Updated Space with Gradio interface and configuration

app.py

@@ -2,48 +2,53 @@ import gradio as gr
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 import torch
+import os
 
 # Constants
 MODEL_NAME = "Salesforce/codegen-350M-mono"
 LORA_PATH = "fine-tuned-model"
 
 # Initialize tokenizer and model
+print("Loading tokenizer...")
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
+
+print("Loading base model...")
 base_model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
-    torch_dtype=torch.float16,
+    trust_remote_code=True,
     device_map="auto",
-    trust_remote_code=True
+    torch_dtype=torch.float32  # Changed from float16 to float32 for better compatibility
 )
 
-# Load the fine-tuned model
+print("Loading fine-tuned model...")
 model = PeftModel.from_pretrained(
     base_model,
     LORA_PATH,
-    torch_dtype=torch.float16,
-    device_map="auto"
+    device_map="auto",
+    torch_dtype=torch.float32
 )
 
 def generate_response(task_description, code_snippet, request_type, mode="concise"):
-    # Format the prompt based on request type
-    if request_type == "hint":
-        prompt = f"""Task Description: {task_description}
+    try:
+        # Format the prompt based on request type
+        if request_type == "hint":
+            prompt = f"""Task Description: {task_description}
 
 User's Code:
 {code_snippet}
 
 AI-HR Assistant: Here's a hint to help you:
 HINT:"""
-    elif request_type == "feedback":
-        prompt = f"""Task Description: {task_description}
+        elif request_type == "feedback":
+            prompt = f"""Task Description: {task_description}
 
 User's Code:
 {code_snippet}
 
 AI-HR Assistant: Here's my feedback on your code:
 FEEDBACK:"""
-    else:  # follow-up
-        prompt = f"""Task Description: {task_description}
+        else:  # follow-up
+            prompt = f"""Task Description: {task_description}
 
 User's Code:
 {code_snippet}
@@ -51,28 +56,30 @@ User's Code:
 AI-HR Assistant: Here's a follow-up question to extend your learning:
 FOLLOW-UP:"""
 
-    # Generate response
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=256 if mode == "detailed" else 128,
-            do_sample=True,
-            temperature=0.7,
-            top_p=0.95,
-        )
-    
-    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    
-    # Extract the relevant part of the response
-    if request_type == "hint" and "HINT:" in response:
-        response = response.split("HINT:", 1)[1].strip()
-    elif request_type == "feedback" and "FEEDBACK:" in response:
-        response = response.split("FEEDBACK:", 1)[1].strip()
-    elif request_type == "follow-up" and "FOLLOW-UP:" in response:
-        response = response.split("FOLLOW-UP:", 1)[1].strip()
-    
-    return response
+        # Generate response
+        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=256 if mode == "detailed" else 128,
+                do_sample=True,
+                temperature=0.7,
+                top_p=0.95,
+            )
+        
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        
+        # Extract the relevant part of the response
+        if request_type == "hint" and "HINT:" in response:
+            response = response.split("HINT:", 1)[1].strip()
+        elif request_type == "feedback" and "FEEDBACK:" in response:
+            response = response.split("FEEDBACK:", 1)[1].strip()
+        elif request_type == "follow-up" and "FOLLOW-UP:" in response:
+            response = response.split("FOLLOW-UP:", 1)[1].strip()
+        
+        return response
+    except Exception as e:
+        return f"An error occurred: {str(e)}"
 
 # Create Gradio interface
 with gr.Blocks(title="Live Coding HR Assistant") as demo:
@@ -83,13 +90,13 @@ with gr.Blocks(title="Live Coding HR Assistant") as demo:
         with gr.Column():
             task_description = gr.Textbox(
                 label="Task Description",
-                placeholder="Enter the coding problem or task description...",
+                value="",
                 lines=5
             )
             code_snippet = gr.Code(
                 label="Code Snippet",
                 language="python",
-                placeholder="Enter your code here..."
+                value=""
             )
             request_type = gr.Radio(
                 choices=["hint", "feedback", "follow-up"],
@@ -107,7 +114,7 @@ with gr.Blocks(title="Live Coding HR Assistant") as demo:
             output = gr.Textbox(
                 label="AI Response",
                 lines=8,
-                placeholder="AI response will appear here..."
+                value=""
             )
     
     submit_btn.click(
@@ -115,18 +122,7 @@ with gr.Blocks(title="Live Coding HR Assistant") as demo:
         inputs=[task_description, code_snippet, request_type, mode],
         outputs=output
     )
-    
-    gr.Markdown("""
-    ### How to use
-    1. Enter your coding task description
-    2. Paste your code (if any)
-    3. Choose what you need:
-       - **Hint**: Get a helpful hint without the full solution
-       - **Feedback**: Get constructive feedback on your code
-       - **Follow-up**: Get a follow-up question to extend your learning
-    4. Choose response style (concise or detailed)
-    5. Click "Get Response"
-    """)
 
 # Launch the app
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -6,5 +6,5 @@ peft
 accelerate
 sentencepiece
 python-multipart
-gradio>=4.12.0
+gradio>=4.44.1
 bitsandbytes