sync: merge upstream/master (3 commits, zero TQ3 impact) - #72
Merged
Conversation
* Switch ROCm from 7.2.1 to 7.14 ROCm 7.14 is the first production release using TheRock build system. It can be installed using multi-arch deliverables from wheels, debs, rpms, tarballs or runfiles. Adjust ROCm targets for Linux and Windows to use this instead. * ci: switch all other Windows ROCm jobs to ROCm 7.14 wheels Move the shared windows-setup-rocm composite action from the HIP SDK PRO Edition installer to the multi-arch ROCm wheels (rocm[libraries,devel]). The wheel-install logic that previously lived inline in release.yml is now in the shared action, and both build-cache.yml and release.yml call it. Also migrate the build-cuda-windows.yml hip job to the same wheel-based layout (cache path/key, rocm-sdk environment setup, llvm/bin compiler paths) so it keeps working after the action's contract changed; drop its now-unused ROCm 7.2.1 rocWMMA download and stale include path.
… review fix) No file overlap with fork-specific TQ3_4S/dflash/muse-glimmer work — clean merge. Closes the gap opened since the 2026-08-10 history graft (dd1ea52). # Conflicts: # .github/workflows/build-cuda-windows.yml # .github/workflows/release.yml # include/llama.h
…ped PR The declaration alone would have been orphaned: its implementation depends on a llama_sampler_i::copy_state interface field that is part of upstream PR ggml-org#25532 ('multi-output backend sampling', 24 files, ~1000 lines) - a squashed feature that predates this fork's real base and is out of scope for a 'zero TQ3 impact' sync. copy_state does not exist anywhere in this tree; a half-port would not compile. Nothing in this codebase currently calls llama_sampler_copy, so dropping the declaration is safe. The full feature belongs in the larger upstream content-catch-up, tracked separately. Co-Authored-By: Claude Fable 5 <noreply@anthropic.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
5 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.
Summary
Closes the small gap that opened between origin/main and upstream/master since the 2026-08-10
history graft (
f75b55413, merge-basedd1ea5243). Merges 3 upstream commits, none of whichtouch TQ3_4S/dflash/muse-glimmer code:
030ebb558Address review comment of PR 25532 (llama.h comment cleanup)689e227dbopencl: transpose the K tile in local memory for FA prefill kernels0666ad2b2ci: target ROCm 7.14 for build and releaseConflicts resolved (3, all clean — reasoning below)
include/llama.h: dropped upstream'sllama_sampler_copydeclaration rather than takingit. It's part of upstream PR Backend sampling multi output ggml-org/llama.cpp#25532 ("multi-output backend sampling", 24 files touched,
src/llama-sampler.cpp+443 lines) — a large squashed feature that predates this fork's realcontent base and is unrelated to the 3 commits actually being merged here. Its implementation
depends on a
llama_sampler_i::copy_stateinterface field that does not exist anywhere in thistree; a bare declaration would have been an orphaned, unimplemented public API. Verified nothing
in this codebase currently calls
llama_sampler_copy. The full feature belongs with the largerupstream catch-up (see below), not this scoped sync.
.github/workflows/build-cuda-windows.yml: kept fork-specific-DGGML_HIP_ROCWMMA_FATTN=ON..github/workflows/release.yml: kept fork-specificwindows-hip:release job (no upstreamequivalent exists; verified no name collision elsewhere in the file).
Verification
git diff --stat 86bd2bacf HEAD -- ggml/src/ggml-cuda/{mmq.cu,vecdotq.cuh,quantize.cu}— empty,TQ3_4S CUDA kernels untouched by this merge.
yaml.safe_load.Known follow-up (not in scope here, flagged for the operator)
Ancestry-wise this now makes origin/main current with upstream/master. Content-wise it is not:
the graft's tree was deliberately kept byte-identical to the pre-graft fork state, so it predates
dd1ea5243by ~2 months of real upstream development that was never content-merged, justancestry-spliced. Measured:
git diff --stat dd1ea5243 86bd2bacf= 894 files, +25036/-62272 lines(819 files / +15845/-34233 excluding vendor+docs). Biggest concentration is
tools/ui/(web UI,~40%+ of file churn) and
ggml/src/(~12%, the TQ3-relevant area). The dropped-herellama_sampler_copy/multi-output-backend-sampling feature is one piece of that larger gap. Thisis a separate, much larger undertaking than this PR and needs an explicit decision on approach —
not attempted here.
Co-Authored-By: Claude Fable 5 noreply@anthropic.com