Commit Graph

Select branches

Hide Pull Requests

fable5/prefetch-experts

fable5/vram-ondemand

fable5/vram-ondemand-rebase-20260909

fable5/vram-ondemand-rebased-20260726

gemini

Mono Color

• 70adb1b4ce common: json.h: fix clang lto (#27575) Xuan-Son Nguyen 2026-08-23 01:11:10 +02:00
• 3f545becce vulkan : added the PAD_REFLECT_1D operation (#26586) Safi UllahandJeff Bolz 2026-08-23 00:42:20 +05:00
• b21e4de745 mtmd: use ggml_rope_set_offset (#27521) Xuan-Son Nguyen 2026-08-22 16:33:47 +02:00
• d9f918d2d0 common: add json.h abstraction (#27511) Xuan-Son Nguyen 2026-08-22 16:28:28 +02:00
• 2fb989b9e7 fit: also take into account n_streams (#27496) Xuan-Son NguyenandPascal 2026-08-22 16:16:06 +02:00
• 9fee29e943 arg: remove -no-cnv from cli [no ci] (#27542) Xuan-Son Nguyen 2026-08-22 15:53:56 +02:00
• e85caa81ea ci : Restore ROCm job for Ubuntu (#27399) 3 people 2026-08-22 05:28:30 -05:00
• 2115b73d8e model : support DSpark for bailingmoe3 (#27508) Tiwei Bie 2026-08-22 17:19:48 +08:00
• 54ee5ee643 mtmd: support dots3-note vision+audio (#27524) Xuan-Son NguyenandSigbjørn Skjæret 2026-08-22 10:35:50 +02:00
• 3a653fea93 ci : add older, min and dry-run options to ccache-clear (#27504) Georgi Gerganov 2026-08-22 11:31:30 +03:00
• 369e1cd614 ggml: optimize concat op by replacing per-element memcpy with row-level memcpy (#24575) Kartik SirohiandSigbjørn Skjæret 2026-08-22 14:00:31 +05:30
• 2c6b141efb common : fix draft-mtp with embeddings (#26352, #27299) (#27400) Shahir BIn ZulfikerandSigbjørn Skjæret 2026-08-22 13:44:22 +06:00
• 8672290039 sycl : add Q2_K reordered MMVQ and ESIMD kernels (again) (#27490) Sigbjørn Skjæret 2026-08-22 09:09:26 +02:00
• 3aeb924628 readme : fix server badge alt (#27533) Sigbjørn Skjæret 2026-08-22 09:08:07 +02:00
• 2100e59260 readme : update badges (#27531) Georgi Gerganov 2026-08-22 08:25:00 +03:00
• d775b8967a mtmd: support webp via ffmpeg (#27520) Xuan-Son Nguyen 2026-08-22 01:38:05 +02:00
• 3af988fabc opencl: fold the gpt-oss MoE per-expert bias adds into the epilogue (op/kernel fusion) (#26431) Hongqiang Wang 2026-08-21 14:24:33 -07:00
• 9a286ac98d docs: improve Windows build instructions (#27381) Niklas Wenzel 2026-08-21 20:49:27 +02:00
• a3b9c23ead ci : fix empty release_id in make-release upload step (#27516) Georgi Gerganov 2026-08-21 21:41:25 +03:00
• 5a32f7b66e model: add dots3-note (#27060) Xuan-Son Nguyen 2026-08-21 19:52:34 +02:00
• 873e5d8e39 model: use ggml_rope_set_offset() (#27382) Xuan-Son Nguyen 2026-08-21 18:54:29 +02:00
• d7fa69b7de ci : run ccache-clear as the last step of release jobs (#27503) Georgi GerganovandSigbjørn Skjæret 2026-08-21 18:56:04 +03:00
• bb4caa7540 llama.cpp : bump version to 0.2.0 (#27498) Georgi Gerganov 2026-08-21 15:01:24 +03:00
• c4b0225d85 scripts : add release.sh for release preparation (#27497) Georgi Gerganov 2026-08-21 14:51:26 +03:00
• 5de25a7487 sync : ggml Georgi Gerganov 2026-08-21 14:32:32 +03:00
• 01ff204fbd ggml : bump version to 0.21.0 (ggml/1597) Georgi Gerganov 2026-08-21 13:22:00 +03:00
• 353b32d8b9 ci : remove duplicate flag (#27488) Georgi Gerganov 2026-08-21 14:10:06 +03:00
• 7a0e42fd01 Revert "sycl : add Q2_K reordered MMVQ and ESIMD kernels (#26336)" (#27486) Georgi Gerganov 2026-08-21 14:02:03 +03:00
• 5b6ddc9675 ui: Settings navigation cleanup (#27241) Aleksander Grygier 2026-08-21 12:30:03 +02:00
• e467c2ff61 ci : add nightly-tag.txt to make-release (#27485) Georgi Gerganov 2026-08-21 13:20:44 +03:00
• 1719747451 ci : release clean-up (#27477) Georgi Gerganov 2026-08-21 11:33:40 +03:00
• 62b2269060 kleidiai : add SME2 F32 GEMV kernel support (#26891) Charles Xu 2026-08-21 10:33:30 +02:00
• ff14356e0c sycl : add Q2_K reordered MMVQ and ESIMD kernels (#26336) Todd Malsbary 2026-08-21 01:01:40 -07:00
• 5fff128451 test : make the FA V-is-view-of-K case a test case parameter (#27394) Georgi Gerganov 2026-08-21 10:29:17 +03:00
• 9e89a196b8 sycl : Add Q5_K ESIMD kernel (#26376) Todd Malsbary 2026-08-21 00:23:02 -07:00
• cd26896c19 opencl: keep the vocab-scale K-quant lm_head on the CPU for Adreno A7X (compiler issue workaround) (#26440) Hongqiang WangandLi He 2026-08-20 22:30:17 -07:00
• 1cb3f5eb41 sycl: Update gate logic for Alchemist GPUs regarding OneDNN features. (#26635) HumerousGorgonandNeo Zhang 2026-08-21 13:16:29 +08:00
• 6602dd3389 sycl: fix multiple warnings in compiling sycl backend (#26713) Ian Faust 2026-08-21 07:15:40 +02:00
• 9e96cf77ff sycl : fix load model with mlock issue (#27250) Neo Zhang 2026-08-21 13:14:54 +08:00
• b2e5e9b28b TP: enable tensor split for LFM2/LFM2MOE (#26993) Chris Danis 2026-08-21 01:13:58 -04:00
• a298422da7 docs: fix typos in ET.md (#27457) vk 2026-08-21 10:06:59 +05:30
• 749f688fca ggml: support ggml_rope_set_offset on opencl, sycl, wgpu, hexagon (#27345) Xuan-Son Nguyen 2026-08-21 00:36:57 +02:00
• 0e1d9185c5 ci: use shell script to check cmake pkg (#27414) Eve 2026-08-20 20:01:32 +00:00
• a30273376e metal : clamp K extent in tensor API mat-mat kernel for K not a multiple of 32 (#27450) Georgi Gerganov 2026-08-20 21:31:29 +03:00
• 6503355df0 opencl: fix q6_K flat mul_mat for Adreno A6x/A7x GPUs with older E031 compilers (#26476) Hongqiang Wang 2026-08-20 10:58:35 -07:00
• 6b4fa88a6c opencl: fix local size for norm (#27339) lhez 2026-08-20 10:52:07 -07:00
• 521a64cd01 ui: Stores split refactor (#27240) Aleksander Grygier 2026-08-20 19:02:04 +02:00
• 681c29d36a mtmd: add --mmproj-device argument (#23255) John-Henry LimandXuan Son Nguyen 2026-08-21 00:45:37 +08:00
• 07822bddf8 model : support DSpark for LFM2 models (#27383) Tarek Dakhran 2026-08-20 16:36:57 +02:00
• 78ec4c3780 vulkan: FA MMQ should use fp32 for Q quantization calculations (#27413) Jeff Bolz 2026-08-20 09:18:11 -05:00
• 63b64a50a3 metal : dequant kv cache only for large batches (#27438) Georgi Gerganov 2026-08-20 17:00:54 +03:00
• bf0040e15f CI: Use LLVM's OpenMP over MSVC_DEBUG_non_redist on Windows (#26678) Oliver Simons 2026-08-20 15:42:26 +02:00
• 9855ad69d3 server: (router) lazy-load startup_models after main setup (#27424) Xuan-Son Nguyen 2026-08-20 15:22:16 +02:00
• 8a832e4bf3 server : fix --docker-repo being treated as router mode (#27416) Aritro Bandyopadhyay 2026-08-20 06:37:14 -06:00
• 2b5621094e CUDA: adding switch points per HW and quant type to tune the mvq->MMQ decode crossover (#26079) 3 people 2026-08-20 12:36:21 +00:00
• dc64a1620e common : gracefully fallback on unsupported regex patterns in JSON schema (#26939) Aldehir Rojas 2026-08-20 06:59:03 -05:00
• 70aff25250 metal : dequantize quantized KV to F16 before flash attention (#27390) Georgi Gerganov 2026-08-20 13:43:59 +03:00
• f20395dae5 Revert "tensor-split meta backend fixes (#26502)" (#27433) Georgi Gerganov 2026-08-20 13:35:15 +03:00
• 8497981321 ggml: fix backend split scheduler race condition (#26040) Ruben Ortlam 2026-08-20 10:42:33 +02:00
• a3b1effcda convert: fix get block count error for Nemotron 3 Ultra (#27101) Rock Chen 2026-08-20 15:35:28 +08:00
• d9b6be07d0 ggml-cuda: provide static workspace for cuBLAS handles (#26574) Alexander HeislerandOliver Simons 2026-08-20 03:27:51 -04:00
• 929d47a391 graph : create V as a view of K in the k_iswa build_attn (#27392) Georgi Gerganov 2026-08-20 10:00:35 +03:00
• f466cfa38f spec : avoid binding reference to null pointer (#27404) Georgi Gerganov 2026-08-20 10:00:16 +03:00
• 2cfdb5fc08 vulkan : add source groups for shaders (#26666) Markus Tavenrath 2026-08-20 08:52:28 +02:00
• 9ee9fc04c1 opencl: make the MoE expert scatter deterministic (#26464) Hongqiang Wang 2026-08-19 20:40:19 -07:00
• d59d455fd8 tensor-split meta backend fixes (#26502) Max Krasnyansky 2026-08-19 14:53:27 -07:00
• 990e3bfee3 hexagon: fix FA HMX queue ordering and pack the rescale D matrices (#27042) Yiwei Shao 2026-08-19 14:42:57 -07:00
• b062ba735e opencl: port fused ssm_scan kernel (Mamba-2, d_state in {128, 256}) to GPU (#26439) Hongqiang WangandLi He 2026-08-19 13:35:17 -07:00
• cd644c3954 ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE (#26860) Pascal 2026-08-19 22:03:13 +02:00
• 947fd9bb2b server: refactor sleep handling, allow access /metrics during sleep (#27376) Xuan-Son Nguyen 2026-08-19 20:48:09 +02:00
• ee0ea03adf server : make models endpoints private when authentication is enabled (#26347) s0mecode 2026-08-19 22:44:42 +04:00
• dc72703fc6 vulkan : dequant q8_0 KV once in coopmat1 (#25494) Nathanw1014 2026-08-20 01:14:15 +09:30
• b95502ba9a vulkan: add null checks in ggml_vk_queue_command_pools_cleanup (#27353) Jetson Tan 2026-08-19 23:43:10 +08:00
• 3e7344670a Revert "common: share thread pools when n_threads differ (#27138)" (#27337) Niklas WenzelandMax Krasnyansky 2026-08-19 17:05:48 +02:00
• 7221e24f57 model : GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM (#25505) Gabe GoodhartandSigbjørn Skjæret 2026-08-19 08:53:31 -06:00
• 6cc504a2e9 sycl: report zero devices instead of aborting when the host has none (#27291) Titaniumtown 2026-08-19 07:46:01 -07:00
• 01ac3ad761 ci : add release attestation url (#27389) Sigbjørn Skjæret 2026-08-19 16:40:47 +02:00
• 2e92ecd024 models : remove duplicate metadata load (#27378) Sigbjørn Skjæret 2026-08-19 16:05:50 +02:00
• 645ca2834b ci : re-enable release dependency for sycl (#27385) Sigbjørn Skjæret 2026-08-19 15:44:16 +02:00
• fe8156f789 ggml: add ggml_rope_set_offset (+ metal support) (#27120) Xuan-Son Nguyen 2026-08-19 14:04:57 +02:00
• 77acca437f ui: read persisted settings before the API key probe (#27365) Pascal 2026-08-19 14:02:08 +02:00
• 95c409c136 mtmd: add mtmd_bitmap_set_mergeable (#27348) Xuan-Son Nguyen 2026-08-19 13:48:22 +02:00
• 8ef78e644f metal : dequantize q8_0 using packed types (#27370) Georgi Gerganov 2026-08-19 14:10:05 +03:00
• ee4c505a4f server: add dedup-cache-models preset option (#27346) Xuan-Son Nguyen 2026-08-19 11:04:26 +02:00
• 98d1e92c21 vulkan: tiled transpose for 0<->2 permuted CONT (#26585) Kevin HopperandKevin Hopper 2026-08-19 03:20:21 -05:00
• 5112b9738b ggml-webgpu: add mulmat with overlapping src0/src1 (e.g., for minimax-01) (#27321) Masashi Yoshimura 2026-08-19 16:29:33 +09:00
• 0adcc3bb57 ci : add attestation for signed release artifacts (#25933) Jeremie Miller 2026-08-19 01:23:52 -06:00
• 79fe799194 tests: skip test-unicode build on win32/BUILD_SHARED_LIBS (#27336) Jeff Bolz 2026-08-19 01:50:50 -05:00
• 0329fcdac8 gguf-py : add size guards to GGUFReader (#27188) Oğuzhan AkkayaandSigbjørn Skjæret 2026-08-19 02:35:27 -04:00
• 6d05498314 server: (cosmetic) do not print cmd_child_to_router messages [no release] (#27347) Xuan-Son Nguyen 2026-08-19 01:45:56 +02:00
• 60addddf3c opencl: fix WAR race in the generic FA tile kernels when the WG spans subgroups (#26434) Hongqiang Wang 2026-08-18 13:57:24 -07:00
• 9731ad3f29 mtmd: deepseek-ocr SAM ggml_conv_2d with the im2col kept in F32 (#26727) Saba Fallah 2026-08-18 19:05:19 +02:00
• af5172627d RPC: populate use_count to enable fusion inside backends (#27142) Aman Gupta 2026-08-18 21:08:57 +05:30
• 3dc7285b4f ui: Services consolidation refactor (#27239) Aleksander Grygier 2026-08-18 16:39:32 +02:00
• fdf4c64604 ui: Stores consolidation refactor (#27238) Aleksander Grygier 2026-08-18 16:37:26 +02:00
• 04b569142d common: share thread pools when n_threads differ (#27138) Niklas Wenzel 2026-08-18 16:23:43 +02:00
• 0596704284 quant : Optimise memory usage by evicting weights after processing each layer (#22877) Ed Addario 2026-08-18 15:22:32 +01:00
• 0882c7bc89 sycl: honor GGML_HINT_SRC0_IS_HADAMARD (#27298) Titaniumtown 2026-08-18 06:21:25 -07:00
• afd439df1f unicode : include '~' in collapsed symbol class (#26972) Thiago Padilha 2026-08-18 10:15:22 -03:00
• 169e4a7ff2 readme : update status badges + regen AUTHORS (#27317) Georgi Gerganov 2026-08-18 14:35:04 +03:00