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

# Load tokenizer and model manually
tokenizer = AutoTokenizer.from_pretrained("microsoft/bitnet-b1.58-2B-4T")
model = AutoModelForCausalLM.from_pretrained(
    "microsoft/bitnet-b1.58-2B-4T",
    torch_dtype=torch.float16,
    device_map="auto"  # uses GPU if available
)

def generate_text(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=100,
            do_sample=True,
            temperature=0.7,
            top_p=0.9,
        )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Create Gradio UI
gr.Interface(fn=generate_text, inputs="text", outputs="text", title="Manual DeepSeek").launch()