Starstruk Logo MarkStarstruk

b10760

Pre-release

ggml-org/llama.cppb10760Sep 2, 2026by github-actions[bot]

AI Summary

Fixes critical bugs in the Qwen3-tts-0.6b model implementation, specifically addressing missing tensor handling and F16 overflow issues that caused numerical instability.

Key Highlights

  • Fixed Qwen3-tts-0.6b loading by making the proj_in projection optional.
  • Kept the code predictor FFN down projection in F32 to prevent overflow and NaN values.
  • Improved numerical stability for the 0.6B checkpoint variants.

Full Release Notes

<details open>

mtmd: Fix Qwen3-tts-0.6b (#28231)

* mtmd: load the qwen3-tts code predictor proj_in as optional

The talker and the code predictor share the hidden size on the 0.6B
checkpoints, so the reference builds no small_to_mtp_projection and
the conversion emits no tensor for it. The graph already falls back
to identity when the weight is missing, the loader now agrees.

* mtmd: keep the qwen3-tts code predictor ffn_down in F32

The code predictor carries a massive activation: its layer 2 FFN
intermediate peaks around 1.5e5, well past the 65504 ceiling of F16.
mul_mat casts its input to the weight type, so an F16 ffn_down turns
that peak into inf, the residual follows, and the next rms_norm yields
NaN. Reference forward in float32 gives 145109 against 145396 measured
in the graph.

</details>

**Website:**
- <https://llama.app>

**Attestations:**
- <https://github.com/ggml-org/llama.cpp/attestations/44677706>

**macOS/iOS:**
- [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-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/b10760/llama-b10760-bin-macos-x64.tar.gz)
- [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-xcframework.zip)

**Linux:**
- [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-x64.tar.gz)
- [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-arm64.tar.gz)
- [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-s390x.tar.gz)
- [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-vulkan-x64.tar.gz)
- [Ubuntu arm64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-vulkan-arm64.tar.gz)
- [Ubuntu x64 (ROCm 7.14)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-rocm-7.14-x64.tar.gz)
- [Ubuntu x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-openvino-2026.3.1-x64.tar.gz)
- [Ubuntu x64 (SYCL FP32)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-sycl-fp32-x64.tar.gz)
- [Ubuntu x64 (SYCL FP16)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-ubuntu-sycl-fp16-x64.tar.gz)

**Android:**
- [Android arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-android-arm64.tar.gz)

**Windows:**
- [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-cpu-x64.zip)
- [Windows arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-cpu-arm64.zip)
- [Windows arm64 (OpenCL Adreno)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-opencl-adreno-arm64.zip)
- [Windows x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-cuda-12.4-x64.zip) - [CUDA 12.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b10760/cudart-llama-bin-win-cuda-12.4-x64.zip)
- [Windows x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-cuda-13.3-x64.zip) - [CUDA 13.3 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b10760/cudart-llama-bin-win-cuda-13.3-x64.zip)
- [Windows arm64 (CUDA 13) (preview)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-cuda-13.4-arm64.zip) - [CUDA 13.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/b10760/cudart-llama-bin-win-cuda-13.4-arm64.zip)
- [Windows x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-vulkan-x64.zip)
- [Windows x64 (OpenVINO)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-openvino-2026.3.1-x64.zip)
- [Windows x64 (SYCL)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-sycl-x64.zip)
- [Windows x64 (ROCm 7.14)](https://github.com/ggml-org/llama.cpp/releases/download/b10760/llama-b10760-bin-win-rocm-7.14-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/b10760/llama-b10760-ui.tar.gz)