vllm
vllm copied to clipboard
vllm-project
[Model] Jamba support
Add Jamba support to vLLM,
This PR comprises two parts: the Jamba modeling file and the Mamba memory handling.
Since Jamba is a hybrid model (which alternates between mamba and transformer layers),
In the current PR state, we preallocate a "large" Mamba cache buffer per worker.
Similarly to how KV cache is preallocated in the CacheEngine,
And pass slices of this preallocated buffer into the forward pass according to the required batch size, maintaining this buffer throughout the generation pipeline by keeping track of the occupied slices with a dictionary that is duplicated throughout all of the workers (contrary to how vLLM handles this).
Would appreciate your guidance on handling the Mamba cache memory and wrap up the PR
- [x] Support for sharded Mamba layers (TP).
- [ ] vLLM way for handling the Mamba cache memory.
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,LLMclass, 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.shto 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-requiredlabel 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!
May I get an update regarding the status of this PR? It seems the author stopped working on it?
May I get an update regarding the status of this PR? It seems the author stopped working on it?
We're currently still working on it, The PR works well, we're just waiting for a more suitable cache blocks manger that will be able to accommodate the Mamba cache (seqlen agnostic cache) instead of just managing it directly inside each worker. We're in direct contact with the vLLM team in that regard.
QQ: Does this PR support parallel sampling (i.e., n > 1 in sampling params)? While I don't think it is not necessary to support parallel sampling in this PR, I'd like to know if this case was considered. Supporting parallel sampling might be a bit hard since it requires implementing copy_blocks for the Mamba cache. If this is not trivial, please leave a comment on the code.
QQ: Does this PR support parallel sampling (i.e.,
n> 1 in sampling params)? While I don't think it is not necessary to support parallel sampling in this PR, I'd like to know if this case was considered. Supporting parallel sampling might be a bit hard since it requires implementingcopy_blocksfor the Mamba cache. If this is not trivial, please leave a comment on the code.
This PR does support parallel sampling, We transfer the seq_ids along with the requests_ids into the Jamba inner state mapping and copy the blocks accordingly, reference in the Jamba code