new

app.py

@@ -1,64 +1,77 @@
+# import gradio as gr
+
+# gr.load("models/HuggingFaceH4/zephyr-7b-alpha").launch()
+
+import os
+import numpy as np
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import faiss
 
-"""
-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")
+# Step 1: Load Precomputed Embeddings and Metadata
+def load_embeddings(embeddings_folder='embeddings'):
+    all_embeddings = []
+    metadata = []
+    for file in os.listdir(embeddings_folder):
+        if file.endswith('.npy'):
+            embedding_path = os.path.join(embeddings_folder, file)
+            embedding = np.load(embedding_path)  # Shape: (27, 384)
+            all_embeddings.append(embedding)
+            # Metadata corresponds to each .npy file
+            meta_info = file.replace('.npy', '')  # Example: 'course_1'
+            metadata.extend([meta_info] * embedding.shape[0])  # Repeat metadata for each sub-embedding
 
+    # Flatten list of embeddings to shape (n * 27, 384)
+    all_embeddings = np.vstack(all_embeddings)
+    return all_embeddings, metadata
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+embeddings, metadata = load_embeddings()
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# Step 2: Set Up FAISS Index with Flattened Embeddings
+dimension = embeddings.shape[1]  # Should be 384 after flattening
+index = faiss.IndexFlatL2(dimension)
+index.add(embeddings)
 
-    messages.append({"role": "user", "content": message})
 
-    response = ""
+# Step 3: Load the Language Model
+# model_name = "HuggingFaceH4/zephyr-7b-alpha"
+# tokenizer = AutoTokenizer.from_pretrained(model_name)
+# model = AutoModelForCausalLM.from_pretrained(model_name)
 
-    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
+model_name = "TheBloke/zephyr-7B-beta-GPTQ"
+tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
+model = AutoModelForCausalLM.from_pretrained(model_name, device_map="balanced", trust_remote_code=False)
 
-        response += token
-        yield response
 
+# Step 4: Define the Retrieval Function
+def retrieve_documents(query, top_k=3):
+    query_embedding = np.mean([embeddings[i] for i in range(len(metadata)) if query.lower() in metadata[i].lower()], axis=0)
+    distances, indices = index.search(np.array([query_embedding]), top_k)
+    retrieved_docs = [metadata[idx] for idx in indices[0]]
+    return retrieved_docs
 
-"""
-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)",
-        ),
-    ],
-)
+# Step 5: Define the Response Generation Function
+def generate_response(query):
+    retrieved_docs = retrieve_documents(query)
+    context = " ".join(retrieved_docs)
+    input_text = f"Context: {context}\n\nQuestion: {query}\n\nAnswer:"
+    inputs = tokenizer(input_text, return_tensors="pt")
+    output = model.generate(**inputs, max_length=512)
+    answer = tokenizer.decode(output[0], skip_special_tokens=True)
+    return answer
 
+# Step 6: Create Gradio Interface
+def gradio_interface(query):
+    response = generate_response(query)
+    return response
+
+iface = gr.Interface(
+    fn=gradio_interface,
+    inputs=gr.Textbox(lines=2, placeholder="Enter your query here..."),
+    outputs="text",
+    title="RAG-based Course Search",
+    description="Enter a query to search for relevant courses using Retrieval Augmented Generation."
+)
 
-if __name__ == "__main__":
-    demo.launch()
+if _name_ == "_main_":
+    iface.launch()