dnrops
/
transformers
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Refactor default chat template warnings (#30551) 

* Temporarily silence warnings in apply_chat_template until we can properly deprecate default chat templates

* make fixup

* Move the default chat template warning into apply_chat_template itself

* make fixup
This commit is contained in:
Matt 2024-05-01 08:42:11 +01:00 committed by GitHub
parent 4bc9cb36b7
commit 4b4da18f53
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
20 changed files with 30 additions and 135 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/transformers/models/blenderbot/tokenization_blenderbot.py
View File

@ -411,13 +411,6 @@ class BlenderbotTokenizer(PreTrainedTokenizer):
"""
A very simple chat template that just adds whitespace between messages.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}"

7
src/transformers/models/blenderbot/tokenization_blenderbot_fast.py
View File

@ -293,13 +293,6 @@ class BlenderbotTokenizerFast(PreTrainedTokenizerFast):
"""
A very simple chat template that just adds whitespace between messages.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}"

7
src/transformers/models/blenderbot_small/tokenization_blenderbot_small.py
View File

@ -224,13 +224,6 @@ class BlenderbotSmallTokenizer(PreTrainedTokenizer):
"""
A very simple chat template that just adds whitespace between messages.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}"

7
src/transformers/models/blenderbot_small/tokenization_blenderbot_small_fast.py
View File

@ -104,13 +104,6 @@ class BlenderbotSmallTokenizerFast(PreTrainedTokenizerFast):
"""
A very simple chat template that just adds whitespace between messages.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}"

7
src/transformers/models/bloom/tokenization_bloom_fast.py
View File

@ -155,11 +155,4 @@ class BloomTokenizerFast(PreTrainedTokenizerFast):
"""
A simple chat template that ignores role information and just concatenates messages with EOS tokens.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"

7
src/transformers/models/code_llama/tokenization_code_llama.py
View File

@ -456,13 +456,6 @@ class CodeLlamaTokenizer(PreTrainedTokenizer):
snippet](https://github.com/facebookresearch/llama/blob/556949fdfb72da27c2f4a40b7f0e4cf0b8153a28/llama/generation.py#L320-L362)
in the original repository.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
template = (
"{% if messages[0]['role'] == 'system' %}"
"{% set loop_messages = messages[1:] %}" # Extract system message if it's present

7
src/transformers/models/code_llama/tokenization_code_llama_fast.py
View File

@ -369,13 +369,6 @@ class CodeLlamaTokenizerFast(PreTrainedTokenizerFast):
snippet](https://github.com/facebookresearch/llama/blob/556949fdfb72da27c2f4a40b7f0e4cf0b8153a28/llama/generation.py#L320-L362)
in the original repository.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
template = (
"{% if messages[0]['role'] == 'system' %}"
"{% set loop_messages = messages[1:] %}" # Extract system message if it's present

7
src/transformers/models/cohere/tokenization_cohere_fast.py
View File

@ -247,13 +247,6 @@ class CohereTokenizerFast(PreTrainedTokenizerFast):
'<BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>'
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
default_template = (
"{{ bos_token }}"
"{% if messages[0]['role'] == 'system' %}"

7
src/transformers/models/gpt2/tokenization_gpt2.py
View File

@ -336,11 +336,4 @@ class GPT2Tokenizer(PreTrainedTokenizer):
"""
A simple chat template that ignores role information and just concatenates messages with EOS tokens.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"

8
src/transformers/models/gpt2/tokenization_gpt2_fast.py
View File

@ -147,11 +147,5 @@ class GPT2TokenizerFast(PreTrainedTokenizerFast):
"""
A simple chat template that ignores role information and just concatenates messages with EOS tokens.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"

7
src/transformers/models/gpt_neox/tokenization_gpt_neox_fast.py
View File

@ -234,11 +234,4 @@ class GPTNeoXTokenizerFast(PreTrainedTokenizerFast):
"""
A simple chat template that ignores role information and just concatenates messages with EOS tokens.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"

7
src/transformers/models/gpt_neox_japanese/tokenization_gpt_neox_japanese.py
View File

@ -165,13 +165,6 @@ class GPTNeoXJapaneseTokenizer(PreTrainedTokenizer):
"""
A simple chat template that just adds BOS/EOS tokens around messages while discarding role information.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{{ bos_token + eos_token + message.content + eos_token }}"

7
src/transformers/models/gpt_sw3/tokenization_gpt_sw3.py
View File

@ -301,13 +301,6 @@ class GPTSw3Tokenizer(PreTrainedTokenizer):
This chat template formats messages like an instant messenger chat log, with "User:" and "Bot:" strings
preceding messages. BOS tokens are added between all messages.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{{ eos_token }}{{ bos_token }}"
"{% for message in messages %}"

7
src/transformers/models/gptsan_japanese/tokenization_gptsan_japanese.py
View File

@ -246,13 +246,6 @@ class GPTSanJapaneseTokenizer(PreTrainedTokenizer):
A simple chat template that adds standard BOS, SEP and EOS tokens between messages while discarding role
information.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{% if not loop.first %}{{ bos_token}}{% endif %}"

8
src/transformers/models/idefics2/processing_idefics2.py
View File

@ -284,8 +284,14 @@ class Idefics2Processor(ProcessorMixin):
if self.chat_template is not None:
chat_template = self.chat_template
else:
logger.warning_once(
"No chat template is set for this processor, falling back to a default class-level template. This is "
"very error-prone, because models are often trained with templates different from the class default! "
"Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
chat_template = self.default_chat_template
return self.tokenizer.apply_chat_template(
conversation, chat_template=chat_template, tokenize=tokenize, **kwargs
)

7
src/transformers/models/llama/tokenization_llama.py
View File

@ -429,13 +429,6 @@ class LlamaTokenizer(PreTrainedTokenizer):
snippet](https://github.com/facebookresearch/llama/blob/556949fdfb72da27c2f4a40b7f0e4cf0b8153a28/llama/generation.py#L320-L362)
in the original repository.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
template = (
"{% if messages[0]['role'] == 'system' %}"
"{% set loop_messages = messages[1:] %}" # Extract system message if it's present

7
src/transformers/models/llama/tokenization_llama_fast.py
View File

@ -261,13 +261,6 @@ class LlamaTokenizerFast(PreTrainedTokenizerFast):
snippet](https://github.com/facebookresearch/llama/blob/556949fdfb72da27c2f4a40b7f0e4cf0b8153a28/llama/generation.py#L320-L362)
in the original repository.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
template = (
"{% if messages[0]['role'] == 'system' %}"
"{% set loop_messages = messages[1:] %}" # Extract system message if it's present

7
src/transformers/models/whisper/tokenization_whisper.py
View File

@ -815,13 +815,6 @@ class WhisperTokenizer(PreTrainedTokenizer):
"""
A simple chat template that ignores role information and just concatenates messages with EOS tokens.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"
def get_decoder_prompt_ids(self, task=None, language=None, no_timestamps=True):

7
src/transformers/models/whisper/tokenization_whisper_fast.py
View File

@ -544,13 +544,6 @@ class WhisperTokenizerFast(PreTrainedTokenizerFast):
"""
A simple chat template that ignores role information and just concatenates messages with EOS tokens.
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. "
"This is very error-prone, because models are often trained with templates different from the class "
"default! Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return "{% for message in messages %}" "{{ message.content }}{{ eos_token }}" "{% endfor %}"
# Copied from transformers.models.whisper.tokenization_whisper.WhisperTokenizer.get_decoder_prompt_ids

30
src/transformers/tokenization_utils_base.py
View File

@ -1746,16 +1746,27 @@ class PreTrainedTokenizerBase(SpecialTokensMixin, PushToHubMixin):
if tokenizer_kwargs is None:
tokenizer_kwargs = {}
using_default_template = False
# First, handle the cases when the model has a dict of multiple templates
if isinstance(self.chat_template, dict) or (
self.chat_template is None and isinstance(self.default_chat_template, dict)
):
template_dict = self.chat_template or self.default_chat_template
if self.chat_template is not None:
template_dict = self.chat_template
using_default_dict = False
else:
template_dict = self.default_chat_template
using_default_dict = True
if chat_template is not None and chat_template in template_dict:
# The user can pass the name of a template to the chat template argument instead of an entire template
chat_template = template_dict[chat_template]
if using_default_dict:
using_default_template = True
elif chat_template is None and "default" in template_dict:
chat_template = template_dict["default"]
if using_default_dict:
using_default_template = True
elif chat_template is None:
raise ValueError(
"This model has multiple chat templates with no default specified! Please either pass a chat "
@ -1769,6 +1780,16 @@ class PreTrainedTokenizerBase(SpecialTokensMixin, PushToHubMixin):
chat_template = self.chat_template
else:
chat_template = self.default_chat_template
using_default_template = True
if using_default_template:
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a default class-level template. This is "
"very error-prone, because models are often trained with templates different from the class default! "
"Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
# Compilation function uses a cache to avoid recompiling the same template
compiled_template = self._compile_jinja_template(chat_template)
@ -1840,13 +1861,6 @@ class PreTrainedTokenizerBase(SpecialTokensMixin, PushToHubMixin):
This template formats inputs in the standard ChatML format. See
https://github.com/openai/openai-python/blob/main/chatml.md
"""
logger.warning_once(
"No chat template is set for this tokenizer, falling back to a ChatML template. "
"This is very error-prone, because most models are not trained with a ChatML template!"
"Default chat templates are a legacy feature and will be removed in Transformers v4.43, at which "
"point any code depending on them will stop working. We recommend setting a valid chat template before "
"then to ensure that this model continues working without issues."
)
return (
"{% for message in messages %}"
"{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}"