app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "lakshraina2/leetcodeAI" 

print("Loading model on CPU...")

tokenizer = AutoTokenizer.from_pretrained(model_id, token=False)
# We removed device_map to avoid the Accelerate dependency error
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    dtype=torch.float32, 
    token=False
)

def solve(problem_text):
    prompt = f"### Instruction:\nSolve this LeetCode problem:\n{problem_text}\n\n### Response:\n"
    inputs = tokenizer(prompt, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs, 
            max_new_tokens=512,
            temperature=0.2,
            do_sample=True
        )
    
    solution = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return solution.split("### Response:\n")[-1].strip()

iface = gr.Interface(fn=solve, inputs="text", outputs="text")
iface.launch()