AlphaS2S.py

import tensorflow as tf
from tensorflow.keras import layers, Model
import numpy as np
import tensorflow.keras.backend as K
from tensorflow.keras import mixed_precision
import sentencepiece as spm
import os, json
import requests

print('1')

tf.get_logger().setLevel("ERROR")
SEED = 42
tf.random.set_seed(SEED)
np.random.seed(SEED)
max_len = 256 # 기존 코드에서 200으로 설정됨
batch_size = 4

# TPU 초기화 (기존 코드와 동일)
try:
    resolver = tf.distribute.cluster_resolver.TPUClusterResolver(tpu="local")
    tf.tpu.experimental.initialize_tpu_system(resolver)
    strategy = tf.distribute.TPUStrategy(resolver)
    print("✅ TPU 초기화 완료:", resolver.cluster_spec().as_dict())
    on_tpu = True

except Exception as e:
    print("⚠️ TPU 미사용, GPU/CPU로 진행:", e)
    strategy = tf.distribute.get_strategy()
    on_tpu = False

# Mixed precision (기존 코드와 동일)
policy = mixed_precision.Policy("mixed_bfloat16" if on_tpu else "float32")
mixed_precision.set_global_policy(policy)
print("✅ Mixed precision:", policy)

# =======================
# 1) 파일 다운로드 및 토크나이저 초기화 (기존 코드와 동일)
# =======================

def download_file(url, save_path):
    r = requests.get(url, stream=True)
    r.raise_for_status()
    with open(save_path, "wb") as f:
        for chunk in r.iter_content(8192*2):
            f.write(chunk)
    print(f"✅ {save_path} 저장됨")

DATA_PATH = "converted.jsonl"
TOKENIZER_PATH = "ko_unigram.model"

if not os.path.exists(DATA_PATH):
    download_file(
        "https://huggingface.co/datasets/Yuchan5386/Multiturn/resolve/main/dataset_shuffled.jsonl?download=true",
        DATA_PATH
    )

if not os.path.exists(TOKENIZER_PATH):
    download_file(
        "https://huggingface.co/datasets/Yuchan5386/Multiturn/resolve/main/unigram.model?download=true",
        TOKENIZER_PATH
    )

sp = spm.SentencePieceProcessor(TOKENIZER_PATH)

pad_id = sp.piece_to_id("<pad>") if sp.piece_to_id("<pad>") != -1 else 0
start_id = sp.piece_to_id("<sos>")
context_s_id = sp.piece_to_id("<context>")
context_e_id = sp.piece_to_id("</context>")
user_s_id = sp.piece_to_id("<user>")
user_e_id = sp.piece_to_id("</user>")
end_id = sp.piece_to_id("<eos>")
unk_id = sp.piece_to_id("<unk>")
vocab_size = sp.get_piece_size()
print(f"✅ Vocabulary size: {vocab_size}")

def text_to_ids(text):
    return sp.encode(text, out_type=int)

def ids_to_text(ids):
    return sp.decode(ids)

# =======================
# JSONL → TF Dataset 로드 (ID 레벨 특수 토큰 포함)
# =======================
def jsonl_stream(file_path):
    with open(file_path, "r", encoding="utf-8") as f:
        for line in f:
            data = json.loads(line)
            context = data["context"]
            prompt = data["prompt"]
            answer = data["answer"]

            # =======================
            # Encoder input: ID 레벨에서 특수 토큰 명시
            # =======================
            enc_ids = [context_s_id] + text_to_ids(context) + [context_e_id] + \
                      [user_s_id] + text_to_ids(prompt) + [user_e_id]
            enc_ids = enc_ids[:max_len]  # max_len 제한

            # =======================
            # Decoder input: <sos> + answer
            # =======================
            dec_input_ids = [start_id] + text_to_ids(answer)
            dec_input_ids = dec_input_ids[:max_len]

            # =======================
            # Target: answer + <eos>
            # =======================
            target_ids = text_to_ids(answer) + [end_id]
            target_ids = target_ids[:max_len]

            # =======================
            # Padding
            # =======================
            enc_ids += [pad_id] * (max_len - len(enc_ids))
            dec_input_ids += [pad_id] * (max_len - len(dec_input_ids))
            target_ids += [pad_id] * (max_len - len(target_ids))

            yield (
                tf.convert_to_tensor(enc_ids, dtype=tf.int32),
                tf.convert_to_tensor(dec_input_ids, dtype=tf.int32),
                tf.convert_to_tensor(target_ids, dtype=tf.int32),
            )

# =======================
# TF Dataset 생성
# =======================
dataset = tf.data.Dataset.from_generator(
    lambda: jsonl_stream(DATA_PATH),
    output_signature=(
        tf.TensorSpec(shape=(max_len,), dtype=tf.int32),  # enc_inputs
        tf.TensorSpec(shape=(max_len,), dtype=tf.int32),  # dec_inputs
        tf.TensorSpec(shape=(max_len,), dtype=tf.int32),  # target
    )
)

# 학습을 위해 딕셔너리 형태로 매핑
def map_fn(enc_input, dec_input, dec_target):
    return {"enc_inputs": enc_input, "dec_inputs": dec_input}, dec_target

dataset = dataset.map(map_fn, num_parallel_calls=tf.data.AUTOTUNE)
dataset = dataset.shuffle(1000, seed=SEED).batch(batch_size, drop_remainder=True).prefetch(tf.data.AUTOTUNE)

with strategy.scope():
    dist_dataset = strategy.experimental_distribute_dataset(dataset)

print("✅ ID 레벨 특수 토큰 적용 Dataset 로드 완료:", dist_dataset)

# =======================
# 3) 모델 레이어 (기존 코드 유지)
# =======================

class SwiGLU(layers.Layer):
    def __init__(self, d_model, d_ff):
        super().__init__()
        self.proj = layers.Dense(d_ff)
        self.out = layers.Dense(d_model)
    def call(self, x):
        x_proj = self.proj(x)
        x_val, x_gate = tf.split(x_proj, 2, axis=-1)
        return self.out(x_val * tf.nn.silu(x_gate))
        

class SwiGLU(layers.Layer):
    def __init__(self, d_model, d_ff):
        super().__init__()
        self.proj = layers.Dense(d_ff*2)
        self.out = layers.Dense(d_model)
    def call(self, x):
        x_proj = self.proj(x)
        x_val, x_gate = tf.split(x_proj, 2, axis=-1)
        return self.out(x_val * tf.nn.silu(x_gate))

class EncoderBlock(layers.Layer):
    def __init__(self, d_model, num_heads, dff, dropout=0.1):
        super().__init__()
        self.mha = layers.MultiHeadAttention(num_heads=num_heads, key_dim=d_model)
        self.ffn = SwiGLU(d_model, dff)
        self.norm1 = layers.LayerNormalization(epsilon=1e-6)
        self.norm2 = layers.LayerNormalization(epsilon=1e-6)
        self.dropout1 = layers.Dropout(dropout)
        self.dropout2 = layers.Dropout(dropout)
    def call(self, x, mask=None, training=False):
        attn_out = self.dropout1(self.mha(x, x, x, attention_mask=mask), training=training)
        out1 = self.norm1(x + attn_out)
        ffn_out = self.dropout2(self.ffn(out1), training=training)
        return self.norm2(out1 + ffn_out)

class DecoderBlock(layers.Layer):
    def __init__(self, d_model, num_heads, dff, dropout=0.1):
        super().__init__()
        self.self_mha = layers.MultiHeadAttention(num_heads=num_heads, key_dim=d_model)
        self.cross_mha = layers.MultiHeadAttention(num_heads=num_heads, key_dim=d_model)
        self.ffn = SwiGLU(d_model, dff)
        self.norm1 = layers.LayerNormalization(epsilon=1e-6)
        self.norm2 = layers.LayerNormalization(epsilon=1e-6)
        self.norm3 = layers.LayerNormalization(epsilon=1e-6)
        self.dropout1 = layers.Dropout(dropout)
        self.dropout2 = layers.Dropout(dropout)
        self.dropout3 = layers.Dropout(dropout)
    def call(self, x, enc_out, training=False):
        attn1 = self.dropout1(self.self_mha(x, x, x, use_causal_mask=True), training=training)
        out1 = self.norm1(x + attn1)
        attn2 = self.dropout2(self.cross_mha(out1, enc_out, enc_out), training=training)
        out2 = self.norm2(out1 + attn2)
        ffn_out = self.dropout3(self.ffn(out2), training=training)
        return self.norm3(out2 + ffn_out)

class Transformer(tf.keras.Model):
    def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size, target_vocab_size, max_len=256, dropout=0.1):
        super().__init__()
        self.max_len = max_len
        self.d_model = d_model
        self.enc_embedding = layers.Embedding(input_vocab_size, d_model)
        self.enc_pos_embedding = layers.Embedding(max_len, d_model)
        self.dec_embedding = layers.Embedding(target_vocab_size, d_model)
        self.dec_pos_embedding = layers.Embedding(max_len, d_model)
        self.enc_layers = [EncoderBlock(d_model, num_heads, dff, dropout) for _ in range(num_layers)]
        self.dec_layers = [DecoderBlock(d_model, num_heads, dff, dropout) for _ in range(num_layers)]
        self.final_layer = layers.Dense(target_vocab_size)
    def call(self, inputs, training=False):
        enc_inputs = inputs["enc_inputs"]
        dec_inputs = inputs["dec_inputs"]
        enc_pos = tf.range(tf.shape(enc_inputs)[1])[tf.newaxis, :]
        dec_pos = tf.range(tf.shape(dec_inputs)[1])[tf.newaxis, :]
        x = self.enc_embedding(enc_inputs) + self.enc_pos_embedding(enc_pos)
        for layer in self.enc_layers: x = layer(x, training=training)
        enc_out = x
        y = self.dec_embedding(dec_inputs) + self.dec_pos_embedding(dec_pos)
        for layer in self.dec_layers: y = layer(y, enc_out, training=training)
        return self.final_layer(y)

# 5) 학습 설정 및 실행
# =======================

def masked_loss(y_true, y_pred):
    loss = loss_fn(y_true, y_pred)
    mask = tf.cast(tf.not_equal(y_true, pad_id), tf.float32)
    # mixed_bfloat16 사용 시 나눗셈 시 NaN 방지
    sum_mask = tf.reduce_sum(mask)
    safe_sum_mask = tf.where(sum_mask == 0.0, 1.0, sum_mask)
    masked_loss = tf.reduce_sum(loss * mask) / safe_sum_mask
    return masked_loss

def masked_perplexity(y_true, y_pred):
    loss = loss_fn(y_true, y_pred)
    mask = tf.cast(tf.not_equal(y_true, pad_id), tf.float32)
    sum_mask = tf.reduce_sum(mask)
    safe_sum_mask = tf.where(sum_mask == 0.0, 1.0, sum_mask)
    avg_loss = tf.reduce_sum(loss * mask) / safe_sum_mask
    return tf.exp(tf.minimum(avg_loss, 10.0))  

def create_lr_schedule(initial_lr=5e-5, decay_steps=10000, decay_rate=0.9):
    return tf.keras.optimizers.schedules.ExponentialDecay(
        initial_learning_rate=initial_lr,
        decay_steps=decay_steps,
        decay_rate=decay_rate,
        staircase=False
    )

with strategy.scope():
    # ⚠️ 수정: chat_vocab_size 대신 정의된 vocab_size 사용
    chat_model = Transformer(num_layers=4, d_model=512, num_heads=8, dff=2048, input_vocab_size=vocab_size, target_vocab_size=vocab_size, max_len=256, dropout=0.1)
    
    dummy_input = {
        "enc_inputs": tf.zeros((1, max_len), dtype=tf.int32),
        "dec_inputs": tf.zeros((1, max_len), dtype=tf.int32)
    }
    _ = chat_model(dummy_input)
    
    loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True, reduction='none')


    # 옵티마이저 설정
    optimizer = tf.keras.optimizers.Adam(
        learning_rate=create_lr_schedule(),
        beta_1=0.9,
        beta_2=0.95,
        epsilon=1e-8,
        clipnorm=1.0
    )

    # 모델 컴파일
    chat_model.compile(
        optimizer=optimizer,
        loss=masked_loss,
        metrics=[
            masked_perplexity
        ]
    )
    chat_model.summary()
    print("✅ 모델 컴파일 완료, 학습 시작...")
    # ⚠️ 학습 실행
    history = chat_model.fit(dataset, epochs=1, verbose=1)

# 가중치 저장
chat_model.save_weights("chat_model.weights.h5")
print("\n✅ 모델 가중치 저장 완료!")

# =======================
# 6) 추론 함수 (기존 코드 유지)
# =======================

def generate_text_topp(model, prompt, max_len=150, max_gen=100, p=0.9, temperature=0.8, min_len=20):
    # 인코더 입력은 <start> Prompt <sep> 만 사용
    model_input = text_to_ids(f"<start> {prompt} <sep>")
    model_input = model_input[:max_len]
    generated = list(model_input)  
    for step in range(max_gen):
        current_len = len(generated)
        # 현재까지 생성된 시퀀스를 입력으로 사용
        if current_len > max_len:
            input_seq = generated[-max_len:]
        else:
            input_seq = generated   
        # 패딩
        input_padded = np.pad(input_seq, (0, max_len - len(input_seq)), constant_values=pad_id)
        input_tensor = tf.convert_to_tensor([input_padded])
        # 모델 추론 (enc_inputs, dec_inputs 모두 동일한 시퀀스를 사용)
        dummy_input = {
            "enc_inputs": input_tensor,
            "dec_inputs": input_tensor
        }
        logits = model(dummy_input, training=False)
        # 다음 토큰의 로짓은 시퀀스의 마지막 토큰 위치에서 가져옴 (0-based index: current_len - 1)
        # 하지만 패딩 후 input_tensor의 실제 시퀀스 길이는 len(input_seq)
        next_token_logits = logits[0, len(input_seq) - 1].numpy()
        # 특수 토큰 생성 억제
        next_token_logits[end_id] -= 5.0
        next_token_logits[pad_id] -= 10.0
        probs = tf.nn.softmax(next_token_logits / temperature).numpy()
        sorted_indices = np.argsort(probs)[::-1]
        sorted_probs = probs[sorted_indices]
        # Top-p (Nucleus) Sampling
        cumulative_probs = np.cumsum(sorted_probs)
        cutoff = np.searchsorted(cumulative_probs, p)
        top_indices = sorted_indices[:cutoff + 1]
        top_probs = sorted_probs[:cutoff + 1]
        top_probs /= np.sum(top_probs)
        next_token_id = np.random.choice(top_indices, p=top_probs)
        if next_token_id == end_id and len(generated) >= min_len:
            break    
        generated.append(int(next_token_id))
    # <start> 토큰 제거 및 <sep> 이전 부분 제거
    try:
        sep_index = generated.index(sep_id)
        # <sep> 이후부터 <end> 이전까지의 응답만 반환
        result_ids = generated[sep_index + 1:]
        try:
            end_index = result_ids.index(end_id)
            result_ids = result_ids[:end_index]
        except ValueError:
            pass
        return ids_to_text(result_ids)
    except ValueError:
        return ids_to_text(generated) # <sep>이 없으면 전체 반환

print("\n\n===== 생성 결과 =====")  
# 모델이 1 epoch만 학습되었으므로 의미 있는 결과가 아닐 수 있습니다.
print(generate_text_topp(chat_model, "제가 이따가 버스를 타야 해서 준비 좀 해야겠어요. 재미있는 대화였습니다!", p=0.9))