Pull Requests ggerganov/llama.cpp

ggml : add fused SINKHORN_NORM op (Sinkhorn-Knopp normalization) + us… model testing ggml CUDA

#25421 opened 2026-07-07 22:19 by jadenmach2

server : respect min-step when splitting prompt batches server

#25420 opened 2026-07-07 22:17 by aldehir

metal: Q2_0 backend ggml Apple Metal

#25419 opened 2026-07-07 22:10 by khosravipasha

ggml-webgpu: tune subgroup split (d_split) in flash_attn_vec ggml WebGPU

#25418 opened 2026-07-07 21:56 by yomaytk

server: fix prefill speed regression on session resumption testing server

#25417 opened 2026-07-07 21:31 by SakshamKapoor2911

server: tool role detection and create checkpoints only on user messages server

#25416 opened 2026-07-07 20:36 by Kononnable

Fix DeepSeek4 crafted template testing jinja parser

#25414 opened 2026-07-07 17:45 by pwilkin

GLM 5.2 Indexer support model conversion

#25407 opened 2026-07-07 14:18 by pcuenca

chat: honor explicit deepseek reasoning format testing

#25405 opened 2026-07-07 13:53 by liminfei-amd

Make hip quality check run on all changes devops

#25403 opened 2026-07-07 11:49 by ORippler

DeepSeek-V4 Checkpointing Fix

#25402 opened 2026-07-07 11:49 by danielhanchen

gguf-py: validate n_dims and guard uint64 overflow in reader conversion

#25401 opened 2026-07-07 11:37 by hclsys

ggml-cpu: Added RVV VLEN=1024 vector dot product (vec_dot) kernels for quantized types. ggml

#25397 opened 2026-07-07 10:04 by velonica0

model: add Hy3 (hy_v3) support with MTP speculative decoding model testing conversion

#25395 opened 2026-07-07 09:57 by satindergrewal

ggml-cpu: Incorrect A indexing in simd_gemm scalar tail-column path ggml merge ready

#25390 opened 2026-07-07 07:47 by tyronecai

llama-graph: prevent f16 overflow in MoE down-projection on Metal

#25389 opened 2026-07-07 07:30 by joerowell

readme : add Lynkr to Infrastructure section documentation

#25387 opened 2026-07-07 06:15 by vishalveerareddy123

CUDA: check UMA before cudaMemGetInfo in device_get_memory ggml CUDA

#25384 opened 2026-07-07 05:37 by 123-code

opencl: fix potential crash in aos reconstruct ggml OpenCL

#25383 opened 2026-07-07 05:34 by lhez

vulkan: add SHMEM_STRIDE_PAD/APPLY_SLM_A_RESHAPE for coopmat1 on Intel Xe Vulkan ggml

#25380 opened 2026-07-07 03:26 by fish-jiang

metal : add 64x8 mul_mm tile variant for small batches (q4_0) ggml Apple Metal

#25377 opened 2026-07-07 00:01 by discobot

Add Bodega one to UIs documentation

#25375 opened 2026-07-06 22:44 by Mayimbe07

metal : null-check buffer alloc to fix OOM crash ggml Apple Metal

#25371 opened 2026-07-06 20:48 by ykhrustalev

llama : make all KQ masks f16 if FA is used, remove zero attention bias, remove raw_k repeats in DeepSeek V4 model

#25370 opened 2026-07-06 20:20 by fairydreaming

[RFC] autoparser unable to accommodate non-ASCII reasoning markers, e.g. `◁think▷`

#25369 opened 2026-07-06 20:16 by ulidtko

cuda : add support for f16->f16 GGML_OP_SET_ROWS ggml CUDA

#25367 opened 2026-07-06 17:49 by fairydreaming

vulkan: add UMA dual-access mode for zero-copy CPU access on iGPUs Vulkan ggml

#25365 opened 2026-07-06 16:52 by essentialols

vulkan: add NonTemporal hints to weight loads for AMD L2 cache optimization Vulkan ggml

#25362 opened 2026-07-06 16:32 by essentialols

ggml : make quantization bit-reproducible across compilers and architectures (-ffp-contract=off on ggml-quants.c) ggml

#25353 opened 2026-07-06 12:21 by theadamdanielsson

Add E8 lattice 2-bit KV cache quantization (GGML_TYPE_E8_2) documentation ggml Apple Metal CUDA

#25352 opened 2026-07-06 12:07 by jagmarques