import gradio as gr import torch import sentencepiece as spm from config import AmaiaConfig from model import Amaia torch.serialization.add_safe_globals([AmaiaConfig]) ckpt = torch.load("amaia.pt", map_location="cpu", weights_only=True) state_dict = ckpt["model"] unwrapped = {k.replace("_orig_mod.", ""): v for k, v in state_dict.items()} model = Amaia(ckpt["cfg"]) model.load_state_dict(unwrapped) model.eval() sp = spm.SentencePieceProcessor() sp.Load("amaia.model") def generate(prompt, temperature, max_tokens): ids = sp.Encode(prompt) idx = torch.tensor([ids]) with torch.no_grad(): out = model.generate(idx, int(max_tokens), temperature=float(temperature)) return sp.Decode(out[0].tolist()) gr.Interface( fn=generate, inputs=[ gr.Textbox(label="Prompt", placeholder="Enter your prompt here..."), gr.Slider(0.1, 2.0, value=0.8, step=0.1, label="Temperature"), gr.Slider(50, 500, value=200, step=50, label="Max Tokens"), ], outputs=gr.Textbox(label="Generated Text"), title="Amaia", description="An open-source language model, built from scratch.", ).launch()