Support HQT on VLLM #59

nirda7 · 2024-06-12T11:05:11Z

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.

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!

…use of matmul class

vllm/model_executor/models/llama.py

vllm/utils.py

vllm/worker/habana_model_runner.py

kzawora-intel · 2024-06-19T12:48:28Z

vllm/worker/habana_model_runner.py

+                print('hqt shutdown start')
+                import habana_quantization_toolkit
+                if habana_quantization_toolkit is not None:
+                    habana_quantization_toolkit.finish_measurements(self.model.model)


self.model.model might not exist here

Why?
you wrap it with HpuModelAdapter anyway.

kzawora-intel · 2024-06-19T12:49:21Z

vllm/hpu/ops.py

@@ -38,8 +38,8 @@ def fetch_from_cache(cache, blocks, permutations):
    return [cache.index_select(0, blocks[:, i]).permute(permutations) for i in range(blocks.size(1))]


-@hpu_utils.with_mark_steps
-def paged_attention_v1(query, key_cache, value_cache, head_mapping, scale, block_tables, context_lens, block_size, alibi_slopes, kv_cache_dtype=None) -> None:
+def paged_attention_v1(query, key_cache, value_cache, head_mapping, scale, block_tables, context_lens, block_size, alibi_slopes, kv_cache_dtype=None,


why is @hpu_utils.with_mark_steps removed here?

we want that all convert to\from hf8 will be in the same graph
so we remove this mark step and add one outside the transformer block align to ohf version.

kzawora-intel · 2024-06-19T12:55:02Z

vllm/attention/backends/habana_attn.py

+            block_indices, block_offset = cache_ops.prepare_to_cache(key_cache, attn_metadata.slot_mapping)
+            key_cache = self.key_cache(key, key_cache, block_indices, block_offset)
+            value_cache = self.value_cache(value, value_cache, block_indices, block_offset)


@madamczykhabana can you check if this is fine? aren't we making HabanaPagedAttention.write_to_paged_cache useless here?

madamczykhabana · 2024-06-28T12:16:04Z

Relevant changes moved to #75

dudilester · 2024-07-03T11:33:44Z

vllm/model_executor/layers/layernorm.py

@@ -72,8 +72,8 @@ def forward(
            return x, residual
        if x.device.type == "hpu" and FusedRMSNorm:
            orig_dtype = x.dtype


orig_dtype not used, here and also in line 60

remove expert_max hard code (#47) vLLM-Ext: Full enabling of ALiBi (#34) Add version inference via setuptools-scm (#58) Revert "vLLM-Ext: Full enabling of ALiBi (#34)" (#59) Remove punica_hpu.py from vllm_hpu_extension (#66) Removed previous (not-pipelined) pa implementation (#72) Add flag to enable running softmax in fp32 (#71) Update calibration readme link (#73) allow lm_head quantization in calibration process (#65) Pad to bmin if value is less (#67) Update pyproject.toml (#75) --------- Co-authored-by: Michał Kuligowski <mkuligowski@habana.ai>

nirda7 added 6 commits June 18, 2024 11:39

support hqt on vllm

4b9b955

Support HQT on VLLM - KVCache and Mark Step uses

f3ffc8c

HQT on VLLM - prep model and finish measurements and multi cards run

8ffc3d0

HQT on VLLM - separate kv caches

f5f0972

HQT on VLLM - remove code duplications

c521c4d

HQT on VLLM - move matmul and softmax to hpu utils and revert logits …

64c8c7f

…use of matmul class

nirda7 force-pushed the vllm-hqt-fork branch from ae8593e to 64c8c7f Compare June 18, 2024 08:41

Move model to hpu when HQT is not used

2e291c5