Inference.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 = 150 # 기존 코드에서 200으로 설정됨
batch_size = 128

# 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/TinyInst/resolve/main/output.jsonl?download=true",
        DATA_PATH
    )

if not os.path.exists(TOKENIZER_PATH):
    download_file(
        "https://huggingface.co/datasets/Yuchan5386/TinyInst/resolve/main/ko_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("<start>")
sep_id = sp.piece_to_id("<sep>")
end_id = sp.piece_to_id("<end>")
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)

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 LoU(layers.Layer):
    def __init__(self, d_model, clip_value=5.0, eps=1e-6):
        super().__init__()
        self.d_model = d_model
        self.clip_value = float(clip_value)
        self.eps = float(eps)
        self.Q = layers.Dense(d_model, dtype='float32')
        self.K = layers.Dense(d_model, dtype='float32')
        self.V = layers.Dense(d_model, dtype='float32')
        self.norm = layers.LayerNormalization(epsilon=1e-5, dtype='float32')
        self.norm1 = layers.LayerNormalization(epsilon=1e-5, dtype='float32')
        
        self.glu = SwiGLU(d_model, 320)
    def call(self, x):
        x_f32 = tf.cast(x, tf.float32)
        residual = x_f32
        x_f32 = self.norm1(x)

        q = self.Q(x_f32)
        k = self.K(x_f32)
        V = self.V(x_f32)
        g_q = (tf.nn.tanh(q) + 1.0) / 2.0
        g_k = (tf.nn.tanh(k) + 1.0) / 2.0
        score = g_q * g_k

        score = tf.cumsum(score, axis=1) # (B, L, D)
        
        # 💡 수정된 부분: 현재 토큰까지의 누적합 평균으로 정규화
        seq_len = tf.shape(score)[1]
        # [1, 2, 3, ..., L]을 D_model 차원으로 확장
        count_for_mean = tf.cast(tf.range(seq_len) + 1, score.dtype)
        count_for_mean = tf.reshape(count_for_mean, (1, seq_len, 1))
        
        # 누적합을 현재까지의 토큰 개수로 나누어 평균 누적합 계산 (B, L, D)
        score_mean = score / count_for_mean
        
        # 정규화 분모 설정
        denom = tf.maximum(score_mean, self.eps)
        score_norm = score / denom
        # -----------------------------------------------

        score_clipped = tf.clip_by_value(score_norm, -self.clip_value, self.clip_value)
        x_comb = score_clipped * V
        
        out = self.norm(x_comb + residual)
        out = self.glu(out)
        return tf.cast(out, x.dtype)


class Lo(layers.Layer):
    def __init__(self, d_model):
        super().__init__()
        self.d = layers.Dense(64, activation='silu')
        self.w = layers.Dense(d_model)
        self.norm = layers.LayerNormalization(epsilon=1e-5, dtype='float32')

    def call(self, x):
        p = self.d(x)
        p = self.w(p)
        return self.norm(p) + x

class Block(layers.Layer):
    def __init__(self, d_model):
        super().__init__()
        self.lou = LoU(d_model)
        self.lo = Lo(d_model)

    def call(self, x):
        x = self.lou(x)
        x = self.lo(x)
        return x

class ReLM(tf.keras.Model):
    def __init__(self, vocab_size, max_seq_len, d_model, n_layers, dropout_rate=0.1):
        super().__init__()
        self.token_embedding = layers.Embedding(vocab_size, d_model)
        self.pos_embedding = layers.Embedding(max_seq_len, d_model)
        self.blocks = [Block(d_model) for _ in range(n_layers)]
        self.ln_f = layers.LayerNormalization(epsilon=1e-5, dtype="float32")

    def call(self, x, training=False):
        batch_size, seq_len = tf.shape(x)[0], tf.shape(x)[1]
        positions = tf.range(seq_len)[tf.newaxis, :]
        x = self.token_embedding(x) + self.pos_embedding(positions)
        for block in self.blocks:
            x = block(x)
        x = self.ln_f(x)
        embedding_matrix = tf.cast(self.token_embedding.embeddings, x.dtype)
        logits = tf.matmul(x, embedding_matrix, transpose_b=True)
        return tf.cast(logits, tf.float32)


model = ReLM(
    vocab_size=vocab_size,
    max_seq_len=max_len,
    d_model=256,
    n_layers=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)

chat_model.load_weights('/kaggle/working/chat_model.weights.h5')
print("모델 가중치 로드 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))