[Hardware][CPU] Cross-attention and Encoder-Decoder models support on CPU backend

[Isotr0py]

FILL IN THE PR DESCRIPTION HERE

FIX #9114

• Add cross-attention support for SDPA backend.
• Add Encoder-Decoder models support for CPU backend.

TODO

• Clean up the code
• Enable bart test for CPU CI

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!

[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.

🚀

[Isotr0py]

Seems that mllama also works on CPU now:

$ python examples/offline_inference_vision_language.py -m mllama --num-prompts 1
WARNING 10-07 11:23:31 config.py:352] Async output processing is only supported for CUDA or TPU. Disabling it for other platforms.
INFO 10-07 11:23:31 llm_engine.py:237] Initializing an LLM engine (v0.6.1.post3.dev210+g83caf35e.d20241003) with config: model='/data/LLM-model/Llama-3.2-11B-Vision-Instruct', speculative_config=None, tokenizer='/data/LLM-model/Llama-3.2-11B-Vision-Instruct', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, rope_scaling=None, rope_theta=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=4096, download_dir=None, load_format=LoadFormat.AUTO, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=True, kv_cache_dtype=auto, quantization_param_path=None, device_config=cpu, decoding_config=DecodingConfig(guided_decoding_backend='outlines'), observability_config=ObservabilityConfig(otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=0, served_model_name=/data/LLM-model/Llama-3.2-11B-Vision-Instruct, use_v2_block_manager=True, num_scheduler_steps=1, chunked_prefill_enabled=False multi_step_stream_outputs=True, enable_prefix_caching=False, use_async_output_proc=False, use_cached_outputs=False, mm_processor_kwargs=None)
WARNING 10-07 11:23:32 cpu_executor.py:354] Environment variable VLLM_CPU_KVCACHE_SPACE (GB) for CPU backend is not set, using 4 by default.
INFO 10-07 11:23:32 selector.py:183] Cannot use _Backend.FLASH_ATTN backend on CPU.
INFO 10-07 11:23:32 selector.py:128] Using Torch SDPA backend.
INFO 10-07 11:23:36 selector.py:183] Cannot use _Backend.FLASH_ATTN backend on CPU.
INFO 10-07 11:23:36 selector.py:128] Using Torch SDPA backend.
Loading safetensors checkpoint shards:   0% Completed | 0/5 [00:00<?, ?it/s]
Loading safetensors checkpoint shards:  20% Completed | 1/5 [00:55<03:42, 55.62s/it]
Loading safetensors checkpoint shards:  40% Completed | 2/5 [01:51<02:47, 55.70s/it]
Loading safetensors checkpoint shards:  60% Completed | 3/5 [02:46<01:50, 55.38s/it]
Loading safetensors checkpoint shards:  80% Completed | 4/5 [03:41<00:55, 55.29s/it]
Loading safetensors checkpoint shards: 100% Completed | 5/5 [03:56<00:00, 40.76s/it]
Loading safetensors checkpoint shards: 100% Completed | 5/5 [03:56<00:00, 47.30s/it]

INFO 10-07 11:27:33 cpu_executor.py:212] # CPU blocks: 1638
WARNING 10-07 11:27:35 preprocess.py:87] Falling back on <BOS> for decoder start token id because decoder start token id is not available.
Processed prompts: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [02:34<00:00, 154.93s/it, est. speed input: 0.06 toks/s, output: 0.41 toks/s]
 The image shows a close-up of a tree branch with green leaves and a few brown branches. The tree branch is in the center of the image, and it is covered in small, green leaves. There are also a few brown branches visible on the tree. The background of the image is a light gray color, and

[Isotr0py]

This aims to fix a bug introduced in #8729.

[DarkLight1337]

Assuming that the tests pass, looks OK to me. However, I'm worried that the number of model runners may become unmaintainable as we support different type of models (decoder-only, encoder-decoder, and embedding) for each hardware backend (CPU, GPU, TPU, ...). Is there a good way to decouple them? @cadedaniel may be able to give some code suggestions.

Edit: In any case, let's do those refactorings in another PR.