Spaces:
Running on Zero
Running on Zero
File size: 4,960 Bytes
98d1e65 9cde5e3 1717c0c 137f7ef 98d1e65 137f7ef 98d1e65 137f7ef 98d1e65 9cde5e3 98d1e65 9cde5e3 98d1e65 1717c0c 98d1e65 9cde5e3 98d1e65 9cde5e3 137f7ef 9cde5e3 98d1e65 9cde5e3 98d1e65 9cde5e3 98d1e65 9cde5e3 98d1e65 1717c0c 9cde5e3 1717c0c 9cde5e3 1717c0c 98d1e65 1717c0c 9cde5e3 1717c0c 98d1e65 1717c0c 98d1e65 1717c0c 98d1e65 1717c0c 98d1e65 1717c0c 98d1e65 1717c0c 98d1e65 9cde5e3 1717c0c | 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 | """Single Head transformer"""
import torch
import torch.nn as nn
from src.transformer.transformer import Transformer
from src.transformer.input_representation import InputRepresentation
class TinyTransformerLM(nn.Module):
def __init__(
self,
config,
vocab_size,
):
super().__init__()
print("Initializing TinyTransformerLM")
self.input_representation = InputRepresentation(
vocab_size=vocab_size,
context_length=config.context_length,
d_model=config.d_model,
)
self.transformer = Transformer(
d_model=config.d_model,
d_ff=config.d_ff,
context_length=config.context_length,
num_layers=config.num_layers,
number_of_heads=config.number_of_heads,
dropout=config.dropout,
#padding mask here
)
self.head = nn.Linear(
config.d_model,
vocab_size,
)
def forward(
self,
tokens: torch.Tensor,
) -> torch.Tensor:
x = self.input_representation(tokens)
x = self.transformer(x)
logits = self.head(x)
return logits
class TinyClassifier(nn.Module):
def __init__(
self,
config,
vocab_size,
num_classes=4,
):
super().__init__()
print("Initializing TinyClassifier")
self.input_representation = InputRepresentation(
vocab_size=vocab_size,
context_length=config.context_length,
d_model=config.d_model,
)
self.transformer = Transformer(
d_model=config.d_model,
d_ff=config.d_ff,
context_length=config.context_length,
num_layers=config.num_layers,
number_of_heads=config.number_of_heads,
dropout=config.dropout,
)
self.head = nn.Linear(
config.d_model,
num_classes,
)
def forward(
self,
tokens: torch.Tensor,
padding_mask: torch.Tensor | None = None,
) -> torch.Tensor:
x = self.input_representation(tokens)
x = self.transformer(
x,
padding_mask=padding_mask,
)
if padding_mask is None:
x = x.mean(dim=1)
else:
mask = padding_mask.unsqueeze(-1)
x = x * mask
token_count = mask.sum(
dim=1
).clamp(min=1)
x = x.sum(dim=1) / token_count
logits = self.head(x)
return logits
class TinyAnsweringModel(nn.Module):
def __init__(
self,
config,
vocab_size,
):
super().__init__()
print("Initializing TinyAnsweringModel")
self.input_representation = InputRepresentation(
vocab_size=vocab_size,
context_length=config.context_length,
d_model=config.d_model
)
self.transformer = Transformer(
d_model=config.d_model,
d_ff=config.d_ff,
context_length=config.context_length, #max_length used for truncating and padding input
num_layers=config.num_layers,
number_of_heads=config.number_of_heads,
dropout=config.dropout,
)
#TWO output projections
#Start of the answer
self.start_head = nn.Linear(
config.d_model, #IN features
1, #OUT features
)
# Predict where the answer ends.
self.end_head = nn.Linear(
config.d_model,
1, #why 1 ? should it not be max_length (or max_length)
)
def forward(
self,
tokens: torch.Tensor,
padding_mask: torch.Tensor,
) -> tuple[torch.Tensor, torch.Tensor]:
x = self.input_representation(tokens)
x = self.transformer(
x,
padding_mask=padding_mask,
)
start_logits = self.start_head(x).squeeze(-1)
end_logits = self.end_head(x).squeeze(-1)
return start_logits, end_logits
# Load language modelfrom checkpoint
def load_model(
checkpoint_path,
config,
vocab_size,
type="GENERATOR"
):
if (type == "GENERATOR"):
model = TinyTransformerLM(
config=config,
vocab_size=vocab_size,
)
elif (type == "CLASSIFIER"):
model = TinyClassifier(
config=config,
vocab_size=vocab_size,
)
elif (type == "ANSWERING"):
model = TinyAnsweringModel(
config=config,
vocab_size=vocab_size,
)
else:
print("UNKNOWN MODEL")
checkpoint = torch.load(
checkpoint_path,
map_location="cpu",
)
model.load_state_dict(checkpoint)
model.eval()
return model
|