File size: 5,715 Bytes
19e72b3 | 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 | import torch
from torch import nn
from argparse import Namespace
import xformers.ops as xops
from transformers.activations import ACT2FN
class PatchEmbedding(nn.Module):
def __init__(self, config):
super().__init__()
self.proj = nn.Conv2d(config.in_channels, config.hidden_size, kernel_size=config.patch_size, stride=config.patch_size)
self.cls_embedding = nn.Parameter(torch.zeros(1, config.hidden_size))
self.position_embedding = nn.Embedding(config.num_positions, config.hidden_size)
def forward(self, images: "tensor(B, C, H, W)") -> "tensor(B, L, D)":
x = self.proj(images)
x = x.flatten(2).transpose(1, 2)
cls_token = self.cls_embedding.expand(x.shape[0], -1, -1)
x = torch.cat((cls_token, x), dim=1)
x += self.position_embedding.weight.unsqueeze(0)
return x
class Attention(nn.Module):
def __init__(self, config):
super().__init__()
self.num_heads = config.num_heads
head_dim = config.hidden_size // config.num_heads
self.scale = head_dim ** -0.5
self.query_key_value = nn.Linear(config.hidden_size, config.hidden_size * 3)
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.output_dropout = torch.nn.Dropout(config.dropout_prob)
def forward(self, x: "tensor(B, L, D)") -> "tensor(B, L, D)":
B, L, _ = x.shape
qkv = self.query_key_value(x)
qkv = qkv.reshape(B, L, 3, self.num_heads, -1).permute(2, 0, 1, 3, 4) # 3, B, L, H, D
q, k, v = qkv[0], qkv[1], qkv[2]
out = xops.memory_efficient_attention(
q, k, v, scale=self.scale,
)
output = self.dense(out.view(B, L, -1))
output = self.output_dropout(output)
return output
def attention(self, q, k, v):
attn_weights = torch.matmul(q * self.scale, k.transpose(-2, -1))
attn_weights = attn_weights.softmax(dim=-1)
output = torch.matmul(attn_weights, v)
return output
class MLP(nn.Module):
def __init__(self, config):
super().__init__()
self.config = config
self.activation_fn = ACT2FN[config.hidden_act]
self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
def forward(self, x: torch.Tensor) -> torch.Tensor:
x = self.fc1(x)
x = self.activation_fn(x)
x = self.fc2(x)
return x
class TransformerLayer(nn.Module):
def __init__(self, config):
super().__init__()
self.input_layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
self.attention = Attention(config)
self.mlp = MLP(config)
self.post_attention_layernorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
def forward(self, hidden_states):
attention_input = hidden_states
attention_output = self.input_layernorm(self.attention(attention_input))
hidden_states = attention_input + attention_output
mlp_input = hidden_states
mlp_output = self.post_attention_layernorm(self.mlp(mlp_input))
output = mlp_input + mlp_output
return output
class Transformer(nn.Module):
def __init__(self, config):
super().__init__()
self.layers = nn.ModuleList([TransformerLayer(config) for _ in range(config.num_hidden_layers)])
def forward(self, hidden_states):
for layer_module in self.layers:
hidden_states = layer_module(hidden_states)
return hidden_states
class GLU(nn.Module):
def __init__(self, config, in_features):
super().__init__()
self.linear_proj = nn.Linear(in_features, config.hidden_size, bias=False)
self.norm1 = nn.LayerNorm(config.hidden_size)
self.act1 = nn.GELU()
self.act2 = nn.functional.silu
self.dense_h_to_4h = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
self.dense_4h_to_h = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
def forward(self, x):
x = self.linear_proj(x)
x = self.act1(self.norm1(x))
x = self.act2(self.gate_proj(x)) * self.dense_h_to_4h(x)
x = self.dense_4h_to_h(x)
return x
class EVA2CLIPModel(nn.Module):
def __init__(self, config):
super().__init__()
vision_config = Namespace(**config.vision_config)
self.patch_embedding = PatchEmbedding(vision_config)
self.transformer = Transformer(vision_config)
self.linear_proj = GLU(config, in_features=vision_config.hidden_size)
self.conv = nn.Conv2d(in_channels=vision_config.hidden_size, out_channels=vision_config.hidden_size, kernel_size=2, stride=2)
self.boi = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
self.eoi = nn.Parameter(torch.zeros(1, 1, config.hidden_size))
def forward(self, images: "tensor(B, C, H, W)") -> "tensor(B, L, D)":
x = self.patch_embedding(images)
x = self.transformer(x)
x = x[:, 1:]
b, s, h = x.shape
grid_size = int(s**0.5)
x = x.view(b, grid_size, grid_size, h).permute(0, 3, 1, 2)
x = self.conv(x)
x = x.flatten(2).transpose(1, 2)
x = self.linear_proj(x)
boi = self.boi.expand(x.shape[0], -1, -1)
eoi = self.eoi.expand(x.shape[0], -1, -1)
x = torch.cat((boi, x, eoi), dim=1)
return x
|