File size: 7,987 Bytes
a7e89c0 | 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 | """
text_detector_model.py
======================
Standalone model definition for HybridAITextDetector.
Import this in both training scripts and the Gradio app.
Architecture:
DeBERTa-v3-small → [BiLSTM | CNN | Transformer] → CrossAttentionFusion → Classifier
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import AutoModel
# ─── Constants ───────────────────────────────────────────────────────────────
MODEL_NAME = "microsoft/deberta-v3-small"
MAX_LENGTH = 128
NUM_CLASSES = 1 # binary: sigmoid output
# ─── Sub-modules ─────────────────────────────────────────────────────────────
class AttentionPool(nn.Module):
"""Soft attention pooling over a sequence of vectors."""
def __init__(self, dim: int):
super().__init__()
self.attn = nn.Linear(dim, 1)
def forward(self, x: torch.Tensor, mask: torch.Tensor = None) -> torch.Tensor:
weights = self.attn(x) # (B, T, 1)
if mask is not None:
weights = weights.masked_fill(mask.unsqueeze(-1) == 0, float("-inf"))
weights = torch.softmax(weights, dim=1) # (B, T, 1)
return (weights * x).sum(dim=1) # (B, dim)
class BiLSTMBranch(nn.Module):
"""2-layer Bidirectional LSTM with Attention Pooling."""
def __init__(self, input_dim: int, hidden_dim: int = 128):
super().__init__()
self.lstm = nn.LSTM(
input_dim, hidden_dim,
num_layers=2,
batch_first=True,
dropout=0.2,
bidirectional=True,
)
self.pool = AttentionPool(hidden_dim * 2)
self.proj = nn.Linear(hidden_dim * 2, 128)
def forward(self, x: torch.Tensor, mask: torch.Tensor = None) -> torch.Tensor:
out, _ = self.lstm(x) # (B, T, 256)
pooled = self.pool(out, mask) # (B, 256)
return F.gelu(self.proj(pooled)) # (B, 128)
class CNNBranch(nn.Module):
"""Multi-kernel 1D CNN with Global MaxPooling."""
def __init__(self, input_dim: int):
super().__init__()
self.conv3 = nn.Conv1d(input_dim, 64, kernel_size=3, padding=1)
self.conv5 = nn.Conv1d(input_dim, 64, kernel_size=5, padding=2)
self.conv7 = nn.Conv1d(input_dim, 64, kernel_size=7, padding=3)
self.bn3 = nn.BatchNorm1d(64)
self.bn5 = nn.BatchNorm1d(64)
self.bn7 = nn.BatchNorm1d(64)
self.proj = nn.Linear(192, 128)
def forward(self, x: torch.Tensor) -> torch.Tensor:
x_t = x.permute(0, 2, 1) # (B, D, T)
c3 = F.gelu(self.bn3(self.conv3(x_t)))
c5 = F.gelu(self.bn5(self.conv5(x_t)))
c7 = F.gelu(self.bn7(self.conv7(x_t)))
p3 = c3.max(dim=-1).values
p5 = c5.max(dim=-1).values
p7 = c7.max(dim=-1).values
cat = torch.cat([p3, p5, p7], dim=-1) # (B, 192)
return F.gelu(self.proj(cat)) # (B, 128)
class TransformerBranch(nn.Module):
"""Lightweight Transformer Encoder with Attention Pooling."""
def __init__(self, input_dim: int):
super().__init__()
self.proj_in = nn.Linear(input_dim, 128)
encoder_layer = nn.TransformerEncoderLayer(
d_model=128, nhead=4,
dim_feedforward=256,
dropout=0.1,
batch_first=True,
norm_first=True,
)
self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=2)
self.pool = AttentionPool(128)
def forward(self, x: torch.Tensor, mask: torch.Tensor = None) -> torch.Tensor:
x = F.gelu(self.proj_in(x)) # (B, T, 128)
src_key_padding_mask = (mask == 0) if mask is not None else None
out = self.transformer(x, src_key_padding_mask=src_key_padding_mask)
return self.pool(out, mask) # (B, 128)
class CrossAttentionFusion(nn.Module):
"""Fuse 3 branch outputs via multi-head self-attention (3-token sequence)."""
def __init__(self, dim: int = 128):
super().__init__()
self.q = nn.Linear(dim, dim)
self.k = nn.Linear(dim, dim)
self.v = nn.Linear(dim, dim)
self.scale = dim ** 0.5
self.proj = nn.Linear(dim, dim)
def forward(
self,
lstm_out: torch.Tensor,
cnn_out: torch.Tensor,
trans_out: torch.Tensor,
) -> torch.Tensor:
stacked = torch.stack([lstm_out, cnn_out, trans_out], dim=1) # (B, 3, 128)
Q = self.q(stacked)
K = self.k(stacked)
V = self.v(stacked)
attn = torch.softmax(torch.bmm(Q, K.transpose(1, 2)) / self.scale, dim=-1)
out = torch.bmm(attn, V).mean(dim=1) # (B, 128)
return F.gelu(self.proj(out))
# ─── Main Model ──────────────────────────────────────────────────────────────
class HybridAITextDetector(nn.Module):
"""
Hybrid AI-generated text detector.
Inputs
------
input_ids : (B, T) long tensor
attention_mask : (B, T) long tensor — 1 = real token, 0 = pad
token_type_ids : (B, T) long tensor
Output
------
logits : (B, 1) float — apply sigmoid to get P(AI-generated)
"""
def __init__(self):
super().__init__()
self.deberta = AutoModel.from_pretrained(MODEL_NAME)
# Freeze first 6 transformer layers
for name, param in self.deberta.named_parameters():
if any(f"layer.{i}." in name for i in range(6)):
param.requires_grad = False
else:
param.requires_grad = True
hidden = self.deberta.config.hidden_size # 768 for deberta-v3-small
self.lstm_branch = BiLSTMBranch(hidden)
self.cnn_branch = CNNBranch(hidden)
self.trans_branch = TransformerBranch(hidden)
self.fusion = CrossAttentionFusion(dim=128)
self.classifier = nn.Sequential(
nn.LayerNorm(128),
nn.Linear(128, 128),
nn.GELU(),
nn.Dropout(0.4),
nn.Linear(128, 64),
nn.GELU(),
nn.Dropout(0.3),
nn.Linear(64, 1),
)
def forward(
self,
input_ids: torch.Tensor,
attention_mask: torch.Tensor,
token_type_ids: torch.Tensor,
) -> torch.Tensor:
out = self.deberta(
input_ids=input_ids,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
)
hidden = out.last_hidden_state # (B, T, 768)
lstm_out = self.lstm_branch(hidden, attention_mask)
cnn_out = self.cnn_branch(hidden)
trans_out = self.trans_branch(hidden, attention_mask)
fused = self.fusion(lstm_out, cnn_out, trans_out)
return self.classifier(fused) # (B, 1)
# ─── Convenience inference helper ────────────────────────────────────────────
def load_model(checkpoint_path: str, device: torch.device = None) -> HybridAITextDetector:
"""Load a trained HybridAITextDetector from a .pt checkpoint."""
if device is None:
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = HybridAITextDetector()
model.load_state_dict(torch.load(checkpoint_path, map_location=device))
model.eval().to(device)
return model
|