[BugFix] Apply get_cached_tokenizer to the tokenizer setter of LLM

[DriverSong]

Change the func to get lengh of tokenizer from len(tokenizer) to tokenizer.vocab_size

FIX the issue decribed in #5206 #5240

FIX #5206
FIX #5240

BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE

---

PR Checklist (Click to Expand)

Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

• [Bugfix] for bug fixes.
• [CI/Build] for build or continuous integration improvements.
• [Doc] for documentation fixes and improvements.
• [Model] for adding a new model or improving an existing model. Model name should appear in the title.
• [Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
• [Kernel] for changes affecting CUDA kernels or other compute kernels.
• [Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
• [Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
• [Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

• We adhere to Google Python style guide and Google C++ style guide.
• Pass all linter checks. Please use format.sh to format your code.
• The code need to be well-documented to ensure future contributors can easily understand the code.
• Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
• Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

• After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
• After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
• After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
• Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

[youkaichao]

Stange, this property should already be cached:

vllm/vllm/transformers_utils/tokenizer.py

Lines 17 to 55 in 7a64d24

	def get_cached_tokenizer(
	tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
	) -> Union[PreTrainedTokenizer, PreTrainedTokenizerFast]:
	"""Get tokenizer with cached properties.
	This will patch the tokenizer object in place.
	By default, transformers will recompute multiple tokenizer properties
	each time they are called, leading to a significant slowdown. This
	function caches these properties for faster access."""
	tokenizer_all_special_ids = set(tokenizer.all_special_ids)
	tokenizer_all_special_tokens_extended = (
	tokenizer.all_special_tokens_extended)
	tokenizer_all_special_tokens = set(tokenizer.all_special_tokens)
	tokenizer_len = len(tokenizer)
	class CachedTokenizer(tokenizer.__class__): # type: ignore
	@property
	def all_special_ids(self):
	return tokenizer_all_special_ids
	@property
	def all_special_tokens(self):
	return tokenizer_all_special_tokens
	@property
	def all_special_tokens_extended(self):
	return tokenizer_all_special_tokens_extended
	def __len__(self):
	return tokenizer_len
	CachedTokenizer.__name__ = f"Cached{tokenizer.__class__.__name__}"
	tokenizer.__class__ = CachedTokenizer
	return tokenizer

[DriverSong]

Stange, this property should already be cached:

vllm/vllm/transformers_utils/tokenizer.py

Lines 17 to 55 in 7a64d24

	def get_cached_tokenizer(
	tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
	) -> Union[PreTrainedTokenizer, PreTrainedTokenizerFast]:
	"""Get tokenizer with cached properties.
	This will patch the tokenizer object in place.
	By default, transformers will recompute multiple tokenizer properties
	each time they are called, leading to a significant slowdown. This
	function caches these properties for faster access."""
	tokenizer_all_special_ids = set(tokenizer.all_special_ids)
	tokenizer_all_special_tokens_extended = (
	tokenizer.all_special_tokens_extended)
	tokenizer_all_special_tokens = set(tokenizer.all_special_tokens)
	tokenizer_len = len(tokenizer)
	class CachedTokenizer(tokenizer.__class__): # type: ignore
	@property
	def all_special_ids(self):
	return tokenizer_all_special_ids
	@property
	def all_special_tokens(self):
	return tokenizer_all_special_tokens
	@property
	def all_special_tokens_extended(self):
	return tokenizer_all_special_tokens_extended
	def __len__(self):
	return tokenizer_len
	CachedTokenizer.__name__ = f"Cached{tokenizer.__class__.__name__}"
	tokenizer.__class__ = CachedTokenizer
	return tokenizer

I change the tokenizer by the func set_tokenizer of llm after LLM init. Maybe we should change this func as below

from vllm.transformers_utils.tokenizer import get_cached_tokenizer

    def set_tokenizer(
        self,
        tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast],
    ) -> None:
        self.llm_engine.tokenizer.tokenizer = get_cached_tokenizer(tokenizer)

[DriverSong]

@youkaichao
I've tested upper change of set_tokenizer on qwen1.5-0.5B with the AdvertiseGen dataset., the generate time decreased to 47 seconds.

Processed prompts:  95%|█████████▍| 1014/1070 [00:43<00:01, 45.00it/s]
Processed prompts:  96%|█████████▌| 1022/1070 [00:43<00:00, 52.35it/s]
Processed prompts:  96%|█████████▌| 1028/1070 [00:44<00:00, 44.96it/s]
Processed prompts:  97%|█████████▋| 1034/1070 [00:44<00:00, 40.40it/s]
Processed prompts:  97%|█████████▋| 1043/1070 [00:44<00:00, 49.90it/s]
Processed prompts:  98%|█████████▊| 1049/1070 [00:44<00:00, 36.90it/s]
Processed prompts:  99%|█████████▊| 1054/1070 [00:44<00:00, 28.55it/s]
Processed prompts:  99%|█████████▉| 1058/1070 [00:45<00:00, 27.48it/s]
Processed prompts:  99%|█████████▉| 1062/1070 [00:45<00:00, 23.82it/s]
Processed prompts: 100%|█████████▉| 1065/1070 [00:45<00:00, 21.18it/s]
Processed prompts: 100%|█████████▉| 1068/1070 [00:45<00:00, 20.53it/s]
Processed prompts: 100%|██████████| 1070/1070 [00:47<00:00, 22.66it/s]
master-0: [2024-06-04 15:00:18,031] [INFO] [launch.py:351:main] Process 1405 exits successfully.

[youkaichao]

better add a check, if the tokenizer is already cached, then just set it.

[DriverSong]

A check is added. While the CachedTokenizer is dynamic, I have no choice but compare the class name. Is there any good ideal?

[mgoin]

I think the string compare is fine here. If you wanted to be extremely safe you could change CachedTokenizer to be VLLMCachedTokenizer

[youkaichao]

LGTM in general, please address the comments.

[mgoin]

I think the string compare is fine here. If you wanted to be extremely safe you could change CachedTokenizer to be VLLMCachedTokenizer