[Attention] MLA with chunked prefill

[LucasWilkinson]

Need to do more benchmarking to see if this makes sense to be on by default in V0, but lays the groundwork for a V1 implementation. (#13111 may help performance)

lm_eval --model vllm --model_args pretrained=deepseek-ai/DeepSeek-V2-Lite-Chat,tensor_parallel_size=2,dtype=auto,gpu_memory_utilization=0.9,trust_remote_code=True,max_model_len=16384,enable_chunked_prefill=False --task gsm8k --num_fewshot=5 --limit 100

vllm (pretrained=deepseek-ai/DeepSeek-V2-Lite-Chat,tensor_parallel_size=2,dtype=auto,gpu_memory_utilization=0.9,trust_remote_code=True,max_model_len=16384,enable_chunked_prefill=False), gen_kwargs: (None), limit: 100.0, num_fewshot: 5, batch_size: 1
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value|   |Stderr|
|-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  | 0.66|±  |0.0476|
|     |       |strict-match    |     5|exact_match|↑  | 0.66|±  |0.0476|


lm_eval --model vllm --model_args pretrained=deepseek-ai/DeepSeek-V2-Lite-Chat,tensor_parallel_size=2,dtype=auto,gpu_memory_utilization=0.9,trust_remote_code=True,max_model_len=16384,enable_chunked_prefill=True --task gsm8k --num_fewshot=5 --limit 100


vllm (pretrained=deepseek-ai/DeepSeek-V2-Lite-Chat,tensor_parallel_size=2,dtype=auto,gpu_memory_utilization=0.9,trust_remote_code=True,max_model_len=16384,enable_chunked_prefill=True), gen_kwargs: (None), limit: 100.0, num_fewshot: 5, batch_size: 1
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value|   |Stderr|
|-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  | 0.66|±  |0.0476|
|     |       |strict-match    |     5|exact_match|↑  | 0.66|±  |0.0476|

Shout-out to @pathorn for assisting with hardening this PR

Future work:

• Allocate the worst case result of self.kv_b_proj(kv_c_normed) in the profile run
• [Attention] MLA with chunked prefill #12639 (comment)
• Improved algo for allocating workspace amongst batch elements
• Improve how the workspace is allocated

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @LucasWilkinson.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @LucasWilkinson.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[tlrmchlsmth]

ok yeah that makes sense for some of the red tests

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @LucasWilkinson.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[ZhongYingMatrix]

Hi @LucasWilkinson thx for ur wonderful work!
I am a little confused on the backend that got from get_attn_backend_cls.
Since we should set VLLM_USE_V1 to use chunked prefill, from here, we would get vllm.v1.attention.backends.flash_attn.FlashAttentionBackend instead of vllm.attention.backends.triton_mla.TritonMLABackend?

[oreo-wjx]

@LucasWilkinson A dumb question, how are these magic number(8 full length request & 4 pages) been decided?

[LucasWilkinson]

arbitrarily; just whatever seemed reasonable, we want want it so under common loads theres enough work space that we don't have to chunk the context but not so large that we materially impact KV-cache space. Feel free to tune these and open a PR if you come up with better numbers; we'd appreciate the contribution!