Starstruk Logo MarkStarstruk

b9866

ggml-org/llama.cppb9866Jul 3, 2026by github-actions[bot]

AI Summary

Enables topk-moe fusion for 288 experts in CUDA, previously only supported for power-of-2 counts or 576. This improves decode performance for large MoE models.

Key Highlights

  • Enables topk-moe fusion for 288 expert models (e.g., Step-3.7-Flash)
  • Adds missing template instantiation for 288 expert counts
  • Measured ~2.4% decode speedup at shallow context
  • Fusion only affects decode routing, not prompt processing

New Features

  • CUDA optimization for 288-expert Mixture of Experts models

Full Release Notes

<details open>

cuda: enable topk-moe fusion for 288 experts (#25267)

* cuda: enable topk-moe fusion for 288 experts

The topk-moe fusion only accepted power-of-2 expert counts (or the
special-cased 576), so models with 288 experts (e.g. Step-3.7-Flash)
fell back to the unfused per-layer routing chain: softmax/sigmoid,
argsort, get_rows, sum_rows, div, clamp, scale. At batch size 1 that
is ~330 extra tiny graph nodes per token.

288 is a multiple of the warp size, so the existing kernel already
handles it; this adds the missing template instantiation and accepts
288 in the eligibility check.

Measured on gfx1151 with Step-3.7-Flash IQ4_XS (llama-bench,
-b 4096 -ub 4096 -fa 1 -dio 1 -ctk q8_0 -ctv q8_0; machine idle,
before/after paired so pp4096 stays matched as a load control):

  test            | before         | after
  ----------------+----------------+----------------
  pp4096          | 460.99 ± 0.45  | 462.47 ± 0.34   (unchanged)
  tg128           |  19.10 ± 0.04  |  19.56 ± 0.03   (+2.4%)
  tg128 @ d30000  |  12.68 ± 0.04  |  12.69 ± 0.03   (unchanged)

Prompt processing is unaffected (the fusion only touches decode
routing). The decode gain is ~+2.4% at shallow context and fades with
depth: by 30k tokens each step is attention-bound over the KV cache,
so removing the fixed routing overhead is no longer visible.

Assisted-By: Claude Fable 5 <noreply@anthropic.com>

* Update tests/test-backend-ops.cpp

Co-authored-by: Oliver Simons <osimons@nvidia.com>

* Add comment for case 288 in topk-moe.cu

---------

Co-authored-by: Oliver Simons <osimons@nvidia.com>

</details>

**macOS/iOS:**
- [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-macos-arm64.tar.gz)
- macOS Apple Silicon (arm64, KleidiAI enabled) [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23780)
- [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-macos-x64.tar.gz)
- [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-xcframework.zip)

**Linux:**
- [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-x64.tar.gz)
- [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-arm64.tar.gz)
- [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-s390x.tar.gz)
- [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-vulkan-x64.tar.gz)
- [Ubuntu arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-vulkan-arm64.tar.gz)
- [Ubuntu x64 (ROCm 7.2)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-rocm-7.2-x64.tar.gz)
- [Ubuntu x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-openvino-2026.2.1-x64.tar.gz)
- [Ubuntu x64 (SYCL FP32)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-sycl-fp32-x64.tar.gz)
- [Ubuntu x64 (SYCL FP16)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-ubuntu-sycl-fp16-x64.tar.gz)

**Android:**
- [Android arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-android-arm64.tar.gz)

**Windows:**
- [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-cpu-x64.zip)
- [Windows arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-cpu-arm64.zip)
- [Windows arm64 (OpenCL Adreno)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-opencl-adreno-arm64.zip)
- [Windows x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-cuda-12.4-x64.zip) - [CUDA 12.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b9866/cudart-llama-bin-win-cuda-12.4-x64.zip)
- [Windows x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-cuda-13.3-x64.zip) - [CUDA 13.3 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b9866/cudart-llama-bin-win-cuda-13.3-x64.zip)
- [Windows x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-vulkan-x64.zip)
- [Windows x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-openvino-2026.2.1-x64.zip)
- [Windows x64 (SYCL)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-sycl-x64.zip)
- [Windows x64 (HIP)](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-bin-win-hip-radeon-x64.zip)

**openEuler:**
- [DISABLED](https://github.com/ggml-org/llama.cpp/pull/23705)
- openEuler x86 (310p)
- openEuler x86 (910b, ACL Graph)
- openEuler aarch64 (310p)
- openEuler aarch64 (910b, ACL Graph)

**UI:**
- [UI](https://github.com/ggml-org/llama.cpp/releases/download/b9866/llama-b9866-ui.tar.gz)