github-featured/ggml-org--ggml
2
0
Fork 0
Code Issues Pull Requests Deployments Models Agent Notes Packages Projects Releases Wiki Activity

Commit Graph

  • d99724f2b1 readme : update following llama.cpp style (#1599) master Georgi Gerganov 2026-08-21 15:39:22 +03:00
  • 43724a637d ci : run builds only on source and build file changes (#1600) Georgi Gerganov 2026-08-21 15:39:10 +03:00
  • 6116c6d952 ci : run builds only on source and build file changes gg/ci-paths Georgi Gerganov 2026-08-21 15:35:44 +03:00
  • 8599e0ea37 sync : llama.cpp v0.21.0 Georgi Gerganov 2026-08-21 14:18:23 +03:00
  • 9a538ea58b Revert "sycl : add Q2_K reordered MMVQ and ESIMD kernels (llama/26336)" (llama/27486) Georgi Gerganov 2026-08-21 14:02:03 +03:00
  • 28d7ab840e ci : extract build job into separate release workflow (#1598) Georgi Gerganov 2026-08-21 14:16:54 +03:00
  • 91147d0cbe ggml : bump version to 0.21.0 (#1597) Georgi Gerganov 2026-08-21 13:22:00 +03:00
  • 19a38580f5 scripts : restore release.sh preparation script (#1596) Georgi Gerganov 2026-08-21 13:21:17 +03:00
  • bb83ce089e ci : replace release workflow with make-release (#1595) Georgi Gerganov 2026-08-21 13:03:43 +03:00
  • 5da4cf17e6 ci : split self-hosted jobs into separate workflow (#1594) Georgi Gerganov 2026-08-21 13:02:42 +03:00
  • 33be41c550 sync : llama.cpp Georgi Gerganov 2026-08-21 11:36:25 +03:00
  • 451e766e34 kleidiai : add SME2 F32 GEMV kernel support (llama/26891) Charles Xu 2026-08-21 10:33:30 +02:00
  • 59bbab9c66 sycl : add Q2_K reordered MMVQ and ESIMD kernels (llama/26336) Todd Malsbary 2026-08-21 01:01:40 -07:00
  • 5a232ab4e3 test : make the FA V-is-view-of-K case a test case parameter (llama/27394) Georgi Gerganov 2026-08-21 10:29:17 +03:00
  • 14dfda3a24 sycl : Add Q5_K ESIMD kernel (llama/26376) Todd Malsbary 2026-08-21 00:23:02 -07:00
  • df8336ef44 opencl: keep the vocab-scale K-quant lm_head on the CPU for Adreno A7X (compiler issue workaround) (llama/26440) Hongqiang Wang 2026-08-20 22:30:17 -07:00
  • 9969469283 sycl: Update gate logic for Alchemist GPUs regarding OneDNN features. (llama/26635) HumerousGorgon 2026-08-21 13:16:29 +08:00
  • 1d30b1b585 sycl: fix multiple warnings in compiling sycl backend (llama/26713) Ian Faust 2026-08-21 07:15:40 +02:00
  • 7378ac178c sycl : fix load model with mlock issue (llama/27250) Neo Zhang 2026-08-21 13:14:54 +08:00
  • 4979ee24eb ggml: support ggml_rope_set_offset on opencl, sycl, wgpu, hexagon (llama/27345) Xuan-Son Nguyen 2026-08-21 00:36:57 +02:00
  • 33c9ea5e14 metal : clamp K extent in tensor API mat-mat kernel for K not a multiple of 32 (llama/27450) Georgi Gerganov 2026-08-20 21:31:29 +03:00
  • 9e7a4c2cec opencl: fix q6_K flat mul_mat for Adreno A6x/A7x GPUs with older E031 compilers (llama/26476) Hongqiang Wang 2026-08-20 10:58:35 -07:00
  • 505842b49e opencl: fix local size for norm (llama/27339) lhez 2026-08-20 10:52:07 -07:00
  • b0d45de2f6 vulkan: FA MMQ should use fp32 for Q quantization calculations (llama/27413) Jeff Bolz 2026-08-20 09:18:11 -05:00
  • 3821f6ed36 metal : dequant kv cache only for large batches (llama/27438) Georgi Gerganov 2026-08-20 17:00:54 +03:00
  • 7159fc6a22 CI: Use LLVM's OpenMP over MSVC_DEBUG_non_redist on Windows (llama/26678) Oliver Simons 2026-08-20 15:42:26 +02:00
  • 02a0ab2b47 CUDA: adding switch points per HW and quant type to tune the mvq->MMQ decode crossover (llama/26079) Pranesh Gonegandla 2026-08-20 12:36:21 +00:00
  • e4159ddd9f metal : dequantize quantized KV to F16 before flash attention (llama/27390) Georgi Gerganov 2026-08-20 13:43:59 +03:00
  • 9a1d2348dd Revert "tensor-split meta backend fixes (#26502)" (llama/27433) Georgi Gerganov 2026-08-20 13:35:15 +03:00
  • 0ab579a594 ggml: fix backend split scheduler race condition (llama/26040) Ruben Ortlam 2026-08-20 10:42:33 +02:00
  • ee401c8b63 ggml-cuda: provide static workspace for cuBLAS handles (llama/26574) Alexander Heisler 2026-08-20 03:27:51 -04:00
  • 6a6a4c19c9 vulkan : add source groups for shaders (llama/26666) Markus Tavenrath 2026-08-20 08:52:28 +02:00
  • e3e8dc16f5 opencl: make the MoE expert scatter deterministic (llama/26464) Hongqiang Wang 2026-08-19 20:40:19 -07:00
  • 7c6058cfaa tensor-split meta backend fixes (llama/26502) Max Krasnyansky 2026-08-19 14:53:27 -07:00
  • 37c79ff5a5 hexagon: fix FA HMX queue ordering and pack the rescale D matrices (llama/27042) Yiwei Shao 2026-08-19 14:42:57 -07:00
  • 585e463c85 opencl: port fused ssm_scan kernel (Mamba-2, d_state in {128, 256}) to GPU (llama/26439) Hongqiang Wang 2026-08-19 13:35:17 -07:00
  • ff4f3e43bc ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE (llama/26860) Pascal 2026-08-19 22:03:13 +02:00
  • 480cc77ce9 vulkan : dequant q8_0 KV once in coopmat1 (llama/25494) Nathanw1014 2026-08-20 01:14:15 +09:30
  • 8fc9724d4d vulkan: add null checks in ggml_vk_queue_command_pools_cleanup (llama/27353) Jetson Tan 2026-08-19 23:43:10 +08:00
  • f673848de8 sycl: report zero devices instead of aborting when the host has none (llama/27291) Titaniumtown 2026-08-19 07:46:01 -07:00
  • aa9d222649 ggml: add ggml_rope_set_offset (+ metal support) (llama/27120) Xuan-Son Nguyen 2026-08-19 14:04:57 +02:00
  • 868403cf66 metal : dequantize q8_0 using packed types (llama/27370) Georgi Gerganov 2026-08-19 14:10:05 +03:00
  • 55e4b3b491 vulkan: tiled transpose for 0<->2 permuted CONT (llama/26585) Kevin Hopper 2026-08-19 03:20:21 -05:00
  • 52b66a4ed4 ggml-webgpu: add mulmat with overlapping src0/src1 (e.g., for minimax-01) (llama/27321) Masashi Yoshimura 2026-08-19 16:29:33 +09:00
  • 981a41bb6f opencl: fix WAR race in the generic FA tile kernels when the WG spans subgroups (llama/26434) Hongqiang Wang 2026-08-18 13:57:24 -07:00
  • f16a1a2c23 RPC: populate use_count to enable fusion inside backends (llama/27142) Aman Gupta 2026-08-18 21:08:57 +05:30
  • b74262dd4b sycl: honor GGML_HINT_SRC0_IS_HADAMARD (llama/27298) Titaniumtown 2026-08-18 06:21:25 -07:00
  • 0c5470cbd1 sync : llama.cpp sync-llama.cpp-26-08-21 Georgi Gerganov 2026-08-21 11:36:25 +03:00
  • 14a5374bcd kleidiai : add SME2 F32 GEMV kernel support (llama/26891) Charles Xu 2026-08-21 10:33:30 +02:00
  • 57bb67edca sycl : add Q2_K reordered MMVQ and ESIMD kernels (llama/26336) Todd Malsbary 2026-08-21 01:01:40 -07:00
  • 5f6811aa88 test : make the FA V-is-view-of-K case a test case parameter (llama/27394) Georgi Gerganov 2026-08-21 10:29:17 +03:00
  • 8609089400 sycl : Add Q5_K ESIMD kernel (llama/26376) Todd Malsbary 2026-08-21 00:23:02 -07:00
  • df78ff0d8c opencl: keep the vocab-scale K-quant lm_head on the CPU for Adreno A7X (compiler issue workaround) (llama/26440) Hongqiang Wang 2026-08-20 22:30:17 -07:00
  • 018a590c1f sycl: Update gate logic for Alchemist GPUs regarding OneDNN features. (llama/26635) HumerousGorgon 2026-08-21 13:16:29 +08:00
  • 059f694e43 sycl: fix multiple warnings in compiling sycl backend (llama/26713) Ian Faust 2026-08-21 07:15:40 +02:00
  • 02353bd4a1 sycl : fix load model with mlock issue (llama/27250) Neo Zhang 2026-08-21 13:14:54 +08:00
  • 7bd17f52b9 ggml: support ggml_rope_set_offset on opencl, sycl, wgpu, hexagon (llama/27345) Xuan-Son Nguyen 2026-08-21 00:36:57 +02:00
  • 0885eaeb63 metal : clamp K extent in tensor API mat-mat kernel for K not a multiple of 32 (llama/27450) Georgi Gerganov 2026-08-20 21:31:29 +03:00
  • 80580c9d9b opencl: fix q6_K flat mul_mat for Adreno A6x/A7x GPUs with older E031 compilers (llama/26476) Hongqiang Wang 2026-08-20 10:58:35 -07:00
  • b39458cffd opencl: fix local size for norm (llama/27339) lhez 2026-08-20 10:52:07 -07:00
  • 7d46867d46 vulkan: FA MMQ should use fp32 for Q quantization calculations (llama/27413) Jeff Bolz 2026-08-20 09:18:11 -05:00
  • a840bb5ed0 metal : dequant kv cache only for large batches (llama/27438) Georgi Gerganov 2026-08-20 17:00:54 +03:00
  • be75e18641 CI: Use LLVM's OpenMP over MSVC_DEBUG_non_redist on Windows (llama/26678) Oliver Simons 2026-08-20 15:42:26 +02:00
  • 624f5f7529 CUDA: adding switch points per HW and quant type to tune the mvq->MMQ decode crossover (llama/26079) Pranesh Gonegandla 2026-08-20 12:36:21 +00:00
  • f951abfe45 metal : dequantize quantized KV to F16 before flash attention (llama/27390) Georgi Gerganov 2026-08-20 13:43:59 +03:00
  • f0fe6e37a9 Revert "tensor-split meta backend fixes (#26502)" (llama/27433) Georgi Gerganov 2026-08-20 13:35:15 +03:00
  • 7780b91063 ggml: fix backend split scheduler race condition (llama/26040) Ruben Ortlam 2026-08-20 10:42:33 +02:00
  • 459c2e6533 ggml-cuda: provide static workspace for cuBLAS handles (llama/26574) Alexander Heisler 2026-08-20 03:27:51 -04:00
  • 2e5518bfff vulkan : add source groups for shaders (llama/26666) Markus Tavenrath 2026-08-20 08:52:28 +02:00
  • 6ffafcd3d0 opencl: make the MoE expert scatter deterministic (llama/26464) Hongqiang Wang 2026-08-19 20:40:19 -07:00
  • ebf748480b tensor-split meta backend fixes (llama/26502) Max Krasnyansky 2026-08-19 14:53:27 -07:00
  • 03ae9b1307 hexagon: fix FA HMX queue ordering and pack the rescale D matrices (llama/27042) Yiwei Shao 2026-08-19 14:42:57 -07:00
  • 95b4869e03 opencl: port fused ssm_scan kernel (Mamba-2, d_state in {128, 256}) to GPU (llama/26439) Hongqiang Wang 2026-08-19 13:35:17 -07:00
  • a35d88f741 ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE (llama/26860) Pascal 2026-08-19 22:03:13 +02:00
  • f19a2427bf vulkan : dequant q8_0 KV once in coopmat1 (llama/25494) Nathanw1014 2026-08-20 01:14:15 +09:30
  • d67a7043c0 vulkan: add null checks in ggml_vk_queue_command_pools_cleanup (llama/27353) Jetson Tan 2026-08-19 23:43:10 +08:00
  • fb375c332b sycl: report zero devices instead of aborting when the host has none (llama/27291) Titaniumtown 2026-08-19 07:46:01 -07:00
  • 74854f3bfc ggml: add ggml_rope_set_offset (+ metal support) (llama/27120) Xuan-Son Nguyen 2026-08-19 14:04:57 +02:00
  • f32bc933c9 metal : dequantize q8_0 using packed types (llama/27370) Georgi Gerganov 2026-08-19 14:10:05 +03:00
  • cff6d3bc57 vulkan: tiled transpose for 0<->2 permuted CONT (llama/26585) Kevin Hopper 2026-08-19 03:20:21 -05:00
  • 8e6b48e9ec ggml-webgpu: add mulmat with overlapping src0/src1 (e.g., for minimax-01) (llama/27321) Masashi Yoshimura 2026-08-19 16:29:33 +09:00
  • dbb3dcbcdf opencl: fix WAR race in the generic FA tile kernels when the WG spans subgroups (llama/26434) Hongqiang Wang 2026-08-18 13:57:24 -07:00
  • 4d9f8ef10b RPC: populate use_count to enable fusion inside backends (llama/27142) Aman Gupta 2026-08-18 21:08:57 +05:30
  • baff7421b7 sycl: honor GGML_HINT_SRC0_IS_HADAMARD (llama/27298) Titaniumtown 2026-08-18 06:21:25 -07:00
  • 8c63e70982 ggml : bump version to 0.20.2 (#1589) v0.20.2 Georgi Gerganov 2026-08-18 11:28:01 +03:00
  • 3f7220233d sync : llama.cpp Georgi Gerganov 2026-08-18 11:26:05 +03:00
  • a7cf17e7d9 CUDA: MMVQ nwarps=8 for bs=1 for dense models on DGX Spark (llama/26843) ynankani 2026-08-18 04:15:53 +00:00
  • 0213cc144e cuda : skip UMA override for HIP builds (llama/27083) Mario Limonciello 2026-08-17 11:35:36 -05:00
  • 3834fd814e ggml : bump version to 0.20.1 (#1587) v0.20.1 Georgi Gerganov 2026-08-17 09:45:11 +03:00
  • 287b79e262 sync : llama.cpp Georgi Gerganov 2026-08-17 09:13:58 +03:00
  • 16ab19de2a sycl: fix thread/block count in quantized cpy kernel launches (llama/27160) Titaniumtown 2026-08-16 21:32:16 -07:00
  • 9a05cb960a support OP OPT_STEP_ADAMW, OPT_STEP_SGD (llama/25268) Neo Zhang 2026-08-17 12:31:29 +08:00
  • 7ea4440eb8 vulkan: add SHMEM_STRIDE_PAD/APPLY_SLM_A_RESHAPE for coopmat1 on Intel Xe (llama/25380) Jiang, Fish 2026-08-15 17:35:05 +08:00
  • 908b21f4bb fixed indent Masato Nakasaka 2026-08-14 11:47:55 +09:00
  • 882faa956c Fixed gating logic for problematic Intel driver version Masato Nakasaka 2026-08-14 10:39:51 +09:00
  • 2d191b5dee ggml : bump version to 0.20.0 (#1584) v0.20.0 Georgi Gerganov 2026-08-14 18:13:36 +03:00
  • acfb0dde9a sync : llama.cpp Georgi Gerganov 2026-08-14 17:45:16 +03:00
  • 66d09faeed ggml : recurrent state rollback for ggml_ssm_scan (llama/26623) lnigam 2026-08-14 19:50:40 +05:30
  • ba3437b95b sycl: fuse mul_mat(gate) + mul_mat(up) + GLU for q4_K dense FFN (llama/26779) Titaniumtown 2026-08-13 23:26:23 -07:00
  • c4f07930ed ggml: force single thread on wasi (llama/25686) Mendy Berger 2026-08-14 02:16:20 -04:00