Spaces:
Running on Zero
Running on Zero
File size: 3,122 Bytes
834d1cf | 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 | import torch
from torch import nn
from typing import Optional
class QwenImageMAE(nn.Module):
def __init__(
self,
):
super().__init__()
from transformers import ViTMAEConfig, ViTMAEModel
config=ViTMAEConfig(**{
"architectures": [
"ViTMAEForPreTraining"
],
"attention_probs_dropout_prob": 0.0,
"decoder_hidden_size": 512,
"decoder_intermediate_size": 2048,
"decoder_num_attention_heads": 16,
"decoder_num_hidden_layers": 8,
"hidden_act": "gelu",
"hidden_dropout_prob": 0.0,
"hidden_size": 1024,
"image_size": 224,
"initializer_range": 0.02,
"intermediate_size": 4096,
"layer_norm_eps": 1e-12,
"mask_ratio": 0.0,
"model_type": "vit_mae",
"norm_pix_loss": False,
"num_attention_heads": 16,
"num_channels": 3,
"num_hidden_layers": 24,
"patch_size": 16,
"qkv_bias": True,
"torch_dtype": "bfloat16",
"attn_implementation": "sdpa"
}
)
self.model = ViTMAEModel(config)
self.config=config
def forward(self, pixel_values):
outputs=self.model(pixel_values,interpolate_pos_encoding=True)
return outputs.last_hidden_state
def new_forward(self, pixel_values,sample1_shapes):
outputs=self.model.new_forward(pixel_values,sample1_shapes,interpolate_pos_encoding=True)
return outputs.last_hidden_state
|