github llama.cpp · 3d ago · inference

llama.cpp b10831 adds TQ1_0 Vulkan support and fixes Metal fallback

from English

The llama.cpp b10831 release introduces TQ1_0 quantization support for the Vulkan backend, enabling operations such as matrix multiplication, mat-vec, dequantization, and get_rows. This update also corrects the Metal backend to properly decline unsupported TQ1_0 operations, ensuring they fall back to CPU execution.

Vulkan implementation packs TQ1_0 powers of 3 into a 32-bit constant to avoid register issues.
Shared decode helpers in types.glsl replace duplicated logic across various shader files.
A bug in the standalone dequant shader's workgroup denominator calculation is fixed.
Metal backend now rejects TQ1_0 for GET_ROWS and mat-mul, matching existing NVFP4 handling.
Test cases are trimmed and comments standardized to ASCII.

The Vulkan changes have been verified on AMD gfx1151 with all relevant backend operations passing tests.

Importance 1/3 Trust 1/3 llama.cpp Inference efficiency Local inference

Read original

github llama.cpp · 2d ago · 2 views

llama.cpp disables lazy tensor loading by default on iGPUs

The llama.cpp project has disabled lazy tensor loading by default for integrated GPUs (iGPUs) in release b10867. This change addresses a regression introduced in the previous "AUTO" lazy mode setting.

github llama.cpp · 2d ago · 3 views

llama.cpp b10864 fixes checkpoint eviction for hybrid models

The llama.cpp project released build b10864, which includes a fix for the checkpoint spacing eviction logic in `create_checkpoint()`. Previously, the eviction rule kept the oldest checkpoint and erased later ones within `checkpoint_min_step`, causing prompts shorter than that threshold to drop the checkpoint needed for resumption. This forced hybrid and recurrent models to unnecessarily re-prefill from the previous checkpoint.

github llama.cpp · 3d ago · 6 views

llama.cpp b10844 adds Vulkan DeepSeek-V4 hyper-connection fused ops

The llama.cpp project released version b10844, introducing support for DeepSeek-V4 hyper-connection fused operations (DSV4_HC_COMB/PRE/POST) in the Vulkan backend. This update brings Vulkan to parity with CUDA and Metal regarding these specific optimizations.

github llama.cpp · 3d ago · 4 views

llama.cpp b10842 adds gfx90c HIP support

The llama.cpp project released build b10842, which introduces support for the gfx90c architecture via HIP.

media Hugging Face Forums · 3d ago · 8 views

AI Model Radar launches with measured hardware fit for open models

A new site called AI Model Radar helps users determine which open models actually fit their GPU or Mac by using measured data rather than estimates. The platform calculates weight sizes based on real published GGUF files and computes context cache layer-by-layer from model configurations.