vulkan: fix build-breaking merge artifact; versioning scheme; CI polish - #35
Merged
Conversation
The inkling merge resolution dropped the closing brace of the turbo-type guard in ggml_backend_vk_device_supports_op (case GGML_OP_SET_ROWS): the inner %128 head-dim check closed, but the outer turbo if never did. Every subsequent function body nested one level deeper and both Vulkan builds (linux, windows) failed with a cascade of 'function-definition is not allowed here' errors from 066cc29 onward. Also corrects the wrong MERGE_NOTES claim that Vulkan lacks TURBO_WHT — fork kernels for TURBO_WHT / turbo set_rows / GATED_DELTA_NET exist and predate the merge; only turbo3 FA SPIR-V and banded/lightning kernels are genuinely absent. dev-build concurrency now serializes pushes per ref.
github-actions
Bot
added
documentation
The server exposes llama_build_info() as the OpenAI-compatible system_fingerprint and clients parse it (server unit tests assert it starts with 'b'; the Atomic-Chat plugin gates features on the b-number). The turboquant semver stays available via turboquant_version() and the --version output. Caught by unit/test_chat_completion.py on the dev staging run — exactly what the channel is for.
…ase notes Version/tag format is now b<upstream>-X.Y.Z (e.g. b10018-1.0.0): the upstream llama.cpp build the fork is based on, plus the fork's own semver (minor for features like new turbo-ops backends, patch for fixes). The upstream part is derived from 'git rev-list --count' of the merge-base with the upstream mirror branch and is updated by hand in upstream-sync PRs. --version now reports 'version: b10018-1.0.0 (build N, commit sha)'. All release/dev-channel notes rewritten in English to match the project's existing release style.
worthant
changed the title
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.
Summary
Pilot run of the new dev → master flow. Three things bundled:
1. Vulkan build fix
The inkling merge resolution (
066cc29) dropped the closing brace of the turbo-type guard inggml_backend_vk_device_supports_op(caseGGML_OP_SET_ROWS,ggml/src/ggml-vulkan/ggml-vulkan.cpp). Every subsequent function nested one level deeper → cascade offunction-definition is not allowed hereerrors → both Vulkan builds (linux-x64, windows-x64) red since the merge; the066cc29release set shipped without Vulkan binaries. Fix: one brace.Also corrects MERGE_NOTES.md: Vulkan DOES carry fork kernels for TURBO_WHT / turbo set_rows / GATED_DELTA_NET (pre-merge code, still present); only turbo3 FA SPIR-V and banded/lightning kernels are genuinely absent.
2. Versioning
TURBOQUANT_VERSIONfile, format<upstream-base>-<fork-semver>(b10018-1.0.0): upstream llama.cpp build the fork is based on + fork semver--version→version: b10018-1.0.0 (build N, commit sha)llama_build_info()kept in upstreamb<N>-<sha>format on purpose — it is the OpenAIsystem_fingerprintand clients parse it (caught byunit/test_chat_completion.pyon the first CI round of this PR)scripts/turboquant-release.sh+release-turboquant.yml: tagb10018-X.Y.Z→ one release with ALL platform archives3. CI polish
devserialize instead of racing fordev-latestVerification
ggml-vulkan.cpprestored (file-level depth 0, all switch cases at expected depth)windows-x64 (vulkan)andlinux-x64-vulkanchecks — first time since the inkling merge