fix(ci): bump pinned Spyre RPMs to unbreak fp8 _scaled_mm compile - #818
Closed
sducouedic wants to merge 1 commit into
Closed
fix(ci): bump pinned Spyre RPMs to unbreak fp8 _scaled_mm compile#818sducouedic wants to merge 1 commit into
sducouedic wants to merge 1 commit into
Conversation
The fused _scaled_mm fp8 kernel fails to compile under dxp_standalone on the current pins. The RPM set here compiles it; ibm-flex and ibm-spyre-comms are the only packages moving commit. Signed-off-by: Sophie du Couédic <sducouedic@users.noreply.github.com>
Contributor
|
👋 Hi! Thank you for contributing. We also recommend installing prek and configuring it to check your code before every local commit. |
HarikrishnanBalagopal
added a commit
to HarikrishnanBalagopal/spyre-inference
that referenced
this pull request
Sep 9, 2026
HarikrishnanBalagopal
added a commit
to HarikrishnanBalagopal/spyre-inference
that referenced
this pull request
Sep 9, 2026
…ply torch-spyre#818) Signed-off-by: Harikrishnan Balagopal <harikrishmenon@gmail.com>
This was referenced Sep 9, 2026
6 tasks
joerunde
added a commit
that referenced
this pull request
Sep 10, 2026
## Description Bumps `torch-spyre` from `7697811` to `c3d949a` — all 13 upstream commits since the previous pin. The tip of `main` compiled cleanly against the currently-installed RPMs (no bisect needed). Also refreshes the full `uv.lock` (`uv lock --upgrade`); `vllm` is held at its pinned git tag (`v0.28.0`) and intentionally **not** upgraded. NB: We had a prior bump PR in today (#818) that was attempting to clear out a regression. That didn't work so we instead disabled tests (#821), but we still do need to keep upgrading ### Notable upstream changes in this range - **Enable sliding window attention** (#3405) — new attention path. - **Lx dump restore** (#3879) — dump/restore support. - **Demote non-critical warnings to info/debug level** (#4374) and **remove `_get_env_bool` / legacy `SPYRE_INDUCTOR_LOG` helpers** (#4241) — logging cleanup and removal of legacy env-var helpers. - **fix attention view error** (#4377) and **codegen: point `maskingConstId_` at samv-maskvalue, not constant 0** (#4400) — attention/masking correctness fixes. - **fix(inductor): preserve offsets in coarse read copies** (#4376) — inductor lowering fix (tip). ### Transitive dependency refresh 5 transitive packages moved via `uv lock --upgrade`: transformers 5.16.1 → 5.17.0, openai 3.9.0 → 3.11.0, xgrammar 0.2.3 → 0.2.6, multidict 6.7.1 → 6.8.0, filelock 3.32.5 → 3.32.6. `vllm` (v0.28.0) and `torch-spyre` unchanged (pinned). ### Installed `ibm-*` packages on the build host ``` ibm-aiu-toolbox-e2e-2.0.0-0.main.1+28.47d9b91_101.el10.x86_64 ibm-deeptools-2.0.0-0.main.1+2429.561f418_335.el10.x86_64 ibm-deeptools-devel-2.0.0-0.main.1+2429.561f418_335.el10.x86_64 ibm-flex-2.0.0-0.main.1+553.8581a91_384.el10.x86_64 ibm-flex-devel-2.0.0-0.main.1+553.8581a91_384.el10.x86_64 ibm-libaiupti-2.0.0-0.main.1+27.e289f7d_13.el10.x86_64 ibm-senlib-core-2.0.0-0.main.1+268.c58bbc0_251.el10.x86_64 ibm-senlib-dd2-2.0.0-0.main.1+268.c58bbc0_251.el10.x86_64 ibm-senlib-headers-2.0.0-0.main.1+268.c58bbc0_251.el10.x86_64 ibm-spyre-comms-1.0.0-0.main.1+146.a445334_182.el10.x86_64 ibm-spyre-comms-devel-1.0.0-0.main.1+146.a445334_182.el10.x86_64 ``` `spyre-rpms.lock` bumped to match: flex `7c58aa7` → `8581a91` and spyre-comms `d3875d2` → `a445334` (new commits); deeptools/senlib/aiu-toolbox/libaiupti advanced to newer builds of the same commit. No downgrades. ## Test Plan - [x] `uv lock` resolves cleanly to `c3d949a` - [x] `uv lock --upgrade` refreshes all transitives (vllm held at its git tag) - [x] `uv sync --frozen` builds the torch-spyre C++ extension successfully - [x] Smoke test (`test_basic_model_load`) passes locally - [ ] Full CI suite passes (pushed for CI validation) - [x] `spyre-rpms.lock` updated — no downgrades (commit-count check passed); resolves + downloads on all arches **Reviewer note:** when pulling this branch onto an existing checkout, `rm -rf /tmp/torchinductor_*` before running tests — the cache bakes in references to internals that were renamed/removed across the bump. --------- Signed-off-by: Joe Runde <joe@joerun.de>
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.
Description
tests/test_fp8_linear.py::TestSpyreFp8LinearKernel::test_scaled_mm_apply*[1]fails onmainwithdxp_standaloneexiting 1 on the fused_scaled_mm + quantize_fp8_with_scalekernel. It is not a code regression: the same commit (d97de1f) passed on 09-08 evening and started failing on 09-09 morning, and the CI runner image changed across that boundary. This bumpsspyre-rpms.lockto the RPM set installed on my dev pod, which compiles the kernel.ibm-flexandibm-spyre-commsare the only packages moving commit; the rest are build-number-only bumps of the same commits.Which package carries the fix is not isolated — every pin here is newer than the current lock, so this shows the failure is gone in the newer set rather than pinpointing the cause.
Related Issues
Relates to the shared failure on #812, #814, #815.
Test Plan
pytest tests/test_fp8_linear.py -m "not upstream"withSPYRE_KERNEL_CACHE=1and a cleared inductor cache, against the pinned torch-spyre rev7697811— 17 passed, including the two cases failing in CI.bash format.sh— no changes to this diff.[packages]commit and no commit-count downgrades.Checklist
bash format.sh)Signed-off-by:line (DCO compliance)