|
|
|
from transformers import PretrainedConfig |
|
|
|
|
|
class DownsamplerConfig(PretrainedConfig): |
|
model_type = 'downsampler' |
|
_auto_class = 'AutoConfig' |
|
|
|
def __init__( |
|
self, |
|
kernel_size=1, |
|
stride=1, |
|
visual_hidden_size=4096, |
|
llm_hidden_size=4096, |
|
depth=2, |
|
hidden_act='gelu', |
|
bias=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.kernel_size = kernel_size |
|
self.stride = stride |
|
super().__init__(**kwargs) |
|
|
