new

app.py

@@ -1,77 +1,84 @@
-# import gradio as gr
-
-# gr.load("models/HuggingFaceH4/zephyr-7b-alpha").launch()
-
-import os
-import numpy as np
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer
-import faiss
-
-# 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
-
-embeddings, metadata = load_embeddings()
-
-# 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)
-
+from huggingface_hub import InferenceClient
+import pandas as pd
+import torch
+from sentence_transformers import SentenceTransformer
 
-# Step 3: Load the Language Model
-# model_name = "HuggingFaceH4/zephyr-7b-alpha"
-# tokenizer = AutoTokenizer.from_pretrained(model_name)
-# model = AutoModelForCausalLM.from_pretrained(model_name)
+# Initialize the Hugging Face API client
+client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
-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)
+# Load and preprocess course data for search functionality
+courses_df = pd.read_csv("courses_data.csv")  # Assuming courses_data.csv is already scraped
+model = SentenceTransformer('distilbert-base-nli-stsb-mean-tokens')
+courses_df['embedding'] = courses_df['description'].apply(lambda x: model.encode(x, convert_to_tensor=True))
 
+# Define a helper function to search for courses based on user query
+def search_courses(query, top_k=5):
+    query_embedding = model.encode(query, convert_to_tensor=True)
+    cosine_scores = torch.nn.functional.cosine_similarity(query_embedding, torch.stack(courses_df['embedding'].tolist()))
+    top_results = torch.topk(cosine_scores, k=top_k)
+    
+    results = []
+    for idx in top_results.indices:
+        course = courses_df.iloc[idx.item()]
+        results.append(f"**{course['title']}**\n{course['description']}\nNumber of Lessons: {course['lessons']}")
+    return "\n\n".join(results)
 
-# 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
+# Modify respond function to include course search when "search: [query]" is detected
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    # Check if the message is a course search command
+    if message.lower().startswith("search:"):
+        query = message[len("search:"):].strip()  # Extract search query
+        response = search_courses(query)  # Perform course search
+        yield response
+        return
 
-# 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
+    # Standard chat processing using Hugging Face Inference API for general chat
+    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})
 
-# Step 6: Create Gradio Interface
-def gradio_interface(query):
-    response = generate_response(query)
-    return response
+    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
 
-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."
+# Gradio chat interface setup
+demo = gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are a helpful assistant that provides detailed information about courses on Analytics Vidhya. Please assist the user by answering questions about available courses, course content, instructors, ratings, pricing, and any other relevant details. If a user asks about course recommendations, suggest relevant courses based on their needs or preferences.", 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)",
+        ),
+    ],
+    description="Type 'search: [your query]' to search for courses on Analytics Vidhya, or chat with the assistant."
 )
 
-if _name_ == "_main_":
-    iface.launch()
+# Launch the app
+if __name__ == "__main__":
+    demo.launch()