fix: Bump GGML to fix broken quantized matmul on Vulkan#27
Open
richiejp wants to merge 1 commit into
Open
Conversation
Updates the ggml submodule from 5cecdad (2026-02-07) to ggml master 3af5f576 (2026-06-12, v0.15.1). The old ggml's Vulkan quantized mat-vec (MMVQ / integer-dot path) returns wrong results on Intel Arc/ANV, which shows up as garbage/"broken record" TTS output for quantized models. The newer ggml fixes the MMVQ correctness on ANV: test-backend-ops MUL_MAT now passes all quantized types (q4_0/q4_1/q4_K/q5_0/q5_1/q5_K/q8_0, incl. n=1 mat-vec) on an A770 against the CPU reference. qwen3-tts.cpp compiles cleanly against the new ggml (no API drift) and the CLI runs. flake.nix adds a dev shell (cmake, ninja, vulkan-loader/headers, shaderc, glslang) so the CPU+Vulkan build has all deps. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Author
|
Ping @predict-woo |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This fixes a "broken-record" bug on my Intel A770 simply by updating GGML.
Also it adds a Nix flake which I use to compile the project. I can remove it
if you don't want a Nix dev env adding.
Updates the ggml submodule from 5cecdad (2026-02-07) to ggml master
3af5f576 (2026-06-12, v0.15.1). The old ggml's Vulkan quantized mat-vec
(MMVQ / integer-dot path) returns wrong results on Intel Arc/ANV, which
shows up as garbage/"broken record" TTS output for quantized models. The
newer ggml fixes the MMVQ correctness on ANV: test-backend-ops MUL_MAT now
passes all quantized types (q4_0/q4_1/q4_K/q5_0/q5_1/q5_K/q8_0, incl. n=1
mat-vec) on an A770 against the CPU reference.
qwen3-tts.cpp compiles cleanly against the new ggml (no API drift) and the
CLI runs.
flake.nix adds a dev shell (cmake, ninja, vulkan-loader/headers, shaderc,
glslang) so the CPU+Vulkan build has all deps.
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com