or_create_collection():
+    try:
+        collection = client.get_collection(name=COLLECTION_NAME)
+        print("Collection loaded successfully.")
+        return collection
+    except ValueError:  # Collection doesn't exist
+        print("Creating new collection...")
+        collection = client.create_collection(name=COLLECTION_NAME, overwrite=True)
 
-    messages.append({"role": "user", "content": message})
+        # Load data and add to the collection
+        ds = load_dataset("rwmasood/hadith-qa-pair")
+        device = 'cuda' if torch.cuda.is_available() else 'cpu'
+        embedding_model = SentenceTransformer('all-MiniLM-L6-v2').to(device)  # Using a local name to avoid shadowing
 
-    response = ""
+        for split in ds.keys():
+            documents = [
+                f"Hadith: {row['hadith-eng']}\nQuestion: {row['question']}\nReference: {row['reference']}"
+                for row in ds[split]
+            ]
+            ids = [f"{split}_{i}" for i in range(len(documents))]
 
-    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
+            # Compute embeddings using CUDA
+            embeddings = embedding_model.encode(documents, convert_to_tensor=True, device=device)
 
-        response += token
-        yield response
+            collection.add(
+                documents=documents,
+                ids=ids,
+                embeddings=embeddings.cpu().numpy()
+            )
 
+        print(f"Collection created with {collection.count()} documents.")
+        return collection
 
-"""
-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)",
-        ),
-    ],
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+# Load or create the collection
+collection = load_or_create_collection()
+
+# Debugging print
+print(f"Number of documents in collection: {collection.count()}")
+
+# Model and Tokenizer Loading
+model_name = "meta-llama/Llama-3.2-3B-Instruct"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+llm = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.float16, pad_token_id=tokenizer.eos_token_id)
+
+# Embedding Model
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+model = SentenceTransformer('all-MiniLM-L6-v2').to(device)
+
+
+# Helper Functions (Querying, Generation, Grading) - No changes needed
+def query_collection(query_text, top_k=3):
+    query_embedding = model.encode(query_text, convert_to_tensor=True, device=device).cpu().numpy()
+    results = collection.query(
+        query_embeddings=[query_embedding],
+        n_results=top_k
+    )
+    return results
+
+def speculative_generation(context, question, num_candidates=3):
+    responses = []
+    for _ in range(num_candidates):
+        prompt = f"Context:\n{context}\n\nQuestion:\n{question}\n\nAnswer:"
+        inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+        try:
+            outputs = llm.generate(**inputs, max_length=2048, num_return_sequences=1, num_beams=5, temperature=0.9, pad_token_id=tokenizer.eos_token_id)
+            response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+            responses.append(response)
+        except Exception as e:
+            print(f"Error during generation: {e}")
+            responses.append("An error occurred during generation.")
+    return responses
+
+def grade_responses(responses, query):
+    best_score = -1
+    best_response = ""
+    for response in responses:
+        score = 0
+        score += sum(1 for word in query.lower().split() if word in response.lower())
+        query_embedding = model.encode(query, convert_to_tensor=True, device=device)
+        response_embedding = model.encode(response, convert_to_tensor=True, device=device)
+        similarity = torch.nn.functional.cosine_similarity(query_embedding, response_embedding, dim=0).item()
+        score += similarity * 10
+        if score > best_score:
+            best_score = score
+            best_response = response
+    return best_response
+
+def chatbot_response(user_query, top_k=3, num_candidates=3):
+    results = query_collection(user_query, top_k)
+    context = "\n\n".join(results['documents'][0])
+    speculative_responses = speculative_generation(context, user_query, num_candidates)
+    best_response = grade_responses(speculative_responses, user_query)
+    return best_response
+
+# Chatbot Function (Adjusted for Error Handling and Default Message)
+def chatbot(query):
+    if not query.strip():
+        return "Please ask a question about hadiths."
+
+    try:
+        answer = chatbot_response(query)
+        if "don't know" in answer.lower() or "not sure" in answer.lower():
+            return "Sorry. I don't have information about the hadiths related. It might be a dhoif, or maudhu, or I just don't have the knowledge."
+        else:
+            return answer
+    except Exception as e:
+        print(f"Error in chatbot: {e}")
+        return f"An error occurred: {e}"
+
+# Gradio Interface
+if __name__ == "__main__":  # Ensures this only runs when the script is executed directly
+    iface = gr.Interface(
+        fn=chatbot,
+        inputs=gr.Textbox(lines=7, placeholder="Ask me a question about hadiths...", label="Question"),
+        outputs=gr.Textbox(label="Answer"),
+        title="Hadith QA Chatbot",
+        description="Ask questions related to Hadiths."
+    )
+    iface.launch()