SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[modular] Fix the prefix-based renaming if the old and new model share a common name suffix (#37829)

Browse Source 
* first try

* Fix and set examples

* style

* fix

* Update modular_test_detr.py

* Update image_processing_new_imgproc_model.py

* Update modular_model_converter.py
This commit is contained in:
Cyril Vallez
2025-04-29 10:43:23 +02:00
committed by GitHub
parent a847d4aa6b
commit 4602059aae
16 changed files with 2340 additions and 939 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
examples/modular-transformers/modeling_dummy_bert.py
View File

@@ -6,7 +6,7 @@
# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
import math
import os
from typing import Optional, Union
from typing import Optional, Tuple, Union
import torch
from packaging import version
@@ -136,9 +136,9 @@ class DummyBertSelfAttention(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> tuple[torch.Tensor]:
) -> Tuple[torch.Tensor]:
mixed_query_layer = self.query(hidden_states)
# If this is instantiated as a cross-attention module, the keys
@@ -245,9 +245,9 @@ class DummyBertSdpaSelfAttention(DummyBertSelfAttention):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> tuple[torch.Tensor]:
) -> Tuple[torch.Tensor]:
if self.position_embedding_type != "absolute" or output_attentions or head_mask is not None:
# TODO: Improve this warning with e.g. `model.config._attn_implementation = "manual"` once implemented.
logger.warning_once(
@@ -386,9 +386,9 @@ class DummyBertAttention(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> tuple[torch.Tensor]:
) -> Tuple[torch.Tensor]:
self_outputs = self.self(
hidden_states,
attention_mask,
@@ -454,9 +454,9 @@ class DummyBertLayer(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_value: Optional[tuple[tuple[torch.FloatTensor]]] = None,
past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
output_attentions: Optional[bool] = False,
) -> tuple[torch.Tensor]:
) -> Tuple[torch.Tensor]:
# decoder uni-directional self-attention cached key/values tuple is at positions 1,2
self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
self_attention_outputs = self.attention(
@@ -532,12 +532,12 @@ class DummyBertEncoder(nn.Module):
head_mask: Optional[torch.FloatTensor] = None,
encoder_hidden_states: Optional[torch.FloatTensor] = None,
encoder_attention_mask: Optional[torch.FloatTensor] = None,
past_key_values: Optional[tuple[tuple[torch.FloatTensor]]] = None,
past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = False,
output_hidden_states: Optional[bool] = False,
return_dict: Optional[bool] = True,
) -> Union[tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
all_hidden_states = () if output_hidden_states else None
all_self_attentions = () if output_attentions else None
all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
@@ -626,6 +626,46 @@ class DummyBertPooler(nn.Module):
return pooled_output
class DummyBertPredictionHeadTransform(nn.Module):
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
if isinstance(config.hidden_act, str):
self.transform_act_fn = ACT2FN[config.hidden_act]
else:
self.transform_act_fn = config.hidden_act
self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
hidden_states = self.dense(hidden_states)
hidden_states = self.transform_act_fn(hidden_states)
hidden_states = self.LayerNorm(hidden_states)
return hidden_states
class DummyBertLMPredictionHead(nn.Module):
def __init__(self, config):
super().__init__()
self.transform = DummyBertPredictionHeadTransform(config)
# The output weights are the same as the input embeddings, but there is
# an output-only bias for each token.
self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
self.bias = nn.Parameter(torch.zeros(config.vocab_size))
# Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
self.decoder.bias = self.bias
def _tie_weights(self):
self.decoder.bias = self.bias
def forward(self, hidden_states):
hidden_states = self.transform(hidden_states)
hidden_states = self.decoder(hidden_states)
return hidden_states
def load_tf_weights_in_dummy_bert(model, config, tf_checkpoint_path):
"""Load tf checkpoints in a pytorch model."""
try:
@@ -726,6 +766,8 @@ class DummyBertPreTrainedModel(PreTrainedModel):
elif isinstance(module, nn.LayerNorm):
module.bias.data.zero_()
module.weight.data.fill_(1.0)
elif isinstance(module, DummyBertLMPredictionHead):
module.bias.data.zero_()
DUMMY_BERT_START_DOCSTRING = r"""