generate.py

import torch
import torch.nn.functional as F
from transformers import GPT2Tokenizer
from evo_decoder import EvoDecoder
from search_utils import web_search

# 🔧 Load model and tokenizer
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
tokenizer.pad_token = tokenizer.eos_token

model = EvoDecoder(
    vocab_size=tokenizer.vocab_size,
    d_model=256,
    nhead=4,
    num_layers=3,
    dim_feedforward=512
).to(device)

model.load_state_dict(torch.load("evo_decoder.pt", map_location=device))
model.eval()

@torch.no_grad()
def generate_response(question, context="", use_rag=False, temperature=1.0):
    if not context and use_rag:
        context = web_search(question)

    prompt = f"Context: {context}\nQuestion: {question}\nAnswer:"
    input_ids = tokenizer.encode(prompt, return_tensors="pt").to(device)

    for _ in range(128):
        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[0], skip_special_tokens=True)
    return output[len(prompt):].strip()