File size: 10,849 Bytes
6a0cf5a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 | """ループドTransformer本体(HF ``PreTrainedModel`` 互換・単一実装)。
設計 docs/architecture.md §2 の Llama 系レシピ(RMSNorm / RoPE / SwiGLU /
bias なし / weight tying)を素の PyTorch で実装。``forward`` は K 回ループする
形で書き、``k=1`` で標準Transformerに厳密に縮退する(``tests/test_k1_equivalence.py``)。
このファイルは ``save_pretrained`` 時に checkpoint へ複製され、公式
evaluation-pipeline(別プロセス・``trust_remote_code=True``)から import される。
そのため **torch / transformers / 標準ライブラリ以外に依存しない**こと。
probing 用のループ毎中間表現は HF 標準 ``hidden_states``(層ごと)と混ぜず、
別フィールド ``loop_hidden_states`` に格納する。
"""
from __future__ import annotations
import math
from dataclasses import dataclass
import torch
import torch.nn.functional as F
from torch import nn
from transformers.modeling_outputs import ModelOutput
from transformers.modeling_utils import PreTrainedModel
from .configuration_babyloop import BabyloopConfig
# --- ビルディングブロック(Llama系レシピ)---------------------------------
class RMSNorm(nn.Module):
def __init__(self, dim: int, eps: float):
super().__init__()
self.eps = eps
self.weight = nn.Parameter(torch.ones(dim))
def forward(self, x: torch.Tensor) -> torch.Tensor:
dtype = x.dtype
x = x.float()
x = x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
return (x.to(dtype)) * self.weight
def _rotate_half(x: torch.Tensor) -> torch.Tensor:
x1, x2 = x.chunk(2, dim=-1)
return torch.cat((-x2, x1), dim=-1)
def _apply_rope(x: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor) -> torch.Tensor:
# x: (B, H, T, head_dim); cos/sin: (1, 1, T, head_dim)
return x * cos + _rotate_half(x) * sin
class Attention(nn.Module):
"""RoPE 付き causal Multi-Head Attention(dropout なし)。"""
def __init__(self, config: BabyloopConfig):
super().__init__()
self.n_heads = config.n_heads
self.head_dim = config.d_model // config.n_heads
self.qkv_proj = nn.Linear(config.d_model, 3 * config.d_model, bias=config.bias)
self.o_proj = nn.Linear(config.d_model, config.d_model, bias=config.bias)
def forward(self, x, cos, sin, attn_bias):
B, T, C = x.shape
q, k, v = self.qkv_proj(x).split(C, dim=-1)
q = q.view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
k = k.view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
v = v.view(B, T, self.n_heads, self.head_dim).transpose(1, 2)
q = _apply_rope(q, cos, sin)
k = _apply_rope(k, cos, sin)
out = F.scaled_dot_product_attention(
q, k, v, attn_mask=attn_bias, is_causal=attn_bias is None
)
out = out.transpose(1, 2).reshape(B, T, C)
return self.o_proj(out)
class SwiGLU(nn.Module):
def __init__(self, config: BabyloopConfig):
super().__init__()
self.gate_proj = nn.Linear(config.d_model, config.ffn_hidden, bias=config.bias)
self.up_proj = nn.Linear(config.d_model, config.ffn_hidden, bias=config.bias)
self.down_proj = nn.Linear(config.ffn_hidden, config.d_model, bias=config.bias)
def forward(self, x):
return self.down_proj(F.silu(self.gate_proj(x)) * self.up_proj(x))
class Block(nn.Module):
"""Pre-Norm 残差ブロック: h += Attn(RMSNorm(h)); h += SwiGLU(RMSNorm(h))。"""
def __init__(self, config: BabyloopConfig):
super().__init__()
self.attn_norm = RMSNorm(config.d_model, config.rms_eps)
self.attn = Attention(config)
self.mlp_norm = RMSNorm(config.d_model, config.rms_eps)
self.mlp = SwiGLU(config)
def forward(self, h, cos, sin, attn_bias):
h = h + self.attn(self.attn_norm(h), cos, sin, attn_bias)
h = h + self.mlp(self.mlp_norm(h))
return h
# --- 出力コンテナ -----------------------------------------------------------
@dataclass
class LoopedModelOutput(ModelOutput):
last_hidden_state: torch.FloatTensor | None = None
hidden_states: tuple[torch.FloatTensor, ...] | None = None
loop_hidden_states: tuple[torch.FloatTensor, ...] | None = None
@dataclass
class LoopedCausalLMOutput(ModelOutput):
loss: torch.FloatTensor | None = None
logits: torch.FloatTensor | None = None
hidden_states: tuple[torch.FloatTensor, ...] | None = None
loop_hidden_states: tuple[torch.FloatTensor, ...] | None = None
# --- PreTrainedModel ラッパ -------------------------------------------------
class LoopedPreTrainedModel(PreTrainedModel):
config_class = BabyloopConfig
base_model_prefix = "model"
supports_gradient_checkpointing = False
def _init_weights(self, module):
std = 0.02
if isinstance(module, nn.Linear):
nn.init.normal_(module.weight, mean=0.0, std=std)
if module.bias is not None:
nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
nn.init.normal_(module.weight, mean=0.0, std=std)
elif isinstance(module, RMSNorm):
nn.init.ones_(module.weight)
class LoopedModel(LoopedPreTrainedModel):
"""重み共有 core ブロックを K 回反復するバックボーン(lm_head なし)。"""
def __init__(self, config: BabyloopConfig):
super().__init__(config)
self.config = config
self.embed_tokens = nn.Embedding(config.vocab_size, config.d_model)
self.prelude = nn.ModuleList(Block(config) for _ in range(config.n_prelude))
self.core = nn.ModuleList(Block(config) for _ in range(config.n_core))
self.coda = nn.ModuleList(Block(config) for _ in range(config.n_coda))
self.final_norm = RMSNorm(config.d_model, config.rms_eps)
head_dim = config.d_model // config.n_heads
inv_freq = 1.0 / (
config.rope_base ** (torch.arange(0, head_dim, 2, dtype=torch.float32) / head_dim)
)
self.register_buffer("inv_freq", inv_freq, persistent=False)
self.post_init()
def get_input_embeddings(self):
return self.embed_tokens
def set_input_embeddings(self, value):
self.embed_tokens = value
def _rope(self, T: int, device, dtype):
t = torch.arange(T, device=device, dtype=torch.float32)
freqs = torch.outer(t, self.inv_freq.to(device))
emb = torch.cat((freqs, freqs), dim=-1)
return emb.cos().to(dtype)[None, None], emb.sin().to(dtype)[None, None]
def _attn_bias(self, attention_mask, T, device, dtype):
# padding が無ければ None を返し、SDPA の is_causal 経路(高速)に乗せる。
if attention_mask is None or bool((attention_mask == 1).all()):
return None
causal = torch.ones(T, T, device=device, dtype=torch.bool).triu(1)
key_pad = attention_mask.to(device) == 0 # (B, T)
mask = causal[None, None] | key_pad[:, None, None, :]
bias = torch.zeros(mask.shape, device=device, dtype=dtype)
return bias.masked_fill(mask, torch.finfo(dtype).min)
def forward(
self,
input_ids=None,
attention_mask=None,
inputs_embeds=None,
output_hidden_states=False,
**kwargs,
) -> LoopedModelOutput:
if inputs_embeds is None:
inputs_embeds = self.embed_tokens(input_ids)
h = inputs_embeds
residual_input = inputs_embeds # inject_input 用(③で本格化、①は false)
B, T, _ = h.shape
cos, sin = self._rope(T, h.device, h.dtype)
attn_bias = self._attn_bias(attention_mask, T, h.device, h.dtype)
all_hidden = [h] if output_hidden_states else None
loop_hidden = []
for block in self.prelude:
h = block(h, cos, sin, attn_bias)
if output_hidden_states:
all_hidden.append(h)
for _ in range(self.config.k):
for block in self.core:
h = block(h, cos, sin, attn_bias)
if output_hidden_states:
all_hidden.append(h)
if self.config.inject_input:
h = h + residual_input
loop_hidden.append(h)
for block in self.coda:
h = block(h, cos, sin, attn_bias)
if output_hidden_states:
all_hidden.append(h)
h = self.final_norm(h)
return LoopedModelOutput(
last_hidden_state=h,
hidden_states=tuple(all_hidden) if output_hidden_states else None,
loop_hidden_states=tuple(loop_hidden),
)
class LoopedForCausalLM(LoopedPreTrainedModel):
"""言語モデリングヘッド付き(``AutoModelForCausalLM`` 互換)。"""
_tied_weights_keys = ["lm_head.weight"]
def __init__(self, config: BabyloopConfig):
super().__init__(config)
self.model = LoopedModel(config)
self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False)
self.post_init()
def get_input_embeddings(self):
return self.model.embed_tokens
def set_input_embeddings(self, value):
self.model.embed_tokens = value
def get_output_embeddings(self):
return self.lm_head
def set_output_embeddings(self, new):
self.lm_head = new
def forward(
self,
input_ids=None,
attention_mask=None,
inputs_embeds=None,
labels=None,
output_hidden_states=False,
**kwargs,
) -> LoopedCausalLMOutput:
out = self.model(
input_ids=input_ids,
attention_mask=attention_mask,
inputs_embeds=inputs_embeds,
output_hidden_states=output_hidden_states,
)
logits = self.lm_head(out.last_hidden_state)
loss = None
if labels is not None:
shift_logits = logits[:, :-1, :].contiguous()
shift_labels = labels[:, 1:].contiguous()
loss = F.cross_entropy(
shift_logits.view(-1, shift_logits.size(-1)),
shift_labels.view(-1),
ignore_index=-100,
)
return LoopedCausalLMOutput(
loss=loss,
logits=logits,
hidden_states=out.hidden_states,
loop_hidden_states=out.loop_hidden_states,
)
# 公式 evaluation-pipeline が trust_remote_code で AutoModel 系から読めるよう登録。
# save_pretrained 時に auto_map と本ファイル群が checkpoint へ複製される。
BabyloopConfig.register_for_auto_class()
LoopedModel.register_for_auto_class("AutoModel")
LoopedForCausalLM.register_for_auto_class("AutoModelForCausalLM")
|