Search: “routing”

server: fix router eviction races with the existing queue (#29217) server: route every model load through the queue A model loaded by the fast path has no queue entry, so tick() evicts it at its LOADE

opencl: choose the MoE expert matmul by batch size for speculative decoding/MTP (#27637) opencl: gate the prebuilt q4_0 MoE GEMM on routing count opencl: stop writing zeros into the padded MoE activat

Overview llama.cpp 0.4.1 adds Maple 20B-A1B, Tencent Hy 4, and Spark2.5 support, improves JSON schema handling, chat parsing, logging, and server child-process management, and updates ggml to v0.24.0.

CUDA: size routed MoE MMQ N-tiles from typical expert width on RDNA3 (#28552) Recreated from #24546 Co-authored-by: Carl Philipp Klemm carl@uvos.xyz CUDA: pick MMQ tile size against ncols_opt set on t

model: add NVIDIA Nemotron-3-Puzzle-75B-A9B (NemotronHPuzzle) support (#25444) hparams: add per-layer n_ff_exp/n_expert_used arrays with scalar-or-array loading G1/G2 infrastructure for variable-per-l

sycl: use TILE for quantized KV decode on BMG (#26689) Route quantized KV decode to TILE on Xe2 (BMG) only, keep VEC on other archs until validated there. Website: https://llama.app Attestations: http

model: add Kimi-K3 text model (#26185) model: add Kimi-K3 text model Hybrid KDA (linear) + MLA (full) attention as in Kimi-Linear-48B, plus five things that architecture does not have: cross-layer res

opencl: route large q6_K lm_head to the flat GEMV (#26427) add a direct size condition for large weights; the original dimension condition is insufficient -- q6_K lm_head for gemma-4 E2B has [1536, 26

llama.cpp release b9453 added support for EXAONE 4.5 model implementations with vision capabilities, including markers, projector paths, and routing through a Qwen2.5-VL-style encode path with window

B9286 introduces support for HybridDNATokenizer as its own vocabulary type (LLAMA_VOCAB_TYPE_HYBRIDDNA), enabling proper tokenization of DNA sequences by routing raw text through a DNA-aware splitter

ggml-openvino: update to 2026.4.1, optimize performance, expand ops, improve device listing. (#29852) ggml-openvino : Qwen3.5 MoE perf (#312) Squash of ravi9#312: ggml-openvino: add detailed inference

Overview llama.cpp 0.4.0 adds initial Qwen3.8-Flash-Next and Nemotron-3-Puzzle support, on-demand tensor reading, per-slot server context limits, video input options, and a ggml update to 0.23.0 with

opencl: use a better matmul path on two Adreno GPU generations (#27640) opencl: default the Adreno xmem F16xF32 GEMM on for X2E kernel_mul_mm_f16_f32_l4_lm is the slowest matmul this backend has on Ad

rpc : turn GGML_RPC_DEBUG into a verbosity level and add logs GGML_RPC_DEBUG is now parsed as a number: 0/unset disables debug logs, 1-3 emit increasingly detailed output (events, per-command trace, t

Overview llama.cpp v0.6.0 introduces the new llama_batch_ext extended batch API (with llama_process) for mixed token/embedding inputs and MTP/deepstack state embeddings, adds support for the GLM-5.3-F

add GLM-5.3-Flash (GLM5-Next) support (#27773) Rebase GLM-Next support onto master, and migrate to llama-memory-hybrid-idx Add initial MTP support Merge branch optimizations. Reduce allocated compute

llama: llama_prefetch_rows (#29599) llama: llama_prefetch_rows llama: support row prefetch on Windows Apply the Windows port contributed by @praneshgo unchanged. Source: #29599 (comment) avoid exposin

glm5-next: give dead indexer slots unique scatter rows (#29745) The sparse indexer mask is built with a set_rows scatter. Padded pools, absent sequences and missing tail cells all pointed to the same

Decision models Ollama now supports decision models through /v1/systemone, based on TypeSafe’s Jev API. Decision models return choices, probabilities, and scores instead of text. Use them for tasks su

tests : refactor test-recurrent-state-rollback (#29426) tests : use llama_context_ptr in test-recurrent-state-rollback Replace raw llama_context pointers with llama_context_ptr and drop the manual lla

vulkan : fuse UNARY(GELU|SIGMOID|SILU|SOFTPLUS) + MUL (#27220) vulkan : fuse UNARY(SIGMOID|SILU|SOFTPLUS) + MUL vulkan : fuse UNARY(SIGMOID|SILU|SOFTPLUS) + MUL implement fusion in unary.comp behind U

model: add Qwen3.8-Flash-Next (qwen4exp) (#27742) gguf: add qwen4exp (Qwen3.8-Flash-Next) arch and converter Adds the GGUF-side plumbing for HF model_type qwen4_exp: MODEL_ARCH.QWEN4EXP plus tensors f

ggml-hexagon: add HTP unary ops for ABS and LOG (#27786) Add HVX-accelerated implementations for GGML_OP_LOG and GGML_UNARY_OP_ABS on the HTP backend. Register HTP_OP_UNARY_ABS and HTP_OP_UNARY_LOG in

metal: per-op source split + parallel compile (#26561) metal : per-op source split + parallel compile (#24021) preliminary extract common header op source split split metallib into 8 libs && load in p

CUDA: adding switch points per HW and quant type to tune the mvq->MMQ decode crossover (#26079) CUDA: runtime GGML_CUDA_MMVQ_MAX to tune the mvq->MMQ decode crossover Add a runtime override of the mul

tests: add model resolution test on synthetic repo listings (#26172) tests: add model resolution test on synthetic repo listings Include download.cpp and arg.cpp inside a namespace with hf_cache monke

SYCL: add oneMKL GEMM flash attention for XMX-accelerated prompt proc… (#25025) SYCL: add oneMKL GEMM flash attention for XMX-accelerated prompt processing fattn-mkl: fix interleaved dst layout in nor

opencl: skip the Adreno KQ/KQV image kernels for multi-stream batches (#26189) The Adreno KQ/KQV image1d kernels (ggml_cl_mul_mat_kq_kqv_adreno) ignore dim 3 entirely: the sub-buffer covers only nb02*

What's Changed launch: keep Claude Code channels available by @hoyyeva in #17210 cmd: remove dead agent prompt wrappers by @ParthSareen in #17227 agent: reorder working directory instruction by @Parth

llama.cpp release **b8739** is a build of the open-source C/C++ LLM inference engine that introduces HIP backend support for the CDNA4 (gfx950) GPU architecture, enabling hardware acceleration on A...