llama.cpp 修復 Metal 後端 mul_mat_id 重複 id 錯誤
llama.cpp 建置版本 b11569 修復了 Metal 後端 mul_mat_id 在 id 重複時輸出錯誤、丟行乃至溢位到下一個 expert 切片的問題。新實現先做一遍預掃描，為每個 expert 建立緊湊的 (token, slot) 條目列表，並借鑑 Vulkan 方案讓 map0 額外輸出 tile 描述符，kernel_mul_mm_id 按描述符處理單個 tile，重複 id 由此被隱式處理。
MoE 模型在 Metal 後端遇到重複 expert id 時會算錯或丟行，此次修復同時把 grid 縮小 8–16 倍、pp 提升約 2%，對在 Mac 上用 llama.cpp 跑 MoE 模型的人最有用。
原標題：b11569
閱讀原文
| 評分 | 46 / 58（平均 52，門檻 60） |
|---|---|
| 狀態 | 未入選 |
原文
metal : fix mul_mat_id with duplicate ids ( #30373 )
metal : fix mul_mat_id limits and kernel name
metal : fix mul_mat_id with duplicate ids
kernel_mul_mm_id_map0 assumed at most one match per token and a fixed
capacity of ne21 entries per expert, which produced wrong or dropped
output rows and could overflow into the slice of the next expert when
the ids contain duplicates.
build a compact per-expert list of the (token, slot) entries instead:
pass 1 counts the entries of every expert, thread 0 computes the
offsets of the expert slices (tpe now holds [base, count])
pass 2 writes the entries with one unconditional store per token
(conditional stores of divergent lanes are very slow), the dummy
value of the tokens without a match lands on a padding slot of the
slice that is never read - duplicates are handled by a rare branch
kernel_mul_mm_id reads the base/count of its expert and loops over
the entry tiles that do not fit into the grid
The ids buffer is sized as ne20*ne21 + ne02 (entries + padding).
883/883 test-backend-ops MUL_MAT_ID tests pass, llama-bench is within
run-to-run variance of the previous behavior.
Assisted-by: pi:llama.cpp/MiMo-V2.6-Flash-MOPD
metal : make mul_mat_id map0 ne02 a function constant
use FOR_UNROLL for the id loops of kernel_mul_mm_id_map0 and turn ne02
into the function constant FC_mul_mm_id_map0_ne02 (FC_MUL_MM + 7), so
that the threadgroup offsets of cnt/bas/sids and the prefix-sum bound
fold at compile time.
the pipeline name now encodes ne02 (same convention as the ssm_conv
nc/silu constants) and the map0 pipeline is compiled with the constant
value set. ne02 is no longer part of ggml_metal_kargs_mul_mm_id_map0 -
the dispatch grid still comes from the tensor ne[2] on the host.
no measurable perf change: 883/883 test-backend-ops MUL_MAT_ID tests
pass, op-level and llama-bench results are within run variance.
Assisted-by: pi:llama.cpp/MiMo-V2.6-Flash-MOPD
cont : clean-up + TODO
metal : dispatch mul_mat_id entry tiles from the prepass
follow the approach of the vulkan implementation ( #29998 ): instead of
launching a grid over experts x ne21 entry tiles and letting most of
the threadgroups early exit, kernel_mul_mm_id_map0 now also emits a
row map - the total tile count plus one (expert, first entry, valid
rows) descriptor per 32-entry tile that has work.
kernel_mul_mm_id is launched with a static upper bound on the number
of tiles (full tiles over all entries plus at most one partial tile
per expert), reads its descriptor and processes a single tile - the
tail early exits. This removes the round loop that was needed to cover
duplicate ids past the grid (register pressure: th_max 768 -> back to
the plain mul_mm level) and duplicates are now handled implicitly,
since they only produce more descriptors within the bound.
The grid shrinks by ~8-16x on typical shapes (e.g. 128 experts,
8 used, 512 tokens: 128 tiles instead of 2048 entry tiles, of which
94% only launched to exit) and ne21 is no longer needed by the mm
kernel kargs.
883/883 test-backend-ops MUL_MAT_ID tests pass. llama-bench
(Qwen3.6-35B-A3B Q8_0) is now faster than master on all pp sizes
(pp2048/4096/8192 +2%, pp512 +0.5%), tg32 unchanged.
Assisted-by: pi:llama.cpp/MiMo-V2.6-Flash-MOPD
cont : add smem assert
Website:
https://llama.app
Attestations:
https://github.com/ggml-org/llama.cpp/attestations/55057952
macOS/iOS:
macOS Apple Silicon (arm64)
macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED
macOS Intel (x64)
iOS XCFramework
Linux:
Ubuntu x64 (CPU)
Ubuntu arm64 (CPU)
Ubuntu s390x (CPU)
Ubuntu x64 (Vulkan)
Ubuntu arm64 (Vulkan)
Ubuntu x64 (CUDA 12) - CUDA 12.8 libraries
Ubuntu x64 (CUDA 13) - CUDA 13.4 libraries
Ubuntu arm64 (CUDA 13) - CUDA 13.4 libraries
Ubuntu x64 (ROCm 10.0)
Ubuntu x64 (OpenVINO)
Ubuntu x64 (SYCL FP32)
Ubuntu x64 (SYCL FP16)
Linux arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU) - setup guide
Android:
Android arm64 (CPU)
Android arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU) - setup guide
Windows:
Windows x64 (CPU)
Windows arm64 (CPU)
Windows arm64 (OpenCL Adreno)
Windows x64 (CUDA 12) - CUDA 12.4 DLLs
Windows x64 (CUDA 13) - CUDA 13.4 DLLs
Windows arm64 (CUDA 13) - CUDA 13.4 DLLs
Windows x64 (Vulkan)
Windows arm64 (Vulkan)
Windows x64 (OpenVINO)
Windows x64 (SYCL)
Windows x64 (ROCm 10.0)
openEuler:
DISABLED
openEuler x86 (310p)
openEuler x86 (910b, ACL Graph)
openEuler aarch64 (310p)
openEuler aarch64 (910b, ACL Graph)
UI:
UI
相關報導
llama.cpp releases● 精選AI 評分74
llama.cpp 修復 Metal 後端 MUL_MAT+ADD 融合錯誤
llama.cpp 合併 PR #30100，修復 Metal 後端 MUL_MAT+ADD 融合中殘差運算元的選擇錯誤：當 ADD 的兩個運算元都是矩陣乘法輸出（x = W1@u + W2@v）時，舊邏輯會選中自身從未寫入的輸出緩衝區，導致結果錯誤。Clef 決策模型因此在 Metal 上機率趨近均勻，billing 輸出 0.28，而 CPU 後端為 0.977，且同一檔案在 CPU 上結果正確，說明並非量化問題。
llama.cpp releasesAI 評分46
llama.cpp 新版最佳化 Vulkan 上 MoE 重複專家的矩陣乘
llama.cpp 發布 b11554，Vulkan 後端改進了 MoE 推理中 mul_mat_id 對重複專家 id 的處理：不再逐個迴圈，而是在預計算階段統一處理。原本受限的 hoist row ids 最佳化現在預設開啟，會生成緊湊的 tile 描述列表、按需發射多個 tile，從而收緊 workgroup 數量上界，尾部執行緒可直接提前退出。
llama.cpp releasesAI 評分36
llama.cpp 修復 Vulkan 後端 RDNA4 矩陣向量調優
llama.cpp 發布 b11389 版本，修復了 Vulkan 後端在 RDNA4 架構上的矩陣向量運算調優問題（PR #29934）。該版本繼續提供覆蓋 macOS、Linux、Windows、Android 等平台的預編譯二進位制，包括 Vulkan、CUDA、ROCm、SYCL 等後端，其中 macOS Apple Silicon 的 KleidiAI 啟用版和 openEuler 建置被停用。
llama.cpp releasesAI 評分37
llama.cpp 發布 b11390，修復 CUDA MMQ 記憶體錯誤
llama.cpp 發布建置版本 b11390，主要修復了 CUDA 後端在 n_expert 遠大於 n_ubatch 時的 MMQ 記憶體故障（#29941）。
llama.cpp releasesAI 評分43
llama.cpp b11446 修復 Metal 量化 flash attention 執行緒組記憶體超額
llama.cpp 發布 b11446，Metal 後端修復量化 flash attention 中 threadgroup memory 超額佔用的問題，對應 PR #29340。