perf: EP-aware DBCache default (CPU 0.25 on, GPU off) - #7
Merged
Conversation
Measured: DBCache's split path helps CPU (-45% at nsteps=8) but on GPU the per-step host round-trips regress quantized weights (Vulkan+Q8 nsteps=8: cache ON 1.05s vs OFF 0.88s = +20%; F32: ON 0.62s vs OFF 0.67s = minor benefit). Make the default EP-aware instead of blanket-on: - InferParams.db_cache_threshold default becomes -1 (auto): CPU -> 0.25, GPU backends (vulkan/cuda/metal) -> 0 (off); explicit values unchanged - CLI: --cache-threshold default 'auto'; help updated - README EN/CN: document the EP-aware default
ggml_backend_name returns 'Vulkan0' (etc.), so the exact lowercase match missed; use case-insensitive substring so GPU defaults to DBCache-off.
|
Important Review available on request
Reviews should be triggered manually for repositories with fewer than 10 stars. Select Trigger review above or comment ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Measured (10s clip, seed 42):\n- CPU nsteps=8: cache ON 14.0s vs OFF 25.4s (big win)\n- Vulkan nsteps=8: Q8 cache ON 1.05s vs OFF 0.88s (regression); F32 ON 0.62s vs OFF 0.67s (marginal win)\n\nSo the blanket 'DBCache on for multi-step' (PR #4) default is wrong on GPU, especially with quantized weights (split-path per-step host round trips). Make it EP-aware:\n\n- \db_cache_threshold\ default -1 = auto: CPU -> 0.25, GPU backends (vulkan/cuda/metal) -> 0 (off)\n- explicit --cache-threshold (incl. 0) unchanged\n- README EN/CN updated\n\nVerified locally: CPU n8 default keeps cache (14.1s / 33 notes); Vulkan Q8 n8 default now cache-off 0.912s / 33 notes.