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

MODEL_REPO = "wuhp/myr1"
SUBFOLDER = "myr1"

tokenizer = AutoTokenizer.from_pretrained(
    MODEL_REPO,
    subfolder=SUBFOLDER,
    trust_remote_code=True
)

# If your GPU has <24GB VRAM, consider 8-bit or CPU offloading
model = AutoModelForCausalLM.from_pretrained(
    MODEL_REPO,
    subfolder=SUBFOLDER,
    trust_remote_code=True,
    device_map="auto",           # tries to place layers on GPU, then CPU if needed
    torch_dtype=torch.float16,    # or bfloat16 or float32
    low_cpu_mem_usage=True
)

model.eval()

def generate_text(prompt, max_length=64, temperature=0.7, top_p=0.9):
    print("=== Starting generation ===")
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    try:
        output_ids = model.generate(
            **inputs,
            max_new_tokens=max_length,   # alternative to max_length
            temperature=temperature,
            top_p=top_p,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        print("=== Generation complete ===")
    except Exception as e:
        print(f"Error during generation: {e}")
        return str(e)
    return tokenizer.decode(output_ids[0], skip_special_tokens=True)

demo = gr.Interface(
    fn=generate_text,
    inputs=[
        gr.Textbox(
            lines=4,
            label="Prompt",
            placeholder="Try a short prompt, e.g., Hello!"
        ),
        gr.Slider(8, 512, value=64, step=1, label="Max New Tokens"),
        gr.Slider(0.0, 1.5, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(0.0, 1.0, value=0.9, step=0.05, label="Top-p"),
    ],
    outputs="text",
    title="DeepSeek R1 Demo",
    description="Generates text using the large DeepSeek model."
)

if __name__ == "__main__":
    demo.launch()