File size: 2,642 Bytes
db6ee6a |
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 |
import torch
import torch.nn as nn
import re
from transformers import BertConfig, BertModel
class IdentityMap(nn.Module):
def __init__(self):
super().__init__()
def forward(self, x, *args, **kwargs):
return x
@property
def config(self):
return {"mm_projector_type": 'identity'}
class SimpleResBlock(nn.Module):
def __init__(self, channels):
super().__init__()
self.pre_norm = nn.LayerNorm(channels)
self.proj = nn.Sequential(
nn.Linear(channels, channels),
nn.GELU(),
nn.Linear(channels, channels)
)
def forward(self, x):
x = self.pre_norm(x)
return x + self.proj(x)
def build_vision_projector(config, delay_load=False, **kwargs):
projector_type = getattr(config, 'mm_projector_type', 'linear')
if projector_type == 'linear':
return nn.Linear(config.mm_hidden_size, config.hidden_size)
mlp_gelu_match = re.match(r'^mlp(\d+)x_gelu$', projector_type)
if mlp_gelu_match:
mlp_depth = int(mlp_gelu_match.group(1))
modules = [nn.Linear(config.mm_hidden_size, config.hidden_size)]
for _ in range(1, mlp_depth):
modules.append(nn.GELU())
modules.append(nn.Linear(config.hidden_size, config.hidden_size))
return nn.Sequential(*modules)
if projector_type == 'identity':
return IdentityMap()
raise ValueError(f'Unknown projector type: {projector_type}')
class ImageEmbeddingPooler(nn.Module):
def __init__(self):
super().__init__()
self.embedding_dim = 512
# Configure a new BERT model with 2 hidden layers and without positional embeddings
config = BertConfig(
hidden_size=self.embedding_dim,
num_hidden_layers=2, # Set the number of hidden layers to 2
num_attention_heads=8,
intermediate_size=self.embedding_dim*4,
use_position_embeddings=True,
max_position_embeddings=2304,
use_bfloat16=True,
vocab_size=1,
)
self.bert = BertModel(config)
def forward(self, embeddings, attention_mask):
# embeddings shape: (batch_size, num_images, embedding_dim)
batch_size, num_tokens, _ = embeddings.shape
# Process embeddings through BERT without positional IDs
outputs = self.bert(inputs_embeds=embeddings, attention_mask=attention_mask)
# identity option
outputs = outputs['last_hidden_state'].to(embeddings.dtype)[:, :196]
return outputs
def build_image_pooler(config):
return ImageEmbeddingPooler()
|