[Bugfix / Core] Prefix Caching Guards (merged with main) #4846

zhuohan123 · 2024-05-16T01:58:07Z

Updated version of #3903

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!

rkooo567 · 2024-05-17T15:35:50Z

vllm/config.py

@@ -251,6 +263,18 @@ def get_sliding_window(self) -> Optional[int]:
            return None
        return getattr(self.hf_text_config, "sliding_window", None)

+    def get_sliding_window(self) -> Optional[int]:


How does it work for the model that already has sliding window like mistral?

Im not sure what you mean?

If the user does not specify --disable-sliding-window then we use sliding window if the model supports it

oh maybe it is a dumb question, but my question is for models that has slinding window by default https://huggingface.co/mistralai/Mistral-7B-v0.1/blob/26bca36bde8333b5d7f72e9ed20ccda6a618af24/config.json#L18, if we use --disable-sliding-window, does it work properly?

Yes, specifically what this does is handle a case like Mistral.

--disable-sliding-window means we turn off sliding window and set max_model_len=sliding_window

So in the case of Mistral, we then would treat the model as a 4096 ctx-len model with no sliding window.

The reason for this feature is that if we want to use features that are incompatible with sliding window (e.g. APC or chunked prefill), then there is a pathway to disable sliding window

I see. that makes sense! Thanks for the explanation

rkooo567

LGTM. Minor comments. Didn't review _get_and_verify_max_len and _get_and_verify_dtype assuming it is just code refactored (lmk if it is wrong)

rkooo567 · 2024-05-18T07:11:03Z

vllm/config.py

+        if self.disable_sliding_window:
+            logger.info("Sliding window is disabled per configuration. "
+                        "Model max length will be capped at sliding window "
+                        "length.")


Suggested change

"length.")

"length, %d tokens", self.get_hf_config_sliding_window())

rkooo567 · 2024-05-18T07:11:59Z

vllm/engine/arg_utils.py

+        parser.add_argument('--disable-sliding-window',
+                            action='store_true',
+                            help='Disables sliding window if the model '
+                            'supports sliding window')


Can you mention the model length is capped by the slinding window size?

robertgshaw2-neuralmagic · 2024-05-23T20:14:10Z

because sliding window is propogated to attention, this is going to require me to edit most model files.

Will get back to this tomorrow after I get mistral over the line

…t non-uniform caching

robertgshaw2-neuralmagic · 2024-05-24T17:27:01Z

vllm/model_executor/models/qwen2.py

@@ -173,18 +168,14 @@ def __init__(
        # Requires transformers > 4.32.0
        rope_theta = getattr(config, "rope_theta", 1000000)
        rope_scaling = getattr(config, "rope_scaling", None)
-        use_sliding_window = (config.use_sliding_window


{ "architectures": [ "Qwen2ForCausalLM" ], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 11008, "max_position_embeddings": 32768, "max_window_layers": 28, << qwen2 uses sliding window for some layers "model_type": "qwen2", "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "sliding_window": 32768, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "transformers_version": "4.37.0", "use_cache": true, "use_sliding_window": false, << qwen2 does not use sliding window by default "vocab_size": 151936 }

I think we had a bug in Qwen2 - this path will not be followed very often b/c qwen2 does not use sliding window by default

Currently, if use_sliding_window=True, only some layers will use sliding window. But we have global KV cache management that would treat KVs the same. So I do not see how it is possible that this could be working correctly.

This is not a very common user path because they would have to opt into sliding window on Qwen.

So I disabled this by default.

robertgshaw2-neuralmagic · 2024-05-24T17:44:25Z

Disabling sliding window ended up being more work than I expected because we broke some abstractions where the models are accessing the hf_config to determine whether sliding window is used when passing arguments to attention. As a result, the user's specification is ignored.

So, I updated Attention to use the cache_config's value, which is set properly rather than letting the model definition specify. This enabled me to remove the sliding_window argument from the various layers in the model.

Additionally, I noticed in this that Qwen2 attempts to support having only some layers with sliding window. We do not support this in our KV cache management, so I removed this bug by Failing if the system is configured this way. (note: this is not a popular codepath b/c Qwen2 does not use sliding window by default.

…t#4846) Co-authored-by: rsnm2 <rshaw@neuralmagic.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-neuralmagic@users.noreply.github.com>

robertgshaw2-neuralmagic and others added 10 commits April 7, 2024 21:59

added guards for prefix-caching. added ability to disable sliding window

56680e7

format.sh

64aac2e

added tests

28ae0cc

Merge remote-tracking branch 'upstream/main' into prefix-caching-guards

2a01ae6

merge

cd0f666

removed images

f30c3de

fixed bad merge

da5a982

./format

8502b6a

validated that prefix caching working on turing with recent update

1bef541

Merge branch 'main' into prefix-caching-guards

6620e53

zhuohan123 requested a review from robertgshaw2-neuralmagic May 16, 2024 01:58

zhuohan123 changed the title ~~Prefix caching guards new~~ [Bugfix / Core] Prefix Caching Guards (merged with main) May 16, 2024

zhuohan123 mentioned this pull request May 16, 2024

[Bugfix / Core] Prefix Caching Guards #3903

Closed

rkooo567 reviewed May 17, 2024

View reviewed changes

rkooo567 approved these changes May 18, 2024

View reviewed changes

robertgshaw2-neuralmagic mentioned this pull request May 19, 2024

v0.4.3 Release Tracker #4895

Closed

6 tasks

robertgshaw2-neuralmagic added 2 commits May 23, 2024 11:33

Merge branch 'main' into prefix-caching-guards-new

8efc774

updated to remove sliding window usage in models

033c2c5

robertgshaw2-neuralmagic added 11 commits May 24, 2024 16:33

removed spurious changes

0638960

revert change to make PR easier to read

63c0097

more cleanup

4a3630c

more cleanup

3f73426

stash

6f754c3

cleanup prints and comments to match current for easier review

a497b7b

more cleanup for PR readibility

9fd64fe

more cleanup for PR readibility

1126c5a

more cleanup for PR readibility

8a53180

removed from mixtral, need to fix qwen

034bbde

updated models to remove sliding window. Big update to qwen to preven…

b56352b

…t non-uniform caching

robertgshaw2-neuralmagic reviewed May 24, 2024

View reviewed changes

robertgshaw2-neuralmagic and others added 9 commits May 24, 2024 17:52

format

8225c3f

added test and fixed requirements dev

13797c1

added test

37efe98

format

7b186c2

updated comment

93bce37

updated test

7c8a9d0

format

4285763

fixed logging

84253fd

Update test_disable_sliding_window.py

7a61f51

zhuohan123 merged commit 1102bef into main May 27, 2024
63 checks passed

simon-mo deleted the prefix-caching-guards-new branch October 28, 2024 16:50

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Bugfix / Core] Prefix Caching Guards (merged with main) #4846

[Bugfix / Core] Prefix Caching Guards (merged with main) #4846

zhuohan123 commented May 16, 2024

rkooo567 May 17, 2024

robertgshaw2-neuralmagic May 17, 2024

rkooo567 May 17, 2024

robertgshaw2-neuralmagic May 17, 2024

rkooo567 May 17, 2024

rkooo567 left a comment

rkooo567 May 18, 2024

rkooo567 May 18, 2024

robertgshaw2-neuralmagic commented May 23, 2024

robertgshaw2-neuralmagic May 24, 2024

robertgshaw2-neuralmagic May 24, 2024 •

edited

Loading

robertgshaw2-neuralmagic commented May 24, 2024

	"length.")
	"length, %d tokens", self.get_hf_config_sliding_window())

[Bugfix / Core] Prefix Caching Guards (merged with main) #4846

[Bugfix / Core] Prefix Caching Guards (merged with main) #4846

Conversation

zhuohan123 commented May 16, 2024

PR Title and Classification

Code Quality

Notes for Large Changes

What to Expect for the Reviews

Thank You

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rkooo567 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

robertgshaw2-neuralmagic commented May 23, 2024

Choose a reason for hiding this comment

robertgshaw2-neuralmagic May 24, 2024 • edited Loading

Choose a reason for hiding this comment

robertgshaw2-neuralmagic commented May 24, 2024

robertgshaw2-neuralmagic May 24, 2024 •

edited

Loading