File size: 2,255 Bytes
141b176 |
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 |
"""
Custom Student Model for Knowledge Distillation
"""
import torch
import torch.nn as nn
from transformers import PreTrainedModel, PretrainedConfig
from typing import Dict, Any, List, Optional
class StudentModelConfig(PretrainedConfig):
model_type = "distilled_student"
def __init__(
self,
hidden_size=768,
num_layers=12,
num_attention_heads=12,
intermediate_size=3072,
vocab_size=30522,
max_position_embeddings=512,
modalities=["text"],
**kwargs
):
super().__init__(**kwargs)
self.hidden_size = hidden_size
self.num_layers = num_layers
self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size
self.vocab_size = vocab_size
self.max_position_embeddings = max_position_embeddings
self.modalities = modalities
class StudentModel(PreTrainedModel):
config_class = StudentModelConfig
def __init__(self, config):
super().__init__(config)
self.config = config
self.hidden_size = config.hidden_size
self.num_layers = config.num_layers
self.modalities = config.modalities
# Build model layers based on config
self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size)
self.layers = nn.ModuleList([
nn.TransformerEncoderLayer(
d_model=config.hidden_size,
nhead=config.num_attention_heads,
dim_feedforward=config.intermediate_size,
batch_first=True
) for _ in range(config.num_layers)
])
self.pooler = nn.Linear(config.hidden_size, config.hidden_size)
def forward(self, input_ids=None, attention_mask=None, **kwargs):
if input_ids is not None:
embeddings = self.embeddings(input_ids)
else:
# Handle other modalities
embeddings = kwargs.get('inputs_embeds')
for layer in self.layers:
embeddings = layer(embeddings, src_key_padding_mask=attention_mask)
pooled = self.pooler(embeddings.mean(dim=1))
return {
'last_hidden_state': embeddings,
'pooler_output': pooled
}
|