mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-31 02:02:21 +06:00
move the fix a bit
This commit is contained in:
parent
00afce9837
commit
3ac6c52f34
@ -1103,10 +1103,6 @@ class Emu3PreTrainedModel(PreTrainedModel):
|
||||
_supports_param_buffer_assignment = False
|
||||
_supports_flex_attn = True
|
||||
_supports_attention_backend = True
|
||||
_can_record_outputs: dict[str, tuple[nn.Module, int]] = {
|
||||
"hidden_states": (Emu3DecoderLayer, 0),
|
||||
"attentions": (Emu3Attention, 1),
|
||||
}
|
||||
|
||||
def _init_weights(self, module):
|
||||
std = self.config.get_text_config().initializer_range
|
||||
@ -1158,6 +1154,11 @@ class Emu3RotaryEmbedding(nn.Module):
|
||||
|
||||
@auto_docstring
|
||||
class Emu3TextModel(Emu3PreTrainedModel):
|
||||
_can_record_outputs: dict[str, tuple[nn.Module, int]] = {
|
||||
"hidden_states": (Emu3DecoderLayer, 0),
|
||||
"attentions": (Emu3Attention, 1),
|
||||
}
|
||||
|
||||
def __init__(self, config: Emu3Config):
|
||||
super().__init__(config)
|
||||
self.padding_idx = config.pad_token_id
|
||||
|
@ -846,10 +846,6 @@ class Emu3PreTrainedModel(ChameleonPreTrainedModel, Emu3VQVAE):
|
||||
]
|
||||
_supports_flex_attn = True
|
||||
_supports_attention_backend = True
|
||||
_can_record_outputs: dict[str, tuple[nn.Module, int]] = {
|
||||
"hidden_states": (Emu3DecoderLayer, 0),
|
||||
"attentions": (Emu3Attention, 1),
|
||||
}
|
||||
|
||||
def _init_weights(self, module):
|
||||
std = self.config.get_text_config().initializer_range
|
||||
@ -866,6 +862,11 @@ class Emu3PreTrainedModel(ChameleonPreTrainedModel, Emu3VQVAE):
|
||||
|
||||
|
||||
class Emu3TextModel(LlamaModel, Emu3PreTrainedModel):
|
||||
_can_record_outputs: dict[str, tuple[nn.Module, int]] = {
|
||||
"hidden_states": (Emu3DecoderLayer, 0),
|
||||
"attentions": (Emu3Attention, 1),
|
||||
}
|
||||
|
||||
def __init__(self, config: Emu3Config):
|
||||
super().__init__(config)
|
||||
self.layers = nn.ModuleList(
|
||||
|
Loading…
Reference in New Issue
Block a user