Remove redundant set_active_loras call during warmup #413
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
michalkuligowski
approved these changes
Oct 22, 2024
xuechendi
pushed a commit
to xuechendi/vllm-fork
that referenced
this pull request
Oct 23, 2024
CUDA uses `capture` for warmup runs and `execute_model` for actual runs. During each phase they call `set_active_loras` only once. HPU uses `execute_model` for both warmup and actual runs. Since `execute_model` already takes care of `set_active_loras` internally, the redundant call can be removed. This special handling is redundant and incorrect, as it causes out-of-bound slicing in decode phase reported in HabanaAI#405. This PR removes special handling of `set_active_loras` function call from warmup runs and resolves the issue in HabanaAI#405.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
CUDA uses
capturefor warmup runs andexecute_modelfor actual runs. During each phase they callset_active_lorasonly once. HPU usesexecute_modelfor both warmup and actual runs. Sinceexecute_modelalready takes care ofset_active_lorasinternally, the redundant call can be removed.This special handling is redundant and incorrect, as it causes out-of-bound slicing in decode phase reported in #405.
This PR removes special handling of
set_active_lorasfunction call from warmup runs and resolves the issue in #405.