github llama.cpp · 8h ago · inference

llama.cpp b10655 adds LIGHTNING_INDEXER support for Deepseek V4 ops on Vulkan

from English

The llama.cpp project released build b10655, which introduces support for the LIGHTNING_INDEXER operation within the Vulkan backend to enable Deepseek V4 operations.

Added the LIGHTNING_INDEXer op to the Vulkan backend.
Updated lightning_indexer.comp and ggml-vulkan.cpp to move 128-lane dot-product reduction from a shared-memory tree to subgroupAdd.
Included cleanup changes such as skipping bounds checks and reverting previous FA_K_ONLY modifications.
Restored interleaved K/V buffer ordering and removed FA_K_ONLY logic.

This update provides Vulkan users with the capability to run Deepseek V4 operations using the new LIGHTNING_INDEXER implementation.

Importance 1/3 Trust 1/3 llama.cpp Hugging Face Inference efficiency

Read original

github llama.cpp · 1h ago · 1 view Live

llama.cpp adds Qwen3.8-Flash-Next (qwen4exp) architecture support

The llama.cpp project has added support for the Qwen3.8-Flash-Next (qwen4exp) model architecture, implementing its specific components including hyper-connections, gated delta nets, Mixture of Experts, and PLE n-gram hash embeddings.

github llama.cpp · 8h ago · 3 views

llama.cpp b10653 adds TENSOR_READ_LAZY for model loading

The llama.cpp project has released build b10653, introducing the TENSOR_READ_LAZY feature to its model loader. This change includes the addition of the TENSOR_GET_ROW_LAZY function and a corresponding --tensor-read-lazy command-line flag.

github llama.cpp · 1d ago · 6 views

llama.cpp b10639 fixes Vulkan warptile warp size assumption

The llama.cpp project released build b10639, which includes a fix for the Vulkan backend where warptiles previously assumed warp sizes were less than or equal to 64. This change clamps the warp size to work around issues with larger warps.

media Hugging Face Forums · 1d ago · 4 views

EdgeFirst publishes Rust pipeline validating YOLO models on 20 edge platforms

EdgeFirst has published the EdgeFirst Model Zoo on Hugging Face, featuring YOLOv5, YOLOv8, YOLO11, and YOLO26 models validated across 20 distinct edge platforms. The validation relies on a single Rust pipeline that handles zero-copy DMA pre-processing, model decoding, and post-processing directly on the target hardware.

lab Hugging Face Blog · 2d ago · 1 view

Papers with Code uses Hugging Face services for hybrid paper search

Papers with Code has implemented a hybrid search system combining keyword and semantic retrieval to help users find relevant AI research papers. The architecture leverages Hugging Face Jobs, Storage Buckets, and Inference Endpoints to manage the embedding pipeline and serve low-latency queries.