app.py

import gradio as gr
import torch
from mingru_lm import MinGRU_LM


model = MinGRU_LM(dim=512, num_tokens=256, num_layers=6)
pt_model = "model/best_model.pt"
checkpoint = torch.load(pt_model)
model.load_state_dict(checkpoint['model_state_dict'])

# Move model to GPU if available
device = 'cuda' if torch.cuda.is_available() else 'cpu'
model = model.to(device)

def decode_tokens(tokens):
    return ''.join([chr(token) for token in tokens if token >= 32 and token < 256])  # ASCII-safe decoding

def tokenize_text(text):
    return [ord(char) for char in text if ord(char) < 256]  # ASCII-safe tokenization

def generate_text(start_text, max_length, temperature):
    model.eval()
    
    tokens = tokenize_text(start_text)
    input_tensor = torch.tensor(tokens, dtype=torch.long).unsqueeze(0).to(device)  # Ensure long tensor
    
    generated_tokens = tokens.copy()
    
    with torch.no_grad():
        for _ in range(max_length):
            _, logits = model(input_tensor, labels=None)
            
            last_token_logits = logits[0, -1, :] / temperature
            probs = torch.softmax(last_token_logits, dim=-1)
            next_token = torch.multinomial(probs, num_samples=1).item()
            
            # Only append if it's within the 256-character ASCII range
            if next_token < 256:  
                generated_tokens.append(next_token)
                input_tensor = torch.cat([input_tensor, torch.tensor([[next_token]], device=device)], dim=1)
            else:
                continue  # Skip tokens outside ASCII range
        
    return decode_tokens(generated_tokens)

# Gradio interface
iface = gr.Interface(
    fn=generate_text,
    inputs=[
        gr.Textbox(lines=3, label="Enter your prompt", value="Once upon a time"),
        gr.Slider(minimum=10, maximum=500, value=200, step=1, label="Max Length"),
        gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.1, label="Temperature"),
    ],
    outputs=gr.Textbox(lines=10, label="Generated Text"),
    title="Text Generation with MinGRU_LM",
    description="Enter a prompt and adjust parameters to generate text using the MinGRU_LM model."
)

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