[TPU] Make --list-impls-for-benchmark-ci TPU-aware#3020
Draft
norx1991 wants to merge 1 commit into
Draft
Conversation
On --device tpu, filter the CI impl list to the implementations that actually run on TPU -- Helion (Pallas), the eager baseline, and torch.compile -- and drop Triton/CUDA-specific kernels (triton_*, liger_*, mamba_ssm, flash-linear- attention, cutlass/quack, etc.) which crash on TPU (no active Triton driver). Uses a positive allowlist of TPU-capable execution-mode prefixes, so new GPU-only backends are excluded by default; only a genuinely new TPU-runnable execution mode needs a central tuple edit. Kernels whose baseline is itself a GPU-only kernel (e.g. fp8_attention's triton baseline) are skipped with a warning. Non-TPU behavior is unchanged. Enables benchmark_tpu_bridge.yml to reuse the GPU nightly's auto-discovery (--list-impls) instead of hardcoding per-kernel --only lists.
meta-cla
Bot
added
the
CLA Signed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
--list-impls-for-benchmark-cireads the static dashboard metric map and prints the impls + baseline for a kernel, which the nightly uses to auto-derive--only. On--device tpu, this now filters to the implementations that actually run on TPU — Helion (Pallas), the eager baseline, and torch.compile — and drops Triton/CUDA-specific kernels (triton_*,liger_*,mamba_ssm, flash-linear-attention, cutlass/quack, …) which crash on TPU (no active Triton driver).Uses a positive allowlist of TPU-capable execution-mode prefixes, so new GPU-only backends are excluded by default; only a genuinely new TPU-runnable execution mode needs a central edit. Kernels whose baseline is itself a GPU-only kernel (e.g.
fp8_attention's Triton baseline) are skipped with a warning. Non-TPU behavior is unchanged.Inert until a TPU bridge nightly consumes it (a follow-up will add
benchmark_tpu_bridge.ymlon top of this). Enables that workflow to reuse the GPU nightly's auto-discovery instead of hardcoding per-kernel--onlylists.