Spaces:
Paused
Paused
File size: 1,719 Bytes
b876688 |
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 |
import torch
from torch import nn
import torch.nn.functional as F
from attention import SelfAttention
class CLIPEmbedding(nn.Module):
def __init__(self, n_vocab: int, n_embd: int, n_token: int):
super().__init__()
self.token_embedding = nn.Embedding(n_vocab, n_embd)
self.position_embedding = nn.Parameter(torch.zeros((n_token, n_embd)))
def forward(self, tokens):
x = self.token_embedding(tokens)
x += self.position_embedding
return x
class CLIPLayer(nn.Module):
def __init__(self, n_head: int, n_embd: int):
super().__init__()
self.layernorm_1 = nn.LayerNorm(n_embd)
self.attention = SelfAttention(n_head, n_embd)
self.layernorm_2 = nn.LayerNorm(n_embd)
self.linear_1 = nn.Linear(n_embd, 4 * n_embd)
self.linear_2 = nn.Linear(4 * n_embd, n_embd)
self.activation = nn.GELU()
def forward(self, x):
residue = x
x = self.layernorm_1(x)
x = self.attention(x, causal_mask=True)
x += residue
residue = x
x = self.layernorm_2(x)
x = self.linear_1(x)
x = self.activation(x)
x = self.linear_2(x)
x += residue
return x
class CLIP(nn.Module):
def __init__(self):
super().__init__()
self.embedding = CLIPEmbedding(49408, 768, 77)
self.layers = nn.ModuleList([CLIPLayer(12, 768) for _ in range(12)])
self.layernorm = nn.LayerNorm(768)
def forward(self, tokens: torch.LongTensor) -> torch.FloatTensor:
state = self.embedding(tokens)
for layer in self.layers:
state = layer(state)
output = self.layernorm(state)
return output |