Unverified Commit 06886d5a authored by Sylvain Gugger's avatar Sylvain Gugger Committed by GitHub
Browse files

Only resize embeddings when necessary (#20043)

* Only resize embeddings when necessary

* Add comment
parent 9080607b
...@@ -387,6 +387,10 @@ def main(): ...@@ -387,6 +387,10 @@ def main():
n_params = sum(dict((p.data_ptr(), p.numel()) for p in model.parameters()).values()) n_params = sum(dict((p.data_ptr(), p.numel()) for p in model.parameters()).values())
logger.info(f"Training new model from scratch - Total size={n_params/2**20:.2f}M params") logger.info(f"Training new model from scratch - Total size={n_params/2**20:.2f}M params")
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets. # Preprocessing the datasets.
......
...@@ -378,6 +378,10 @@ def main(): ...@@ -378,6 +378,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForCausalLM.from_config(config) model = AutoModelForCausalLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets. # Preprocessing the datasets.
......
...@@ -389,6 +389,10 @@ def main(): ...@@ -389,6 +389,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForMaskedLM.from_config(config) model = AutoModelForMaskedLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets. # Preprocessing the datasets.
......
...@@ -383,6 +383,10 @@ def main(): ...@@ -383,6 +383,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForMaskedLM.from_config(config) model = AutoModelForMaskedLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets. # Preprocessing the datasets.
......
...@@ -376,6 +376,10 @@ def main(): ...@@ -376,6 +376,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = XLNetLMHeadModel(config) model = XLNetLMHeadModel(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets. # Preprocessing the datasets.
......
...@@ -398,6 +398,10 @@ def main(): ...@@ -398,6 +398,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForMultipleChoice.from_config(config) model = AutoModelForMultipleChoice.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Preprocessing the datasets. # Preprocessing the datasets.
......
...@@ -380,6 +380,10 @@ def main(): ...@@ -380,6 +380,10 @@ def main():
use_auth_token=True if model_args.use_auth_token else None, use_auth_token=True if model_args.use_auth_token else None,
) )
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
if model.config.decoder_start_token_id is None: if model.config.decoder_start_token_id is None:
......
...@@ -422,6 +422,10 @@ def main(): ...@@ -422,6 +422,10 @@ def main():
use_auth_token=True if model_args.use_auth_token else None, use_auth_token=True if model_args.use_auth_token else None,
) )
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
if model.config.decoder_start_token_id is None and isinstance(tokenizer, (MBartTokenizer, MBartTokenizerFast)): if model.config.decoder_start_token_id is None and isinstance(tokenizer, (MBartTokenizer, MBartTokenizerFast)):
......
...@@ -439,6 +439,10 @@ def main(): ...@@ -439,6 +439,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForSeq2SeqLM.from_config(config) model = AutoModelForSeq2SeqLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
if model.config.decoder_start_token_id is None: if model.config.decoder_start_token_id is None:
raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined") raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined")
......
...@@ -414,6 +414,12 @@ def main(): ...@@ -414,6 +414,12 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForTokenClassification.from_config(config) model = AutoModelForTokenClassification.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Model has labels -> use them. # Model has labels -> use them.
......
...@@ -380,6 +380,10 @@ def main(): ...@@ -380,6 +380,10 @@ def main():
use_auth_token=True if model_args.use_auth_token else None, use_auth_token=True if model_args.use_auth_token else None,
) )
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Set decoder_start_token_id # Set decoder_start_token_id
......
...@@ -411,6 +411,10 @@ def main(): ...@@ -411,6 +411,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = AutoModelForSeq2SeqLM.from_config(config) model = AutoModelForSeq2SeqLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# Set decoder_start_token_id # Set decoder_start_token_id
......
...@@ -473,6 +473,10 @@ def main(): ...@@ -473,6 +473,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = TFAutoModelForCausalLM.from_config(config) model = TFAutoModelForCausalLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# endregion # endregion
......
...@@ -489,6 +489,10 @@ def main(): ...@@ -489,6 +489,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = TFAutoModelForMaskedLM.from_config(config) model = TFAutoModelForMaskedLM.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# endregion # endregion
......
...@@ -516,6 +516,10 @@ def main(): ...@@ -516,6 +516,10 @@ def main():
use_auth_token=True if model_args.use_auth_token else None, use_auth_token=True if model_args.use_auth_token else None,
) )
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# endregion # endregion
......
...@@ -385,6 +385,10 @@ def main(): ...@@ -385,6 +385,10 @@ def main():
logger.info("Training new model from scratch") logger.info("Training new model from scratch")
model = TFAutoModelForTokenClassification.from_config(config) model = TFAutoModelForTokenClassification.from_config(config)
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
# endregion # endregion
......
...@@ -469,6 +469,10 @@ def main(): ...@@ -469,6 +469,10 @@ def main():
use_auth_token=True if model_args.use_auth_token else None, use_auth_token=True if model_args.use_auth_token else None,
) )
# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
# on a small vocab and want a smaller embedding size, remove this test.
embedding_size = model.get_input_embeddings().weight.shape[0]
if len(tokenizer) > embedding_size:
model.resize_token_embeddings(len(tokenizer)) model.resize_token_embeddings(len(tokenizer))
if isinstance(tokenizer, tuple(MULTILINGUAL_TOKENIZERS)): if isinstance(tokenizer, tuple(MULTILINGUAL_TOKENIZERS)):
model.config.forced_bos_token_id = forced_bos_token_id model.config.forced_bos_token_id = forced_bos_token_id
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment