test.py

import torch
import json

from language.responder import Responder


# ===== LOAD TOKENIZER =====
class SimpleTokenizer:
    def __init__(self, vocab_path):
        with open(vocab_path, "r") as f:
            self.vocab = json.load(f)
        self.inv_vocab = {v: k for k, v in self.vocab.items()}

    def encode(self, text):
        return [self.vocab.get(t, 1) for t in text.split()]

    def decode(self, ids):
        return " ".join([self.inv_vocab.get(i, "<unk>") for i in ids])


# ===== LOAD MODEL =====
def load_responder(device):
    vocab_path = "artifacts/vocab.json"
    model_path = "artifacts/responder.pt"
    brain_path = "artifacts/replay_buffer.pt"

    tokenizer = SimpleTokenizer(vocab_path)

    # 🔥 GET REAL TARGET_DIM FROM TRAINING DATA
    brain_data = torch.load(brain_path, map_location=device)
    target_dim = brain_data[0][1].shape[-1]

    print(f"[INFO] TARGET_DIM (from training): {target_dim}")

    model = Responder(
        vocab_size=len(tokenizer.vocab),
        embed_dim=192,
        hidden_dim=512,
        target_dim=target_dim
    ).to(device)

    model.load_state_dict(torch.load(model_path, map_location=device))
    model.eval()

    return model, tokenizer


# ===== GENERATE =====
def generate_response(model, tokenizer, text, device):
    with torch.no_grad():

        # 🔥 MUST MATCH hidden_dim (512)
        brain_input = torch.randn(1, 512).to(device)

        output = model.generate_from_brain(
            brain_input=brain_input,
            tokenizer=tokenizer,
            max_len=64,
            intent="general",
            emotion="neutral",
            memory_text=text
        )

        return output


# ===== MAIN LOOP =====
def main():
    device = "cuda" if torch.cuda.is_available() else "cpu"
    print(f"[INFO] Using device: {device}")

    model, tokenizer = load_responder(device)

    print("\n=== RESPONDER TEST MODE ===")
    print("Type 'exit' to quit\n")

    while True:
        user_input = input("You: ")

        if user_input.lower() in ["exit", "quit"]:
            break

        response = generate_response(model, tokenizer, user_input, device)
        print(f"AI: {response}\n")


if __name__ == "__main__":
    main()