Fix failing DeepSpeed model zoo tests (#30112)
* fix sequence length errors * fix label column name error for vit * fix the lm_head embedding!=linear layer mismatches for Seq2Seq models
This commit is contained in:
parent
2f12e40822
commit
4e3490f79b
|
@ -1932,7 +1932,10 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin, PushToHubMix
|
|||
# if word embeddings are not tied, make sure that lm head is resized as well
|
||||
if self.get_output_embeddings() is not None and not self.config.tie_word_embeddings:
|
||||
old_lm_head = self.get_output_embeddings()
|
||||
new_lm_head = self._get_resized_lm_head(old_lm_head, new_num_tokens)
|
||||
if isinstance(old_lm_head, torch.nn.Embedding):
|
||||
new_lm_head = self._get_resized_embeddings(old_lm_head, new_num_tokens)
|
||||
else:
|
||||
new_lm_head = self._get_resized_lm_head(old_lm_head, new_num_tokens)
|
||||
if hasattr(old_lm_head, "_hf_hook"):
|
||||
hook = old_lm_head._hf_hook
|
||||
add_hook_to_module(new_lm_head, hook)
|
||||
|
|
|
@ -236,6 +236,8 @@ def make_task_cmds():
|
|||
--train_file {data_dir_wmt}/train.json
|
||||
--source_lang en
|
||||
--target_lang ro
|
||||
--max_source_length 12
|
||||
--max_target_length 12
|
||||
""",
|
||||
"sum": f"""
|
||||
{scripts_dir}/summarization/run_summarization.py
|
||||
|
@ -269,6 +271,7 @@ def make_task_cmds():
|
|||
--remove_unused_columns False
|
||||
--max_steps 10
|
||||
--image_processor_name {DS_TESTS_DIRECTORY}/vit_feature_extractor.json
|
||||
--label_column_name labels
|
||||
""",
|
||||
}
|
||||
|
||||
|
|
Loading…
Reference in New Issue