File size: 1,590 Bytes
5d2c747 |
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 |
"""
Simple, flexible core models.
"""
import torch
from models.components.layers.transformer_blocks import GenericTransformerBlock
class GenericTransformer(torch.nn.Module):
"""
Generic Transformer Class intended to be used for as
broad a range of transformer models as possible.
"""
def __init__(self, model_cfg):
super().__init__()
# build the transformer
self.transformer = torch.nn.ModuleDict(
{
"drop": torch.nn.Dropout(),
"h": torch.nn.ModuleList(
[
GenericTransformerBlock(
hidden_dim=model_cfg["hidden_dim"],
context_window=model_cfg["context_window"],
use_rope=model_cfg["positional_encoding_type"] == "rope",
ffn_cfg=model_cfg["core_model"]["ffn"],
attn_cfg=model_cfg["core_model"]["attn"],
)
for _ in range(model_cfg["core_model"]["num_layers"])
]
),
}
)
def forward(self, x, attention_mask=None):
"""
Pass an input through the model
Args:
x: torch.tensor(B, S, H)
Returns:
x: torch.tensor(B, S, H)
"""
# apply dropout
x = self.transformer.drop(x)
# pass through the transformer blocks
for block in self.transformer.h:
x = block(x, attention_mask=attention_mask)
return x
|