app.py

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

model_id = "deepseek-ai/deepseek-coder-1.3b-base"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",          # Auto-detect GPU if available
    torch_dtype=torch.float16   # Use FP16 for faster, lower-memory inference
)

pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

def generate_code(prompt):
    response = pipe(prompt, max_new_tokens=200, temperature=0.7, do_sample=True)
    return response[0]["generated_text"]

gr.Interface(
    fn=generate_code,
    inputs=gr.Textbox(lines=4, placeholder="Ask DeepSeek R1 something..."),
    outputs="text",
    title="🧠 DeepSeek Coder R1 (1.3B)",
    description="Running open-source DeepSeek Coder model (1.3B) on Hugging Face Spaces."
).launch()