[Bugfix] Get a specific type of layer from forward context #17222
Conversation
Signed-off-by: Chen Zhang <[email protected]>
Signed-off-by: Chen Zhang <[email protected]>
heheda12345
requested review from
WoosukKwon,
alexm-redhat,
comaniac,
njhill,
robertgshaw2-redhat and
ywang96
as code owners
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run 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 either: Add 🚀 |
| def get_layers_from_vllm_config(vllm_config: VllmConfig, | ||
| layer_type: type[T]) -> dict[str, T]: | ||
| return { | ||
| layer_name: layer | ||
| for layer_name, layer in | ||
| vllm_config.compilation_config.static_forward_context.items() | ||
| if isinstance(layer, layer_type) | ||
| } |
There was a problem hiding this comment.
Are there any caveats to the static forward context? Can we always use this or only sometimes? (I'm not sure) Could you add a comment if there's anything folks should be careful of?
There was a problem hiding this comment.
I've updated the comment to show that there may be different types of layers in forward context.
Signed-off-by: Chen Zhang <[email protected]>
…ect#17222) Signed-off-by: Chen Zhang <[email protected]>
…ect#17222) Signed-off-by: Chen Zhang <[email protected]>
…ect#17222) Signed-off-by: Chen Zhang <[email protected]> Signed-off-by: Agata Dobrzyniewicz <[email protected]>
…ect#17222) Signed-off-by: Chen Zhang <[email protected]> Signed-off-by: Mu Huai <[email protected]>
…ect#17222) Signed-off-by: Chen Zhang <[email protected]> Signed-off-by: Yuqi Zhang <[email protected]>
3 participants
Forward context saves both
AttentionandFusedMOElayers. Add an interface to only get one type of layer from forward context.Fix bug:
VLLM_ATTENTION_BACKEND=FLASHINFER python3 examples/offline_inference/data_parallel.py --model="Qwen/Qwen1.5-MoE-A2.7B-Chat" --dp-size=2 --tp-size=1