Commit Graph

  • 5069c08034 ggml : split graph allocations according to backend max buffer size (llama/15815) Acly 2025-09-24 16:17:49 +02:00
  • 41245891c1 ggml-cpu: Respect cpumask settings (llama/16164) Xiangyan Sun 2025-09-23 01:58:12 -07:00
  • 73e8f3acb8 ggml : fix uninitialized is_on_grid in quantize_row_iq3_xxs_impl (llama/15928) Sigbjørn Skjæret 2025-09-23 10:25:20 +02:00
  • c706a50746 zdnn: refactor codebase + add docs (llama/16178) Aaron Teo 2025-09-23 14:53:05 +08:00
  • d8d31e3638 ggml-cpu : fix typo in gemm comments [no ci] (llama/16189) Daniel Bevenius 2025-09-23 05:59:03 +02:00
  • 4e32ee733b ggml : implement set_rows with i32 index (llama/16159) Sigbjørn Skjæret 2025-09-22 19:13:00 +02:00
  • df672c6372 ggml : extend ggml_can_fuse to work with non-sequential nodes (llama/16123) Georgi Gerganov 2025-09-22 11:12:37 +03:00
  • 973054a8cd ggml : add ggml_op_is_empty (llama/16122) Georgi Gerganov 2025-09-22 11:12:09 +03:00
  • 9f673df08d Vulkan: add conv_transpose_2d operation (llama/16022) Shin-myoung-serp 2025-09-22 17:04:01 +09:00
  • 14723f25a1 vulkan: add RTE variants of exp shader (llama/16165) Jeff Bolz 2025-09-22 00:37:17 -05:00
  • 95b29fab78 vulkan: vec dot matrix multiplication fix (llama/16151) Ruben Ortlam 2025-09-22 07:22:43 +02:00
  • 4b7f09ac0b opencl: fix concat crash on win arm64 with Adreno (llama/15944) lhez 2025-09-21 16:42:10 -07:00
  • 0a7096f4f3 opencl: initial q8_0 mv support (llama/15732) lhez 2025-09-21 14:48:44 -07:00
  • eae2be0ca2 vulkan: optimize UMA buffer operations and fix driver hangs (llama/16059) Giuseppe Scrivano 2025-09-21 08:31:55 +02:00
  • 9a6c2036a9 vulkan: fix validation error about VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR (llama/16086) Jeff Bolz 2025-09-21 01:23:37 -05:00
  • 8d10ded025 ggml : prepare for development of 0.9.2-dev Georgi Gerganov 2025-09-20 16:44:23 +03:00
  • d89164a08d ggml : bump version to 0.9.1 Georgi Gerganov 2025-09-20 16:44:23 +03:00
  • 36778bd8b8 talk-llama : sync llama.cpp Georgi Gerganov 2025-09-20 13:47:47 +03:00
  • 66ad624d5b sync : ggml Georgi Gerganov 2025-09-20 13:46:41 +03:00
  • 76d0934287 vulkan: use vec dot for matrix matrix multiplications (llama/16056) Ruben Ortlam 2025-09-20 10:42:56 +02:00
  • 2ad00d5586 ggml : refactor forward_dup for cpu backend (llama/16062) Xuan-Son Nguyen 2025-09-19 11:31:56 +07:00
  • 4d8cd07825 ggml-amx : fix ggml_amx_init() on generic Linux (llama/16049) Adrien Gallouët 2025-09-18 23:07:26 +02:00
  • 4575f96873 cmake : fix static linking for OpenMP on Unix-like systems (llama/16031) Adrien Gallouët 2025-09-18 23:07:18 +02:00
  • f4a225cea6 opencl: optimize mxfp4 kernels (llama/16037) Shawn Gu 2025-09-18 12:03:34 -07:00
  • 7fcb7e83ec rename optimize_graph to graph_optimize (llama/16082) Jeff Bolz 2025-09-18 13:46:17 -05:00
  • fce6354e0f CUDA: Optimize PAD_REFLECT_1D (llama/15957) Bowen Han 2025-09-18 11:26:03 -07:00
  • 05bdfd4380 CUDA: fix compilation on CC 6.0 (llama/16091) Johannes Gäßler 2025-09-18 19:28:32 +02:00
  • 960aaa9904 metal : use function constants for mul_mv_ext kernels (llama/16074) Georgi Gerganov 2025-09-18 16:28:41 +03:00
  • 225d7c1d5a cuda : add missing F32<->I32 entries in ggml_cuda_cpy_fn (llama/16060) Sigbjørn Skjæret 2025-09-18 13:28:22 +02:00
  • d37f590a77 metal : improve F32, F16 and BF16 mat-vec multiplication (llama/16057) Georgi Gerganov 2025-09-18 12:33:45 +03:00
  • 32b6d9c134 metal : avoid call free for non-owned buffer (llama/16067) Jhen-Jie Hong 2025-09-18 15:06:48 +08:00
  • 1f24b1df4d metal : handle nil cv during pipeline creation (llama/16065) Georgi Gerganov 2025-09-18 10:03:24 +03:00
  • c46adc0817 CANN: Remove print (llama/16044) Chenguang Li 2025-09-18 09:26:33 +08:00
  • 1361f679cc GGML WebGPU: Support for ADD, MUL, RMS_NORM, GET_ROWS operators (llama/16018) Reese Levine 2025-09-17 13:09:40 -07:00
  • eb2c01f92e metal : refactor + optimize v2 (llama/15995) Georgi Gerganov 2025-09-20 13:46:10 +03:00
  • 6458bac4c1 sync : ggml Georgi Gerganov 2025-09-20 13:45:32 +03:00
  • d452f0cf8c CUDA: fix FA occupancy, optimize tile kernel (llama/15982) Johannes Gäßler 2025-09-17 15:32:42 +02:00
  • e96b285011 vulkan: automatically remove unsupported devices (llama/15976) Eve 2025-09-17 07:35:37 +00:00
  • e32c3b0fd3 CANN: Optimize ggml_cann_set_device (llama/15935) Chenguang Li 2025-09-17 14:33:08 +08:00
  • 5c524bb879 ggml : fix padding in timestep embedding kernels (llama/15932) Daniel Bevenius 2025-09-16 15:25:57 +02:00
  • f72ec185fb CUDA: fix im2col_3d to respect non-contiguous inputs (views) (llama/15956) Jake Karnes 2025-09-15 16:28:31 -06:00
  • a642b533a4 SYCL: Add COUNT_EQUAL operator support (llama/15991) yael-works 2025-09-15 19:51:35 +03:00
  • 10bd5d3626 CUDA: some micro-optimizations in mmf.cuh for mul_mat_id (llama/15926) Aman Gupta 2025-09-15 17:35:11 +08:00
  • 82a8c141ea metal : remove memory pools (llama/15966) Georgi Gerganov 2025-09-14 22:02:32 +03:00
  • c36358cb3c Vulkan: Clean up mul_mm shader (llama/15987) Ruben Ortlam 2025-09-14 16:56:28 +02:00
  • 2d3f15607f metal : fix kernel requirements (llama/15983) Georgi Gerganov 2025-09-14 15:33:22 +03:00
  • 7dca05ca77 ggml-zdnn: rm user mapped buffers (llama/15965) Aaron Teo 2025-09-14 13:37:03 +08:00
  • 1789ed3f2c vulkan: fix failing dequant shaders (llama/15862) Jeff Bolz 2025-09-13 16:29:43 +01:00
  • a3defb0a3b vulkan: initialize vulkan-hpp to allow using extension function pointers (llama/15705) Jeff Bolz 2025-09-13 16:23:30 +01:00
  • 2caf15d68a metal : refactor kernel loading (llama/15964) Georgi Gerganov 2025-09-13 16:24:22 +03:00
  • 0d36ba9e1a metal : allow ops to run concurrently (llama/15929) Georgi Gerganov 2025-09-13 13:54:28 +03:00
  • 20a930ec94 metal : fix memory leaks (llama/15962) Georgi Gerganov 2025-09-13 12:45:04 +03:00
  • e902731ccc ggml-zdnn: fix #15414, activate FP16 and BF16 acceleration and incorrect zTensor free (llama/15839) Aaron Teo 2025-09-13 02:39:52 +08:00
  • 424c85f22a Vulkan iGPU device selection overhaul and PCI ID API support (llama/15947) Ruben Ortlam 2025-09-12 13:24:21 +02:00
  • 5a752bab84 vulkan: Make device memory check more portable (llama/15939) Mathieu Baudier 2025-09-12 09:06:20 +02:00
  • cd764eaf2b Revert "sycl: add usage of enqueue_functions extension (llama/14244)" (llama/15910) Neo Zhang Jianyu 2025-09-12 09:15:12 +08:00
  • 555dcb3e01 ggml-backend : add GGML_BACKEND_DEVICE_TYPE_IGPU device type (llama/15797) Diego Devesa 2025-09-11 13:47:38 -07:00
  • f0768eb575 CUDA: larger SRAM reads for tile FA, AMD FP16 dot (llama/15927) Johannes Gäßler 2025-09-11 21:19:58 +02:00
  • 020eb19eb3 ggml-cpu : add check for ARM MATMUL_INT8/i8mm support (llama/15922) Daniel Bevenius 2025-09-11 15:39:12 +02:00
  • b079d9c8b0 kleidiai: fix GGML_ASSERT(*cur_backend_id != -1) failed (llama/15614) Charles Xu 2025-09-11 12:45:40 +02:00
  • dadf73665a CANN: Disable acl_graph for prefill stage (llama/15933) hipudding 2025-09-11 15:59:37 +08:00
  • f5ef0e25e2 CUDA: Add fastdiv to k_bin_bcast*, giving 1-3% E2E performance (llama/15872) Oliver Simons 2025-09-10 22:04:03 +02:00
  • 3617008c37 ggml-cpu : fix padding in ggml_timestep_embedding (llama/15917) Daniel Bevenius 2025-09-10 17:31:40 +02:00
  • c974f63057 sync : ggml Georgi Gerganov 2025-09-20 13:44:48 +03:00
  • 7eae055e61 metal : make the backend async (llama/15906) Georgi Gerganov 2025-09-20 13:44:27 +03:00
  • e2c7f1cccd sync : ggml Georgi Gerganov 2025-09-20 13:43:01 +03:00
  • 4d453b14a9 CANN: Add ROPE sin/cos cache for reuse (llama/15912) Chenguang Li 2025-09-10 18:42:00 +08:00
  • 9b773acac0 CANN: implement LRU cache for ACL graphs (llama/15814) Chenguang Li 2025-09-10 15:29:12 +08:00
  • 7abe187860 vulkan: throw the oom error instead of no memory type found (llama/15905) Ruben Ortlam 2025-09-09 22:26:03 +02:00
  • d0e98656c3 vulkan: Fix OOB accesses in soft_max_back (llama/15861) Jeff Bolz 2025-09-09 07:41:15 -05:00
  • e35d1375ee HIP: use v_dot2_f32_f16 instruction for FA (llama/15884) Johannes Gäßler 2025-09-09 14:04:43 +02:00
  • 7fbbb67b47 Workaround for subgroup arithmetic failing on MoltenVK with AMD GPUs (issue 15846) (llama/15886) lksj92hs 2025-09-09 15:01:15 +03:00
  • 621764b1a5 CUDA: Add mul_mat_id support for the mmf kernel (llama/15767) Aman Gupta 2025-09-09 14:38:02 +08:00
  • 260982232c CUDA: fix GET_ROWS for large tensors (llama/15882) Johannes Gäßler 2025-09-09 08:11:01 +02:00
  • c29cd54818 vulkan: sort graph to allow more parallel execution (llama/15850) Jeff Bolz 2025-09-08 13:10:07 -05:00
  • 70ee808f3d CUDA: generate_cu_files.py - add missing mxfp4 (llama/15880) Aman Gupta 2025-09-09 01:23:46 +08:00
  • ae6cc6a386 cuda : fix supports_op condition for get_rows when number of blocks is too large (llama/15868) Georgi Gerganov 2025-09-08 13:56:51 +03:00
  • e9cb59e970 metal : refactor + optimize (llama/15857) Georgi Gerganov 2025-09-20 13:04:02 +03:00
  • 40bcd1a469 ggml: allow casting between f32 and i32 (llama/15783) Xuan-Son Nguyen 2025-09-08 17:33:01 +07:00
  • 0175a1df8d CUDA: non-contiguous src0 not supported for PAD (llama/15869) Sigbjørn Skjæret 2025-09-08 11:55:44 +02:00
  • d9c0ead2ab CANN: Stream sync between devices for acl_graph (llama/15809) Chenguang Li 2025-09-08 10:03:29 +08:00
  • dfa7722e2e vulkan: support im2col_3d (llama/15795) Jeff Bolz 2025-09-07 13:50:26 -05:00
  • db4f504b69 ggml-cpu: clean up s390x SIMD (llama/15855) Aaron Teo 2025-09-08 02:18:28 +08:00
  • 9523fd8de6 vulkan: Support pad_ext (llama/15794) Jeff Bolz 2025-09-07 12:00:49 -05:00
  • 647e2d7de5 vulkan: Use larger loads in scalar/coopmat1 matmul (llama/15729) Jeff Bolz 2025-09-07 11:53:07 -05:00
  • cda7d4e5ac ggml WebGPU: remove userdata from request adapter callback (llama/15527) Daniel Bevenius 2025-09-07 10:19:45 +02:00
  • cd70d89628 CUDA: faster tile FA (Pascal/AMD), headsize 256 (llama/15769) Johannes Gäßler 2025-09-07 00:26:28 +02:00
  • be2676bb1c kleidiai: generalize compute_forward_kv_cache to compute_forward_fp16 (llama/15817) Charles Xu 2025-09-06 16:08:43 +02:00
  • 69400f16f1 ggml-cpu: document use of "free" memory [no ci] (llama/15834) Johannes Gäßler 2025-09-06 13:28:44 +02:00
  • f499271c4e ggml-cpu: drop support for nnpa intrinsics (llama/15821) Aaron Teo 2025-09-06 11:27:28 +08:00
  • 6ff468cfaa CUDA: fastdiv, launch bounds for mmvq + q8_1 quant (llama/15802) Johannes Gäßler 2025-09-05 16:07:02 +02:00
  • 4d6e1144b1 ggml : introduce semantic versioning (ggml/1336) Daniel Bevenius 2025-09-16 06:16:52 +02:00
  • c80f78cc7b CUDA : conditionally add cuda architectures (ggml/1341) Gregor Jasny 2025-09-10 17:21:11 +02:00
  • ffe560cbb1 metal : Add template specialization for mul_mm_id w/ ne20 == 10 (llama/15799) Gabe Goodhart 2025-09-04 09:53:22 -06:00
  • 3780a3c917 CANN: Refactor ND to NZ workspace to be per-device (llama/15763) Chenguang Li 2025-09-04 20:20:14 +08:00
  • 2228462b19 ggml: add ops for WAN video model (cuda && cpu) (llama/15669) leejet 2025-09-04 16:38:49 +08:00
  • 96efb472b4 CANN: Fix precision issue on 310I DUO multi-devices (llama/15784) hipudding 2025-09-04 15:12:30 +08:00
  • 1569daf524 opencl: add hs=40 to FA (llama/15758) rmatif 2025-09-04 08:30:28 +02:00
  • 5c860e94c6 CANN: fix acl_rstd allocation size in ggml_cann_rms_norm (llama/15760) Chenguang Li 2025-09-04 11:03:02 +08:00
  • 719a05c665 vulkan: fix mmv subgroup16 selection (llama/15775) Ruben Ortlam 2025-09-03 22:55:10 +02:00