generate.py

import torch
import torch.nn.functional as F
from evo_model import EvoDecoder
from transformers import GPT2Tokenizer

# Load tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")

# Load trained model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = EvoDecoder(vocab_size=tokenizer.vocab_size, d_model=512, nhead=8, num_layers=6).to(device)
model.load_state_dict(torch.load("evo_decoder.pt", map_location=device))
model.eval()

@torch.no_grad()
def generate_response(prompt, max_length=50, temperature=1.0):
    model.eval()
    input_ids = tokenizer.encode(prompt, return_tensors="pt").to(device)

    for _ in range(max_length):
        logits = model(input_ids)
        logits = logits[:, -1, :] / temperature
        probs = F.softmax(logits, dim=-1)
        next_token = torch.multinomial(probs, num_samples=1)
        input_ids = torch.cat((input_ids, next_token), dim=1)

        if next_token.item() == tokenizer.eos_token_id:
            break

    output = tokenizer.decode(input_ids.squeeze(), skip_special_tokens=True)
    return output[len(prompt):].strip()