File size: 1,193 Bytes
032e687 | 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 | # Copyright (c) OpenMMLab. All rights reserved.
from transformers import PretrainedConfig
class ProjectorConfig_OMG_LLaVA(PretrainedConfig):
model_type = 'projector'
_auto_class = 'AutoConfig'
def __init__(
self,
visual_hidden_size=4096,
llm_hidden_size=4096,
depth=2,
hidden_act='gelu',
bias=True,
query_channels=256,
feat_channels=1536,
pixel_shuffle_ratio=None,
additional_bg_tokens=10,
visual_prompt_proj=False,
add_cross_attn_layer=False,
**kwargs,
):
self.visual_hidden_size = visual_hidden_size
self.llm_hidden_size = llm_hidden_size
self.depth = depth
self.hidden_act = hidden_act
self.bias = bias
self.query_channels=query_channels
self.feat_channels=feat_channels
if pixel_shuffle_ratio is not None:
self.feat_channels = self.feat_channels * pixel_shuffle_ratio * pixel_shuffle_ratio
self.additional_bg_tokens = additional_bg_tokens
self.visual_prompt_proj = visual_prompt_proj
self.add_cross_attn_layer = add_cross_attn_layer
super().__init__(**kwargs) |