Conversation
The parallel-agnostic fs-tier cache (vllm-project#44733) collapses tp/pp/pcp/dcp and rank out of the cache namespace for a single full-attention group, on the assumption that the offloaded KV blocks are parallelism-invariant. That invariant is not known to hold under the V2 model runner, whose KV layout may differ, so sharing a cache directory across layouts there could alias distinct blocks. Gate the opt-in on the canonical vllm_config.use_v2_model_runner property (not a raw env-var read, so config-driven V2 defaults such as diffusion models are handled correctly). The check sits alongside the existing MLA and multi-group exclusions in FileMapper, keeping the call sites unchanged. Adds a regression test asserting tp/rank are not collapsed when the V2 model runner is active. Co-authored-by: Claude Signed-off-by: Itay Etelis <etelis2019@gmail.com>
Co-authored-by: Claude Signed-off-by: Itay Etelis <etelis2019@gmail.com>
Signed-off-by: Itay Etelis <etelis2019@gmail.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Purpose
Disable the parallel-agnostic fs-tier KV cache (vllm-project#44733) when the V2 model runner
is active. The parallelism-invariant assumption it relies on does not hold under
V2.