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

[Modeling] Fix encoder CPU offloading for whisper (#38994)

Browse Source 
* fix cpu offloading for whisper

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>

* unskip offloading tests

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>

* revert small change

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>

* remove tests

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>

---------

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
This commit is contained in:
Kyle Sayers
2025-06-26 11:56:33 -04:00
committed by GitHub
parent c63cfd6a83
commit 0a8081b03d
2 changed files with 2 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/transformers/models/whisper/modeling_whisper.py
View File

@@ -687,9 +687,9 @@ class WhisperEncoder(WhisperPreTrainedModel):
inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds)) inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))
inputs_embeds = inputs_embeds.permute(0, 2, 1) inputs_embeds = inputs_embeds.permute(0, 2, 1)
embed_pos = self.embed_positions.weight all_positions = torch.arange(self.embed_positions.num_embeddings, device=inputs_embeds.device)
hidden_states = inputs_embeds + embed_pos hidden_states = inputs_embeds + self.embed_positions(all_positions)
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training) hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
encoder_states = () if output_hidden_states else None encoder_states = () if output_hidden_states else None

16
tests/models/whisper/test_modeling_whisper.py
View File

@@ -3356,22 +3356,6 @@ class WhisperEncoderModelTest(ModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs() config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model_forward(*config_and_inputs, use_weighted_layer_sum=True) self.model_tester.create_and_check_model_forward(*config_and_inputs, use_weighted_layer_sum=True)
@unittest.skip(reason="Some undefined behavior encountered with tiny versions of this model. Skip for now.")
def test_cpu_offload(self):
pass
@unittest.skip(reason="Some undefined behavior encountered with tiny versions of this model. Skip for now.")
def test_disk_offload_bin(self):
pass
@unittest.skip(reason="Some undefined behavior encountered with tiny versions of this model. Skip for now.")
def test_disk_offload_safetensors(self):
pass
@unittest.skip(reason="Some undefined behavior encountered with tiny versions of this model. Skip for now.")
def test_model_parallelism(self):
pass
@unittest.skip(reason="Not applicable for an encoder-only acoustic model") @unittest.skip(reason="Not applicable for an encoder-only acoustic model")
def test_inputs_embeds(self): def test_inputs_embeds(self):
# input embeds is meaningless for an encoder-only acoustic model # input embeds is meaningless for an encoder-only acoustic model