Implemented unidirectional attention, moving on
Browse files
model.py
CHANGED
@@ -3,46 +3,97 @@ import torch.nn as nn
|
|
3 |
import torch.functional as F
|
4 |
import torch.optim as optim
|
5 |
import wandb
|
6 |
-
import fancy_einsum
|
7 |
from einops import rearrange, repeat, reduce
|
|
|
8 |
|
9 |
|
10 |
class OsSoluModel(nn.Module):
|
11 |
-
def __init__(self, config) -> None:
|
12 |
super().__init__()
|
|
|
13 |
self.config = config
|
|
|
|
|
14 |
self.transformer_block = TransformerBlock(config)
|
|
|
|
|
15 |
|
16 |
def forward(self, x: t.Tensor) -> t.Tensor:
|
17 |
-
|
|
|
|
|
18 |
|
19 |
|
20 |
class TransformerBlock(nn.Module):
|
21 |
-
def __init__(self, config) -> None:
|
22 |
-
super().__init__()
|
23 |
self.config = config
|
24 |
|
25 |
-
self.
|
26 |
self.linear = nn.Sequential(
|
27 |
nn.Linear(config.d_model, config.d_model),
|
28 |
SoLU(),
|
29 |
)
|
30 |
-
self.layer_norm = nn.LayerNorm(normalized_shape)
|
31 |
self.unembed = nn.Embedding(config.num_embeddings, config.d_model)
|
32 |
|
33 |
def forward(self, x: t.Tensor) -> t.Tensor:
|
34 |
pass
|
35 |
|
36 |
|
37 |
-
class
|
38 |
-
def __init__(self, config) -> None:
|
39 |
super().__init__()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
-
def
|
42 |
-
|
43 |
-
# Apply attention mask
|
44 |
-
# Compute softmax
|
45 |
-
# Apply final einsum
|
46 |
-
# Return attention output
|
47 |
|
48 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
import torch.functional as F
|
4 |
import torch.optim as optim
|
5 |
import wandb
|
6 |
+
import fancy_einsum as einsum
|
7 |
from einops import rearrange, repeat, reduce
|
8 |
+
from utils import OsSoluConfig
|
9 |
|
10 |
|
11 |
class OsSoluModel(nn.Module):
|
12 |
+
def __init__(self, config: OsSoluConfig) -> None:
|
13 |
super().__init__()
|
14 |
+
normalised_shape = None # TODO: normalised_shape should be defined properly
|
15 |
self.config = config
|
16 |
+
self.embed_positions = nn.Embedding(config.max_positional_embeddings, config.d_model)
|
17 |
+
self.embed_tokens = nn.Embedding(config.vocab_size, config.d_model)
|
18 |
self.transformer_block = TransformerBlock(config)
|
19 |
+
self.final_ln = nn.LayerNorm(normalized_shape, config.ln_eps)
|
20 |
+
self.unembed = nn
|
21 |
|
22 |
def forward(self, x: t.Tensor) -> t.Tensor:
|
23 |
+
positional_embeddings = self.embed_positions(t.arange(x.size(1)))
|
24 |
+
token_embeddings = self.embed_tokens(x)
|
25 |
+
embeddings = positional_embeddings + token_embeddings
|
26 |
|
27 |
|
28 |
class TransformerBlock(nn.Module):
|
29 |
+
def __init__(self, config: OsSoluConfig) -> None:
|
30 |
+
super().__init__()
|
31 |
self.config = config
|
32 |
|
33 |
+
self.attention = UnidirectionalAttention(config) if config.self_attention_type == "unidirectional" else RotaryAttention(config)
|
34 |
self.linear = nn.Sequential(
|
35 |
nn.Linear(config.d_model, config.d_model),
|
36 |
SoLU(),
|
37 |
)
|
38 |
+
self.layer_norm = nn.LayerNorm(normalized_shape, config.ln_eps)
|
39 |
self.unembed = nn.Embedding(config.num_embeddings, config.d_model)
|
40 |
|
41 |
def forward(self, x: t.Tensor) -> t.Tensor:
|
42 |
pass
|
43 |
|
44 |
|
45 |
+
class UnidirectionalAttention(nn.Module):
|
46 |
+
def __init__(self, config: OsSoluConfig) -> None:
|
47 |
super().__init__()
|
48 |
+
self.num_heads = config.num_heads
|
49 |
+
self.d_model = config.d_model
|
50 |
+
self.project_q = nn.Linear(config.num_embeddings, config.d_model)
|
51 |
+
self.project_k = nn.Linear(config.num_embeddings, config.d_model)
|
52 |
+
self.project_v = nn.Linear(config.num_embeddings, config.d_model)
|
53 |
+
self.project_out = nn.Linear(config.d_model, config.d_model)
|
54 |
+
self.LARGE_NEGATIVE_VALUE = -1e5
|
55 |
|
56 |
+
def hidden_to_heads(self, tensor: t.Tensor) -> t.Tensor:
|
57 |
+
return rearrange(tensor, "b s (nh hs) -> b nh s hs", nh=self.num_heads)
|
|
|
|
|
|
|
|
|
58 |
|
59 |
+
def compute_pre_softmax_attn_pattern(self, x: t.Tensor) -> t.Tensor:
|
60 |
+
Q = self.project_q(x)
|
61 |
+
K = self.project_k(x)
|
62 |
+
|
63 |
+
Q = self.hidden_to_heads(Q)
|
64 |
+
K = self.hidden_to_heads(K)
|
65 |
+
attention_pattern = einsum("batch num_heads seqlen_q head_size, batch num_heads seqlen_k head_size -> batch num_heads seqlen_q seqlen_k")
|
66 |
+
|
67 |
+
return attention_pattern
|
68 |
+
|
69 |
+
def forward(self, x: t.Tensor) -> t.Tensor:
|
70 |
+
batch, seqlen, hidden_size = x.shape
|
71 |
+
attention_pattern = self.compute_pre_softmax_attn_pattern(x)
|
72 |
+
V = self.project_v(x)
|
73 |
+
|
74 |
+
# Masking attention. Since GPT is unidirectional, it should only attend to previous tokens.
|
75 |
+
if seqlen > 1:
|
76 |
+
fst_range = t.arange(seqlen, device=self.device).unsqueeze(0).T
|
77 |
+
snd_range = t.arange(seqlen, device=self.device).unsqueeze(0)
|
78 |
+
bool_array = fst_range < snd_range
|
79 |
+
attention_score[..., bool_array] = self.LARGE_NEGATIVE_VALUE
|
80 |
+
|
81 |
+
|
82 |
+
attention_pattern = attention_pattern / t.sqrt(t.tensor(self.d_model // self.num_heads))
|
83 |
+
attention_score = attention_pattern.softmax(dim=-1)
|
84 |
+
|
85 |
+
V = self.hidden_to_heads(V)
|
86 |
+
out = einsum("batch num_heads seqlen_q seqlen_k, batch num_heads seqlen_k head_size -> batch num_heads seqlen_q head_size", attention_score, V)
|
87 |
+
out = rearrange("b nh s hs -> b s (nh hs)")
|
88 |
+
out = self.project_out(out)
|
89 |
+
|
90 |
+
|
91 |
+
return out
|
92 |
+
|
93 |
+
class RotaryAttention(nn.Module):
|
94 |
+
def __init__(self, config: OsSoluConfig) -> None:
|
95 |
+
super().__init__()
|
96 |
+
self.config = config
|
97 |
+
|
98 |
+
def forward(self, x: t.Tensor) -> t.Tensor:
|
99 |
+
pass
|
utils.py
CHANGED
@@ -1,10 +1,12 @@
|
|
1 |
@dataclass
|
2 |
class OsSoluConfig:
|
3 |
-
d_model: int = 512
|
4 |
-
vocab_size: int = 65536
|
5 |
-
learning_rate: float = 1e-3
|
6 |
-
num_embeddings: int = 1024
|
7 |
-
num_blocks: int = 1
|
8 |
-
dropout: float = 0.1
|
9 |
-
ln_eps: float = 1e-3
|
10 |
-
num_heads: int = 4
|
|
|
|
|
|
1 |
@dataclass
|
2 |
class OsSoluConfig:
|
3 |
+
d_model: int = 512 # Hidden size of the model.
|
4 |
+
vocab_size: int = 65536 # Vocabulary size of the input sequence. Unsure about this.
|
5 |
+
learning_rate: float = 1e-3 # Learning rate for the optimiser.
|
6 |
+
num_embeddings: int = 1024 # Number of embeddings. Unsure about this.
|
7 |
+
num_blocks: int = 1 # Number of transformer blocks.
|
8 |
+
dropout: float = 0.1 # Probability of dropout.
|
9 |
+
ln_eps: float = 1e-3 # Layer norm epsilon.
|
10 |
+
num_heads: int = 4 # Number of attention heads in each attention layer.
|
11 |
+
self_attention_type: str = "unidirectional" # What type of attention to use: rotary or unidirectional.
|
12 |
+
max_positional_embeddings: int = 1024 # Maximum number of positional embeddings.
|