Commit Graph

  • 242dbc8cd6 test : make the FA V-is-view-of-K case a test case parameter Georgi Gerganov 2026-08-19 18:02:40 +03:00
  • 5676c4bda8 nits Xuan Son Nguyen 2026-08-20 12:52:03 +02:00
  • 70aff25250 metal : dequantize quantized KV to F16 before flash attention (#27390) b10532 Georgi Gerganov 2026-08-20 13:43:59 +03:00
  • f20395dae5 Revert "tensor-split meta backend fixes (#26502)" (#27433) b10531 Georgi Gerganov 2026-08-20 13:35:15 +03:00
  • 166cad8af6 only allow is_first_load to populate it Xuan Son Nguyen 2026-08-20 12:33:28 +02:00
  • 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 Heisler 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) b10524 Hongqiang Wang 2026-08-19 20:40:19 -07:00
  • c47322cc8a server: (router) lazy-load startup_models after main setup Xuan Son Nguyen 2026-08-20 01:56:18 +02:00
  • e12b6c2c04 nits xsn/server_sleep_proc Xuan Son Nguyen 2026-08-20 01:31:47 +02:00
  • 0d9a89c717 fix Xuan Son Nguyen 2026-08-20 01:28:05 +02:00
  • fc9969c725 fix common_params_print_info Xuan Son Nguyen 2026-08-20 01:20:30 +02:00
  • a138f8d630 move llama_backend_init to load_model Xuan Son Nguyen 2026-08-20 01:11:47 +02:00
  • 4c4a0ba8bb add docs Xuan Son Nguyen 2026-08-20 00:34:02 +02:00
  • 437b415fa8 server: add --sleep-mode rst Xuan Son Nguyen 2026-08-19 23:56:38 +02:00
  • 5f9b61d603 add cache serializer Xuan Son Nguyen 2026-08-19 17:01:35 +02:00
  • 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 Wang 2026-08-19 13:35:17 -07:00
  • cd644c3954 ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE (#26860) b10520 Pascal 2026-08-19 22:03:13 +02:00
  • 947fd9bb2b server: refactor sleep handling, allow access /metrics during sleep (#27376) b10519 Xuan-Son Nguyen 2026-08-19 20:48:09 +02:00
  • d9de2b2062 move to a new llama_kv_cache_dsa_iswa Xuan Son Nguyen 2026-08-19 20:47:00 +02:00
  • ee0ea03adf server : make models endpoints private when authentication is enabled (#26347) s0mecode 2026-08-19 22:44:42 +04:00
  • 0baf73ad96 should_reset_buckets xsn/server_refactor_sleep Xuan Son Nguyen 2026-08-19 17:49:59 +02:00
  • dc72703fc6 vulkan : dequant q8_0 KV once in coopmat1 (#25494) b10517 Nathanw1014 2026-08-20 01:14:15 +09:30
  • b95502ba9a vulkan: add null checks in ggml_vk_queue_command_pools_cleanup (#27353) b10516 Jetson Tan 2026-08-19 23:43:10 +08:00
  • c1fc93f297 split metrics and slots tasks / results Xuan Son Nguyen 2026-08-19 17:15:02 +02:00
  • 3e7344670a Revert "common: share thread pools when n_threads differ (#27138)" (#27337) Niklas Wenzel 2026-08-19 17:05:48 +02:00
  • 7221e24f57 model : GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM (#25505) b10514 Gabe Goodhart 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
  • 6bb8fd0f86 Merge branch 'master' into xsn/dots3-note Xuan Son Nguyen 2026-08-19 15:28:00 +02:00
  • 6f2bb72d9d fix rope Xuan Son Nguyen 2026-08-19 15:26:55 +02:00
  • c993f31fb2 model: use ggml_rope_set_offset() Xuan Son Nguyen 2026-08-19 15:18:59 +02:00
  • 2dbbb34514 rm inplace optimization xsn/ggml_rope_off_rest Xuan Son Nguyen 2026-08-19 00:53:01 +02:00
  • efee18d262 ggml: support ggml_rope_set_offset on opencl, sycl, wgpu, hexagon Xuan Son Nguyen 2026-08-19 00:29:53 +02:00
  • fe8156f789 ggml: add ggml_rope_set_offset (+ metal support) (#27120) b10509 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) b10507 Xuan-Son Nguyen 2026-08-19 13:48:22 +02:00
  • 0591080c28 fix a race condition Xuan Son Nguyen 2026-08-19 13:45:12 +02:00
  • 8ca1955127 add test Xuan Son Nguyen 2026-08-19 13:35:16 +02:00
  • dd7803d3d8 fix get_res_model_info Xuan Son Nguyen 2026-08-19 13:29:00 +02:00
  • 7fa2fdbbdf fix Xuan Son Nguyen 2026-08-19 13:27:13 +02:00
  • 2ccdc2f70c updated docs Xuan Son Nguyen 2026-08-19 13:14:59 +02:00
  • 8ef78e644f metal : dequantize q8_0 using packed types (#27370) b10506 Georgi Gerganov 2026-08-19 14:10:05 +03:00
  • f5aef6ab5c metrics task should not reset timer Xuan Son Nguyen 2026-08-19 13:04:33 +02:00
  • abdb18dd0c allow accessing metrics during sleep Xuan Son Nguyen 2026-08-19 12:56:59 +02:00
  • 5b776fdfe8 refactor on_sleeping_state Xuan Son Nguyen 2026-08-19 12:35:18 +02:00
  • 9f710a6e77 add cached responses Xuan Son Nguyen 2026-08-19 12:29:19 +02:00
  • ee4c505a4f server: add dedup-cache-models preset option (#27346) b10505 Xuan-Son Nguyen 2026-08-19 11:04:26 +02:00
  • 6415eeef12 Merge branch 'master' into xsn/ggml_rope_off xsn/ggml_rope_off Xuan Son Nguyen 2026-08-19 10:39:03 +02:00
  • 884729830d ggml: support ggml_rope_set_offset on vulkan (#27344) Xuan-Son Nguyen 2026-08-19 10:38:19 +02:00
  • 98d1e92c21 vulkan: tiled transpose for 0<->2 permuted CONT (#26585) Kevin 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) b10502 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 Akkaya 2026-08-19 02:35:27 -04:00
  • 85dfd2a292 mtmd: add mtmd_bitmap_set_mergeable xsn/mtmd_bitmap_set_mergeable Xuan Son Nguyen 2026-08-19 01:55:21 +02:00
  • 6d05498314 server: (cosmetic) do not print cmd_child_to_router messages [no release] (#27347) b10499 Xuan-Son Nguyen 2026-08-19 01:45:56 +02:00
  • 2256a051cf server: (cosmetic) do not print cmd_child_to_router messages xsn/server_no_spam_log Xuan Son Nguyen 2026-08-19 01:37:06 +02:00
  • 5fd3e06e2a server: add dedup-cache-models preset option xsn/server_dedup_cache_model Xuan Son Nguyen 2026-08-19 01:16:10 +02:00
  • e3d49edbf4 remove inplace optimization xsn/ggml_rope_off_vk Xuan Son Nguyen 2026-08-19 00:47:45 +02:00
  • 435d4116ed add server_schema for tts xsn/server_tts_2 Xuan Son Nguyen 2026-08-19 00:33:15 +02:00
  • 9f26dc0bf2 wip Xuan Son Nguyen 2026-08-19 00:21:13 +02:00
  • 6d4aa04337 ggml: support ggml_rope_set_offset on vulkan Xuan Son Nguyen 2026-08-18 23:54:35 +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
  • 2e26789a2e fix pocket-tts stream Xuan Son Nguyen 2026-08-18 20:58:16 +02:00
  • 14cc550c70 fix server_output_limits Xuan Son Nguyen 2026-08-18 20:26:53 +02:00
  • 2072dfae41 fix penalty_last_n Xuan Son Nguyen 2026-08-18 20:19:20 +02:00
  • d9d747df3b wip Xuan Son Nguyen 2026-08-18 20:16:02 +02:00
  • 3278c339df fix header for pcm Xuan Son Nguyen 2026-08-18 19:23:08 +02:00
  • 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
  • 1090236cb9 address review comments Xuan Son Nguyen 2026-08-18 16:37:18 +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
  • b78b595b73 Merge branch 'master' into xsn/dots3-note Xuan Son Nguyen 2026-08-18 12:13:28 +02:00
  • 9d77fa1725 ci : Update OpenVINO to 2026.3, skip nemotron-h rollback test (#27292) b10488 Zijun Yu 2026-08-18 18:02:22 +08:00
  • c0296022f3 ci: add Windows ARM64 CUDA support to the manual workflow (#27300) shivamkumard-ctrl 2026-08-18 15:25:45 +05:30
  • c450cd68e0 expose seed to api Xuan Son Nguyen 2026-08-18 11:55:43 +02:00
  • 9534049d52 (sip) allow inject messge to generation xsn/server_control_inject Xuan Son Nguyen 2026-08-18 11:51:30 +02:00
  • 7acdbb1f19 mtmd: fix LFM2 image tiling threshold (#27057) b10486 BlackFoil 2026-08-18 18:11:19 +09:00
  • 1511ce3bc3 sync : ggml v0.1.2 b10485 Georgi Gerganov 2026-08-18 11:29:14 +03:00
  • da786dc23e ggml : bump version to 0.20.2 (ggml/1589) Georgi Gerganov 2026-08-18 11:28:01 +03:00
  • 27e345b574 build : fix xcframework + cmake clean-up (#27304) b10483 Georgi Gerganov 2026-08-18 11:16:51 +03:00
  • 8b86400975 ci : create pre-release with change log and nightly link in make-release (#27302) Georgi Gerganov 2026-08-18 10:14:41 +03:00
  • 25ae3a9b33 CUDA: MMVQ nwarps=8 for bs=1 for dense models on DGX Spark (#26843) b10481 ynankani 2026-08-18 04:15:53 +00:00
  • 4daa33ddb0 Merge remote-tracking branch 'origin/master' into xsn/server_tts_2 Xuan Son Nguyen 2026-08-18 01:36:40 +02:00
  • 5234b9d267 demo, wip xsn/server_disk_cache Xuan Son Nguyen 2026-08-18 00:43:50 +02:00