b10536
server: (router) lazy-load startup_models after main setup (#27424) server: (router) lazy-load startup_models after main setup only allow is_first_load to populate it nits nits 2 Website: https://llam
Knowledge catalogue
server: (router) lazy-load startup_models after main setup (#27424) server: (router) lazy-load startup_models after main setup only allow is_first_load to populate it nits nits 2 Website: https://llam
CI: Use LLVM's OpenMP over MSVC_DEBUG_non_redist on Windows (#26678) CI: Use LLVM's OpenMP over MSFT_DEBUG_non_redist on Windows Currently, we ship the non-redist debug version of microsoft's libomp.
vulkan: FA MMQ should use fp32 for Q quantization calculations (#27413) Codex found that qd could be a denorm and 1/qd would overflow. Website: https://llama.app Attestations: https://github.com/ggml-
mtmd: add --mmproj-device argument (#23255) feat: add --mmproj-device arg & backwards compatible MTMD_BACKEND_DEVICE env var feat: load mmproj device backend immediately, add -mmdev shortflag fix: its
metal : clamp K extent in tensor API mat-mat kernel for K not a multiple of 32 (#27450) The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op o
TP: enable tensor split for LFM2/LFM2MOE (#26993) Assisted-by: deepseek-v4-flash Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/42096995 macOS/iOS: macOS A
ci : run ccache-clear as the last step of release jobs (#27503) ci : run ccache-clear as the last step of release jobs Assisted-by: pi:llama.cpp/Qwen3.8-27B update disabled job too to force rebase Co-
model: use ggml_rope_set_offset() (#27382) model: use ggml_rope_set_offset() partially apply to deepseek2 Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/42
ggml: add ggml_rope_set_offset (+ metal support) (#27120) add params cpu kernel metal kernel add test backend ops gate other backends ggml: (cuda) support ggml_rope_set_offset (#27121) rm cuda support
model : GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM (#25505) feat(convert): Add conversion for GraniteSWAForCausalLM Branch: GraniteSWAForCausalLM AI-usage: full (Bob, OpenCode + Qwen3.6-35b) Sig
vulkan: add null checks in ggml_vk_queue_command_pools_cleanup (#27353) Guard against null queue pointers. Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/4
vulkan : dequant q8_0 KV once in coopmat1 (#25494) vulkan : dequant q8_0 KV once in coopmat1 Assisted-by: Claude (Opus 4.8) vulkan : fall back instead of aborting when FA scratch exceeds maxStorageBuf
server: refactor sleep handling, allow access /metrics during sleep (#27376) add cached responses refactor on_sleeping_state allow accessing metrics during sleep metrics task should not reset timer up
ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE (#26860) ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE __ARM_NEON only signals NEON availability. The __fp16 type also needs the IEEE half format, imp
arXiv:2608.18673v1 Announce Type: new Abstract: Active geo-localization enables low-altitude UAVs to search for specified targets from limited local aerial observations, supporting time-sensitive appl
Hi folks, I have a 7800x3d + 32GB RAM + Rtx4090 and > 200TB of storage. I downloaded the Qwen 3.7 35B hauhaucs but it has been a disappointment so far because training data is too outdated (networking
build : fix xcframework + cmake clean-up (#27304) xcframework : fix build mtmd : remove unused include path vendor : use vendor::hash alias target in cmake CMake reserves '::' in target names for impo
mtmd: fix LFM2 image tiling threshold (#27057) mtmd: fix LFM2 image tiling threshold refactor testing fix fix on windows Co-authored-by: Xuan Son Nguyen son@huggingface.co Website: https://llama.app m
ci : Update OpenVINO to 2026.3, skip nemotron-h rollback test (#27292) update to ov-2026.3, update device drivers ci: skip nemotron-h rollback test on OpenVINO The OpenVINO backend does not support SS
arXiv:2608.15574v1 Announce Type: cross Abstract: Video question answering systems built on vision-language models often produce timestamped claims with high confidence even when unsupported by the ci
Yesterday I downloaded Qwen3.8 as MLX from the Ollama library from the commandline on a MAC. Simple 'Ollama Pull xxx' This is the answer I get when I ask it what model it is... I swear to god this is
When Qwen3.8-27B dropped on Thursday the obvious question came up for us : does a 27B dense model actually fit on the 24GB Mac Mini machines? Ran it properly over the weekend on an M4 Pro (24GB unifie
[SYCL] support OP OPT_STEP_ADAMW, OPT_STEP_SGD (#25268) fix conflict fix conflict of ops.md fix conflict of ops.md update the ops.md Co-authored-by: Neo Zhang Jianyu jianyu.zhang@intel.com Website: ht
sycl: fix thread/block count in quantized cpy kernel launches (#27160) Adjusts the thread/block count to be proportional to the size of the quant, reducing under/over subscription. Largest perf improv
ci : push release tag explicitly in release.yml (#27261) Add a 'Create and push git tag' step to the release job, right before the 'Create release' step. The tag is created with git tag and pushed wit
cuda : skip UMA override for HIP builds (#27083) AMD APUs report accurate memory via hipMemGetInfo. Using MemAvailable over-promises on small-carveout systems. fixes #18159 Website: https://llama.app
llama : check LoRA tensor data is within file bounds (#27056) llama : check LoRA tensor data is within file bounds Update src/llama-adapter.cpp Co-authored-by: Sigbjørn Skjæret sigbjorn.skjaeret@huggi
chat: refactor handling supports_string_content / supports_typed_content (#27130) better supports_string_content cap detect test: add 'skip' messages_inp_normalizer Website: https://llama.app macOS/iO
model : add support for MiniMaxText01ForCausalLM and MiniMaxM1ForCausalLM (#27018) llama : support for MiniMax-Text-01 model chore : renames to match the other MiniMax models model : add logits mask a
common: migrate the deprecated --mmap/--no-mmap to --load-mode (#26934) Replace the deprecated --mmap, --no-mmap, --mlock, and --direct-io flags with the unified --load-mode argument across scripts, e
vulkan: add SHMEM_STRIDE_PAD/APPLY_SLM_A_RESHAPE for coopmat1 on Intel Xe (#25380) vulkan: add SHMEM_STRIDE_PAD/APPLY_SLM_A_RESHAPE for coopmat mul_mm on Intel Xe vulkan: fix shmem estimate for Intel
common: support --models-dir loading MTP assistant models (#24431) common: support --models-dir loading MTP assistant models common: preset: check for MTP models with strict prefix common: preset: Tak
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
dflash : clarify output logging of target_layer_ids (#27013) This commit tries to make the logging of target_layer_ids a bit clearer and easier to read. Currently the output generated looks like this:
sycl: fuse the gated-delta-net state writeback cpy (#26643) Port of #23940. Arc Pro B70, Qwen 3.6 27B Q4_K - Medium (48 of its 64 blocks run gated_delta_net), -ngl 99 -fa 1 -ctk f16 -ctv f16 -b 2048 -
sycl: fuse mul_mat(gate) + mul_mat(up) + GLU for q4_K dense FFN (#26779) Measured on Arc Pro B70 (Battlemage, Level Zero), llama-bench -r 20, two interleaved rounds, tg128: qwen2.5-3B-Instruct Q4_K_M
tests : replace personal home directory paths with generic placeholders (#27043) Scrub developer-specific /home// paths from example docs and test fixtures so they don't leak into the tree. examples/t
server: allow accessing /metrics and /slots during llama_decode() (#27041) server_queue::worker call llama_decode inside yield_to_queue also handle process_mtmd_chunk clean up nits rm test Website: ht
ggml : recurrent state rollback for ggml_ssm_scan (#26623) Initial changes for Recurrent state rollback for nemotron for cpu and cuda Removing CPU RS rollback. Will enable it in subsequent PRs additio
chat : pass reasoning_effort to template chat: add reasoning_effort to common_chat_templates_inputs Store OpenAI Chat Completions reasoning_effort and make it available to jinja templates (with model
arXiv:2608.12645v1 Announce Type: new Abstract: LLM judges have become central infrastructure for model evaluations, online grading, and reward modeling. Judges are typically validated by accuracy on
arXiv:2608.13167v1 Announce Type: cross Abstract: When visual evidence is occluded or chaotic, models should abstain. In this paper, we show that Vision-Language Models (VLMs) can internally distingui
arXiv:2608.12333v1 Announce Type: cross Abstract: Vision-language models must associate visual entities with textual attributes. Whether these associations or concept bindings remain stable when the l
ggml-hip : remove -funsafe-math-optimizations (#26696) It enables -fassociative-math, which reassociates FP reductions and can flip greedy argmax on RDNA3.5 (e.g. MTP speculative decode diverging from
sycl : Add DMMV ESIMD Q3_K kernel (#26251) Add DMMV Q4_K and Q6_K ESIMD kernels Configure cmake build with -DGGML_SYCL_ESIMD=ON to enable. Signed-off-by: Todd Malsbary todd.malsbary@intel.com Refactor
spec: enable backend sampling for both dflash & dspark (#26958) dflash: enable backend sampling for both dflash & dspark enable p_min > 0 in backend sampling and add guard cont : add TODO Co-authored-
common : auto-detect spec type from draft GGUF metadata (#26814) common : auto-detect spec type from draft GGUF metadata When -md loads a local draft model without --spec-type, the sidecar inference i
metal : add TQ2_0 support (#26980) metal: add TQ2_0 support Add support for the GGML_TYPE_TQ2_0 (ternary, 2 bits per element) type in the Metal backend. Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731 c
server : serve index.html with no-cache (#27006) index.html was served with max-age=31536000, immutable like the hashed assets, but its name is stable while its contents change every build, so a cache
[SYCL] Support host pinned mem to improve SYCL Host-to-Device Memory Access (#26789) support host pinned mem, ggml_backend_sycl_host_buffer_type_get_max_size, fix the thread-safe issue Website: https:
OpenVINO: Qwen3.5, memory optimization, and test-recurrent-state-rollback (#26952) OpenVINO backend: 1) enable gpt-oss moe on OV bk; 2) enable mxfp4 support OpenVINO backend: disable TOPK_MOE op test
arXiv:2608.11933v1 Announce Type: new Abstract: Zero-shot anomaly detection (ZSAD) aims to identify anomalies in unseen domains, a setting that is particularly critical for industrial and medical appl
mtmd: support pocket-tts (#26871) adapt the api text model ok working impl, need verify and clean up mtmd: build the pocket-tts transposed convolutions as GEMM + col2im ggml_conv_transpose_1d has no g
LFM2.5-VL-3B is a multimodal variant of LFM2.5, a family of hybrid models designed for on-device deployment. It builds on LFM2-VL-3B with further mid- and post-training. LFM2.5-VL-3B can process both
ci : target ROCm 7.14 for build and release (#25775) Switch ROCm from 7.2.1 to 7.14 ROCm 7.14 is the first production release using TheRock build system. It can be installed using multi-arch deliverab
ggml-webgpu: fix CI errors from #25025 and #25262 (#26566) test new flash_attn test rebase and fix to disable subgrou matrices when max_kv_tile == 0 delete log output Add i32 support to cpy and enable
model : fix SWA not being enabled for EXAONE 4.5 (#26848) model : fix SWA not being enabled for EXAONE 4.5 load_arch_hparams tests hparams.n_layer() == 64 before LLM_KV_NEXTN_PREDICT_LAYERS has been r
arXiv:2608.09389v1 Announce Type: cross Abstract: This note aims to serve as an entry point to the literature on learning in games, a topic with significant theoretical appeal and a wide range of appl
model-saver : fix expert shared/chunk FFN length key clobber (#26693) The saver called add_kv with LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH twice, the second time passing n_ff_chexp. gguf_set_val_u32
model: add MTP support for Nemotron model (#26725) model: add MTP support for Nemotron Nano model model: add mtp_flags for nemotron model address review comments Website: https://llama.app macOS/iOS: