Spaces:
Sleeping
Sleeping
File size: 9,637 Bytes
f37be5a | 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 | from __future__ import annotations
import math
from dataclasses import dataclass
import torch
import torch.nn as nn
import torch.nn.functional as F
from src.model.backbone import TransformerBlock
from src.model.config import ModelConfig
@dataclass(frozen=True)
class FogTaskProfile:
name: str
compare_ratio: float
memory_ratio: float
expand_ratio: float
gate_ratio: float
hybrid_start_ratio: float
max_layers: int
adapter_scale: float
@dataclass(frozen=True)
class FogLayerGeometry:
layer_idx: int
stage: str
d_compare: int
d_memory: int
d_expand: int
d_gate: int
residual_scale: float
def _align_to_heads(value: int, n_heads: int) -> int:
aligned = max(n_heads, (value // n_heads) * n_heads)
if aligned < value:
aligned += n_heads
return aligned
def resolve_fog_task_profile(cfg: ModelConfig) -> FogTaskProfile:
profile = cfg.fog_task_profile
if profile == "stories":
return FogTaskProfile(
name="stories",
compare_ratio=0.18,
memory_ratio=0.60,
expand_ratio=1.35,
gate_ratio=0.08,
hybrid_start_ratio=0.67,
max_layers=2,
adapter_scale=0.11,
)
if profile == "code":
return FogTaskProfile(
name="code",
compare_ratio=0.34,
memory_ratio=0.82,
expand_ratio=1.70,
gate_ratio=0.12,
hybrid_start_ratio=0.45,
max_layers=3,
adapter_scale=0.16,
)
if profile == "math":
return FogTaskProfile(
name="math",
compare_ratio=0.42,
memory_ratio=0.70,
expand_ratio=1.95,
gate_ratio=0.14,
hybrid_start_ratio=0.45,
max_layers=3,
adapter_scale=0.18,
)
if profile == "synthetic":
return FogTaskProfile(
name="synthetic",
compare_ratio=0.32,
memory_ratio=0.84,
expand_ratio=2.10,
gate_ratio=0.14,
hybrid_start_ratio=0.0,
max_layers=cfg.n_layers,
adapter_scale=0.20,
)
return FogTaskProfile(
name="balanced",
compare_ratio=cfg.fog_compare_ratio,
memory_ratio=cfg.fog_memory_ratio,
expand_ratio=cfg.fog_expand_ratio,
gate_ratio=cfg.fog_gate_ratio,
hybrid_start_ratio=0.55,
max_layers=min(2, cfg.n_layers),
adapter_scale=0.13,
)
def select_fog_adapter_layers(cfg: ModelConfig, profile: FogTaskProfile) -> list[int]:
start_idx = min(cfg.n_layers - 1, max(0, int(cfg.n_layers * profile.hybrid_start_ratio)))
candidate_layers = list(range(start_idx, cfg.n_layers))
if len(candidate_layers) <= profile.max_layers:
return candidate_layers
return candidate_layers[-profile.max_layers:]
def build_fog_geometries(cfg: ModelConfig) -> list[FogLayerGeometry]:
profile = resolve_fog_task_profile(cfg)
adapter_layers = select_fog_adapter_layers(cfg, profile)
geometries: list[FogLayerGeometry] = []
if not adapter_layers:
return geometries
for adapter_pos, layer_idx in enumerate(adapter_layers):
depth = adapter_pos / max(len(adapter_layers) - 1, 1)
if depth < 0.34:
stage = "early"
compare_ratio = profile.compare_ratio * 0.95
memory_ratio = profile.memory_ratio * 0.90
expand_ratio = profile.expand_ratio * 0.90
gate_ratio = profile.gate_ratio * 0.90
residual_scale = profile.adapter_scale * 0.85
elif depth < 0.67:
stage = "middle"
compare_ratio = profile.compare_ratio
memory_ratio = profile.memory_ratio
expand_ratio = profile.expand_ratio
gate_ratio = profile.gate_ratio
residual_scale = profile.adapter_scale
else:
stage = "late"
compare_ratio = profile.compare_ratio * 1.05
memory_ratio = profile.memory_ratio * 1.05
expand_ratio = profile.expand_ratio * 1.10
gate_ratio = profile.gate_ratio * 1.10
residual_scale = profile.adapter_scale * 1.10
geometries.append(
FogLayerGeometry(
layer_idx=layer_idx,
stage=stage,
d_compare=_align_to_heads(max(cfg.n_heads, int(cfg.d_model * compare_ratio)), cfg.n_heads),
d_memory=_align_to_heads(max(cfg.n_heads, int(cfg.d_model * memory_ratio)), cfg.n_heads),
d_expand=max(cfg.d_model, int(cfg.d_model * expand_ratio)),
d_gate=max(4, int(cfg.d_model * gate_ratio)),
residual_scale=residual_scale,
)
)
return geometries
class FogAttention(nn.Module):
def __init__(self, d_model: int, d_compare: int, d_memory: int, n_heads: int, dropout: float) -> None:
super().__init__()
assert d_compare % n_heads == 0
assert d_memory % n_heads == 0
self.n_heads = n_heads
self.compare_head_dim = d_compare // n_heads
self.memory_head_dim = d_memory // n_heads
self.d_memory = d_memory
self.q_proj = nn.Linear(d_model, d_compare)
self.k_proj = nn.Linear(d_model, d_compare)
self.v_proj = nn.Linear(d_model, d_memory)
self.out_proj = nn.Linear(d_memory, d_model)
self.attn_dropout = nn.Dropout(dropout)
def forward(self, x: torch.Tensor, mask: torch.Tensor) -> torch.Tensor:
b, t, _ = x.shape
q = self.q_proj(x).view(b, t, self.n_heads, self.compare_head_dim).transpose(1, 2)
k = self.k_proj(x).view(b, t, self.n_heads, self.compare_head_dim).transpose(1, 2)
v = self.v_proj(x).view(b, t, self.n_heads, self.memory_head_dim).transpose(1, 2)
scores = torch.matmul(q, k.transpose(-2, -1)) / math.sqrt(self.compare_head_dim)
scores = scores.masked_fill(mask == 0, float("-inf"))
attn = self.attn_dropout(torch.softmax(scores, dim=-1))
y = torch.matmul(attn, v)
y = y.transpose(1, 2).contiguous().view(b, t, self.d_memory)
return self.out_proj(y)
class FogFFN(nn.Module):
def __init__(self, d_model: int, d_expand: int, d_gate: int, dropout: float) -> None:
super().__init__()
self.expand = nn.Linear(d_model, d_expand)
self.gate = nn.Linear(d_model, d_gate)
self.gate_up = nn.Linear(d_gate, d_expand)
self.compress = nn.Linear(d_expand, d_model)
self.dropout = nn.Dropout(dropout)
def forward(self, x: torch.Tensor) -> torch.Tensor:
expanded = F.silu(self.expand(x))
gate = torch.sigmoid(self.gate_up(F.silu(self.gate(x))))
return self.compress(self.dropout(expanded * gate))
class FogAdapterBlock(nn.Module):
def __init__(self, cfg: ModelConfig, geometry: FogLayerGeometry) -> None:
super().__init__()
self.geometry = geometry
self.ln1 = nn.LayerNorm(cfg.d_model)
self.ln2 = nn.LayerNorm(cfg.d_model)
self.attn = FogAttention(cfg.d_model, geometry.d_compare, geometry.d_memory, cfg.n_heads, cfg.dropout)
self.ffn = FogFFN(cfg.d_model, geometry.d_expand, geometry.d_gate, cfg.dropout)
self.drop = nn.Dropout(cfg.dropout)
self.attn_scale = nn.Parameter(torch.tensor(float(geometry.residual_scale)))
self.ffn_scale = nn.Parameter(torch.tensor(float(geometry.residual_scale)))
def forward(self, x: torch.Tensor, mask: torch.Tensor) -> torch.Tensor:
attn_update = self.attn_scale * self.drop(self.attn(self.ln1(x), mask))
ffn_update = self.ffn_scale * self.drop(self.ffn(self.ln2(x + attn_update)))
return attn_update + ffn_update
class FogFlowBackbone(nn.Module):
def __init__(self, cfg: ModelConfig) -> None:
super().__init__()
self.cfg = cfg
self.profile = resolve_fog_task_profile(cfg)
self.tok_emb = nn.Embedding(cfg.vocab_size, cfg.d_model)
self.pos_emb = nn.Embedding(cfg.max_seq_len, cfg.d_model)
self.drop = nn.Dropout(cfg.dropout)
self.blocks = nn.ModuleList([TransformerBlock(cfg, i) for i in range(cfg.n_layers)])
self.layer_geometries = build_fog_geometries(cfg)
self.fog_blocks = nn.ModuleDict(
{str(geom.layer_idx): FogAdapterBlock(cfg, geom) for geom in self.layer_geometries}
)
self.fog_layers = [geom.layer_idx for geom in self.layer_geometries]
self.ln_final = nn.LayerNorm(cfg.d_model)
self.register_buffer(
"_causal_mask",
torch.tril(torch.ones(cfg.max_seq_len, cfg.max_seq_len, dtype=torch.bool)).unsqueeze(0).unsqueeze(0),
persistent=False,
)
def forward(self, input_ids: torch.Tensor) -> dict[str, torch.Tensor | list[torch.Tensor] | list[int] | str]:
_, t = input_ids.shape
pos = torch.arange(t, device=input_ids.device).unsqueeze(0)
x = self.drop(self.tok_emb(input_ids) + self.pos_emb(pos))
layer_outputs = [x]
mask = self._causal_mask[:, :, :t, :t]
for idx, block in enumerate(self.blocks):
x = block(x, layer_outputs)
if str(idx) in self.fog_blocks:
x = x + self.fog_blocks[str(idx)](x, mask)
layer_outputs.append(x)
hidden = self.ln_final(x)
return {
"hidden": hidden,
"layer_outputs": layer_outputs,
"fog_layers": self.fog_layers,
"fog_profile": self.profile.name,
"flow_type": "fog_hybrid",
}
|