GET_ROWS

This release updates the OpenVINO backend to version 2026.3.1, adds support for Whisper.cpp and Qwen3.5 on NPU, and introduces new operations. It also includes significant improvements to static shape handling, token-count independence, and chunked prefill for better performance and accuracy.

This release enhances the OpenVINO backend with support for Qwen3.5, MoE models, and various new operations. It also includes significant memory optimizations for GPU inference and fixes for accuracy and stability issues.

This release significantly enhances CUDA support for GET_ROWS by incorporating k-quant, i-quant, and mxfp4 dequantizers, ensuring all quantized GGML types leverage the direct device path. Bug fixes address incorrect gate applications for certain quant types.

Introduces the initial ET backend for ggml, significantly expanding the supported operations and optimizations. This release includes numerous bug fixes, performance enhancements, and new features for various kernels and backends.

This release introduces numerous new CANN operators and significant performance optimizations for existing operations like GLU and CROSS_ENTROPY_LOSS. Crucially, it fixes a major bug in the ACL graph cache that caused incorrect results when mixing F16 and BF16 tensors.

This release introduces support for conv2d kernels in ggml-webgpu shaders and includes numerous stability fixes related to f16 precision and packed integer handling across various operations. Internal code has been cleaned up, deprecated quant structs removed, and the WebGPU backend instance management improved.

This release addresses numerous quantization precision issues within ggml WebGPU, especially concerning f16 stability and NaN handling. It also improves backend lifecycle management for WebGPU and cleans up deprecated code.