[New Feature][Habana-Main] speculative_decoding HPU support

[xuechendi]

Req - https://jira.habana-labs.com/browse/REQ-289 => target for 1.19

TODO:

• There remains one hardcode to HPUWorker, need to remove

Next Steps:

• 1. submit necessary codes change to vllm-upstream branch => WIP
• 2. support all 3 draft_model_types - mlp_speculator, medusa and others

FILL IN THE PR DESCRIPTION HERE

FIX #xxxx (link existing issues this PR will resolve)

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.

Adding or changing kernels

Each custom kernel needs a schema and one or more implementations to be registered with PyTorch.

• Make sure custom ops are registered following PyTorch guidelines: Custom C++ and CUDA Operators and The Custom Operators Manual
• Custom operations that return Tensors require meta-functions. Meta-functions should be implemented and registered in python so that dynamic dims can be handled automatically. See above documents for a description of meta-functions.
• Use torch.libary.opcheck() to test the function registration and meta-function for any registered ops. See tests/kernels for examples.
• When changing the C++ signature of an existing op, the schema must be updated to reflect the changes.
• If a new custom type is needed, see the following document: Custom Class Support in PT2.

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!

[xuechendi]

Run test with

python examples/offline_inference_spec_decode.py

[xuechendi]

@michalkuligowski , I fixed all comments, some suggestion might not work with existing codes, so I add explanation in the review

[xuechendi]

@michalkuligowski , may you help to trigger the CI again, I fixed yapf detected format issues.

[xuechendi]

@michalkuligowski , I updated the codes according to your last comments. For the draft_model_runner.py importing behavior change issue you mentioned in last comment, since draft_model_runner will be imported by spec_decode_runner and multi_step_decode_runner, we need to prevent the unnecessary importing error termination due to Cuda and ROCm flashattn support.
Since you think the previous fixing by simply changing raise to print is not very good, I provided an alternative fixing in latest commit.

After this PR merged, I will move on to the complete draft model support for medusa and mlp, and I'll revisit this draft_model_runner.py for better support.

[xuechendi]

@michalkuligowski , please help to review.

The previous commit of using platform to switch between cuda-alike and hpu leads to a coding format issue, which yapf and ruff gave me different fixing suggestion and I can't get both of them passed the check.

That is why I have to change it to use try and expect.

[xuechendi]

@michalkuligowski , rebased this PR to latest habana_main.

As mentioned in last commit, in order to pass CICD formatting check, I have to still use "try and except" in draft_model_runner.py. Previous way of using "platform check" causes a formatting conflict issue that yapf prefers one format and isort asks for another one, so I can't make CICD passed.

Please have a review.

[xuechendi]

@michalkuligowski , I submitted my second PR: #461

PR461 is based on PR375, and enabled Medusa and MLP speculator there.

[xuechendi]

@michalkuligowski , I have rebased this PR to latest habana_main - changes are in 20a2e6

Meanwhile, I tested with multi_step_scheduler using cmdline as below => passed (The multi-step-worker in spec_decode is different to the one for multi-step-scheduler => so it will not affect multi-step-scheduler)

VLLM_PROFILER_ENABLED=true VLLM_SKIP_WARMUP=true VLLM_CONTIGUOUS_PA=true VLLM_PA_SOFTMAX_IMPL='wsum' python benchmarks/benchmark_throughput.py \
        --model ${model} \
        --device hpu \
        --backend vllm \
        --num-prompts ${bs} \
        --input_len ${in_len} \
        --output_len ${out_len} \
        --max_model_len ${total_len} \
        --dtype bfloat16 \
        --num_scheduler_steps 16 \
        --tensor_parallel_size 1 \
        --gpu-memory-util 0.9