Commit Graph

  • 17a4cb15b8 ggml-hexagon: Implement true Q8_0 quantization on Hexagon NPU for more accurate mixed-precision matmul operations (llama/17977) Alfred 2025-12-19 12:42:28 -05:00
  • 195d8d0c65 vulkan: Add perf logger mode with concurrency (llama/17944) Jeff Bolz 2025-12-18 23:36:46 -06:00
  • fea481f412 model : add ASR support for LFM2-Audio-1.5B (conformer) (llama/18106) Xuan-Son Nguyen 2025-12-19 00:18:01 +01:00
  • 956fac433b ggml-cpu: extend support for RVV floating-point kernels (llama/17318) Taimur Ahmad 2025-12-18 19:02:09 +05:00
  • 325a9b739c remove i_major_dual (llama/18157) yulo 2025-12-18 19:50:56 +08:00
  • c3a16089e3 ggml-hexagon: swiglu_oai operation (llama/18114) Shouyu 2025-12-17 16:38:21 -05:00
  • c7ccedb5ba ggml-hexagon: gelu operation (llama/17921) Shouyu 2025-12-17 13:39:32 -05:00
  • 1f72f00542 ggml-cpu: ARM64: repack version of q8_0 (dotprod and i8mm) (llama/18096) Alberto Cabrera Pérez 2025-12-17 11:39:13 +00:00
  • 9118c05dc4 HIP: Refactor mma for RDNA and CDNA (llama/17990) yulo 2025-12-17 16:34:54 +08:00
  • 6114e69213 ruby : add Whisper::Token, fix model URI (#3575) KITAITI Makoto 2025-12-24 16:52:16 +09:00
  • 6c22e792cb talk-llama : sync llama.cpp Georgi Gerganov 2025-12-17 15:20:22 +02:00
  • 698348aadc sync : ggml Georgi Gerganov 2025-12-17 15:19:57 +02:00
  • 00108bb713 llama.android : Rewrite Android binding (w/o cpu_features dep) (llama/17413) Naco Siren 2025-12-17 00:14:47 -08:00
  • 41a95b8ba7 ggml : use WARP_SIZE/2 for argmax reduction offset (llama/18092) Aadeshveer Singh 2025-12-17 09:17:01 +05:30
  • 8dd70bdc85 ggml-hexagon: mm for mtmd (llama/17894) Shouyu 2025-12-15 13:53:56 -05:00
  • b90ec07aba metal: use shared buffers on eGPU (llama/17866) Jeremy Demeule 2025-12-15 15:14:49 +01:00
  • aaf3f39b4a llama: automatically set parameters not set by the user in such a way that maximizes GPU utilization (llama/16653) Johannes Gäßler 2025-12-15 09:24:59 +01:00
  • b5e352a52f Support gpt-oss by OPs add-id, mul_mat for mxfp4, swiglu_oai (llama/17826) Neo Zhang Jianyu 2025-12-15 10:35:15 +08:00
  • 3bb4e1e0ac vulkan: fix mul_mat_vec_iq1_s formatting (llama/18026) Ruben Ortlam 2025-12-14 14:52:46 +01:00
  • af2c8cba6f vulkan: Fix data race/hang in scalar/cm1 flash attention (llama/17887) Jeff Bolz 2025-12-14 02:00:00 -06:00
  • 7e5df2975e vulkan: improve mul_mat_vec_iq1_s speed (llama/17874) lovedheart 2025-12-14 08:47:49 +01:00
  • cdadfc3b72 vulkan: faster q6_k matmul (llama/17813) Eve 2025-12-14 07:29:37 +00:00
  • b62ef9af7a ggml : arm repack fix build (llama/0) Georgi Gerganov 2025-12-13 22:54:14 +02:00
  • b901ebe4a3 vulkan: support get_rows for i32 (llama/17941) Jeff Bolz 2025-12-13 03:12:53 -06:00
  • f33446643e vulkan: support GGML_OP_DIAG (llama/17893) Jeff Bolz 2025-12-13 03:07:49 -06:00
  • 939d3085e9 vulkan: Multi-pass softmax for large number of cols (llama/17892) Jeff Bolz 2025-12-13 03:04:29 -06:00
  • 13bb296dbf vulkan: Allow non-pow2 n_experts in topk_moe (llama/17872) Jeff Bolz 2025-12-13 01:40:04 -06:00
  • feb856d4a1 CUDA: fix overflow in MMA kernel without stream-k (llama/17939) Johannes Gäßler 2025-12-12 17:43:58 +01:00
  • db1fcd958f cann : fix ops broken by circular padding guard (llama/17825) Sigbjørn Skjæret 2025-12-12 15:49:27 +01:00
  • 2c782ec325 ggml-cpu : fix RISC-V Q4_0 repack select and RVV feature reporting (llama/17951) ixgbe 2025-12-12 22:26:03 +08:00
  • 25d99e9135 HIP: enable mmf for RDNA3 (llama/17879) yulo 2025-12-12 18:34:33 +08:00
  • e0af519a61 SOLVE_TRI extension to more dimensions (llama/17793) Piotr Wilkin (ilintar) 2025-12-11 17:20:43 +01:00
  • fcaa8f8a7b talk-llama : sync llama.cpp sync-ggml-25-12-17 Georgi Gerganov 2025-12-17 15:20:22 +02:00
  • 82ee376b93 sync : ggml Georgi Gerganov 2025-12-17 15:19:57 +02:00
  • 7e9b6ebb47 llama.android : Rewrite Android binding (w/o cpu_features dep) (llama/17413) Naco Siren 2025-12-17 00:14:47 -08:00
  • 6829ed7187 ggml : use WARP_SIZE/2 for argmax reduction offset (llama/18092) Aadeshveer Singh 2025-12-17 09:17:01 +05:30
  • 5cac224741 ggml-hexagon: mm for mtmd (llama/17894) Shouyu 2025-12-15 13:53:56 -05:00
  • 90e623c297 metal: use shared buffers on eGPU (llama/17866) Jeremy Demeule 2025-12-15 15:14:49 +01:00
  • 542f01ec8b llama: automatically set parameters not set by the user in such a way that maximizes GPU utilization (llama/16653) Johannes Gäßler 2025-12-15 09:24:59 +01:00
  • 6f0e8fecb2 Support gpt-oss by OPs add-id, mul_mat for mxfp4, swiglu_oai (llama/17826) Neo Zhang Jianyu 2025-12-15 10:35:15 +08:00
  • 631c3b3955 vulkan: fix mul_mat_vec_iq1_s formatting (llama/18026) Ruben Ortlam 2025-12-14 14:52:46 +01:00
  • 72c24f6e01 vulkan: Fix data race/hang in scalar/cm1 flash attention (llama/17887) Jeff Bolz 2025-12-14 02:00:00 -06:00
  • a2380a3864 vulkan: improve mul_mat_vec_iq1_s speed (llama/17874) lovedheart 2025-12-14 08:47:49 +01:00
  • 9a211ac983 vulkan: faster q6_k matmul (llama/17813) Eve 2025-12-14 07:29:37 +00:00
  • e500fa6ce4 ggml : arm repack fix build (llama/0) Georgi Gerganov 2025-12-13 22:54:14 +02:00
  • c62c6104cd vulkan: support get_rows for i32 (llama/17941) Jeff Bolz 2025-12-13 03:12:53 -06:00
  • 5cb35693b4 vulkan: support GGML_OP_DIAG (llama/17893) Jeff Bolz 2025-12-13 03:07:49 -06:00
  • 96fa63886c vulkan: Multi-pass softmax for large number of cols (llama/17892) Jeff Bolz 2025-12-13 03:04:29 -06:00
  • 4de5699291 vulkan: Allow non-pow2 n_experts in topk_moe (llama/17872) Jeff Bolz 2025-12-13 01:40:04 -06:00
  • 8d05cf479c CUDA: fix overflow in MMA kernel without stream-k (llama/17939) Johannes Gäßler 2025-12-12 17:43:58 +01:00
  • 240cb2db53 cann : fix ops broken by circular padding guard (llama/17825) Sigbjørn Skjæret 2025-12-12 15:49:27 +01:00
  • a21115d3ce ggml-cpu : fix RISC-V Q4_0 repack select and RVV feature reporting (llama/17951) ixgbe 2025-12-12 22:26:03 +08:00
  • 951f8a978f HIP: enable mmf for RDNA3 (llama/17879) yulo 2025-12-12 18:34:33 +08:00
  • 43be89ef7c SOLVE_TRI extension to more dimensions (llama/17793) Piotr Wilkin (ilintar) 2025-12-11 17:20:43 +01:00
  • 3e79e73eee build: link whisper target against Threads::Threads for FreeBSD support (#3568) Russ 2025-12-17 09:13:38 +00:00
  • 2551e4ce98 server: allow custom temp directory for ffmpeg (#3564) Marcos Del Sol Vives 2025-12-13 08:37:44 +01:00
  • f0c9017a2f ggml : arm repack fix build (#0) sync-ggml-25-12-12 Georgi Gerganov 2025-12-13 08:04:09 +02:00
  • 179d8b1c9c talk-llama : sync llama.cpp Georgi Gerganov 2025-12-12 17:56:43 +02:00
  • 48cdc06e91 sync : ggml Georgi Gerganov 2025-12-12 17:55:11 +02:00
  • 72714d169c whisper : adjust to ggml changes (#0) Georgi Gerganov 2025-12-12 17:54:58 +02:00
  • 324dd21d3c cmake : set CMAKE_RUNTIME_OUTPUT_DIRECTORY for non standalone build (ggml/1394) Congcong Cai 2025-12-12 22:37:38 +08:00
  • 1da1a6865c ggml-alloc : fix reuse-parent logic for misaligned sizes (llama/17884) Georgi Gerganov 2025-12-11 14:30:10 +02:00
  • 0c88de5c69 ggml-hexagon: fix rope failure at test-backend-ops (llama/17565) nullname 2025-12-11 06:45:43 +08:00
  • a2886fba48 Fix race conditions in threadpool when dealing with dynamic/frequent n_threads changes (llama/17748) Max Krasnyansky 2025-12-10 12:32:23 -08:00
  • cd9b8c6d18 ggml : remove GGML_KQ_MASK_PAD constant (llama/17910) Georgi Gerganov 2025-12-10 20:53:16 +02:00
  • ca8ea18d06 cuda : add missing support check for xielu (llama/17895) Sigbjørn Skjæret 2025-12-10 16:16:20 +01:00
  • ea1829134f CUDA: fix unpadded strides in MMA FA kernel (llama/17891) Johannes Gäßler 2025-12-10 12:39:56 +01:00
  • c10b4f9a01 fix softmax for iGPU (llama/17838) Neo Zhang Jianyu 2025-12-10 16:59:57 +08:00
  • 307dc525bb metal: SSM kernel improvements (llama/17876) Gabe Goodhart 2025-12-09 12:30:02 -07:00
  • 2817582be2 Add DIAG for CUDA (llama/17873) Piotr Wilkin (ilintar) 2025-12-09 20:28:57 +01:00
  • 41bbc034f0 ggml : Provide macos-specific backtrace printing to avoid terminal death (llama/17869) Gabe Goodhart 2025-12-09 09:29:07 -07:00
  • b6ae0b29d1 metal : print node names for debugging (llama/17882) Georgi Gerganov 2025-12-09 15:25:49 +02:00
  • ba463fb577 ggml : allow fill node alloc inplace (llama/17870) Sigbjørn Skjæret 2025-12-09 12:23:47 +01:00
  • 79d86a5c2c CANN: add support for partial RoPE and Vision mode (llama/17543) Chenguang Li 2025-12-09 17:53:23 +08:00
  • bef1f5a57e CUDA: fix FP16 overflow in tile FA kernel (llama/17875) Johannes Gäßler 2025-12-09 09:34:02 +01:00
  • 821c2071ab cuda : add FILL op support (llama/17851) Jay Zenith 2025-12-08 05:10:12 -08:00
  • e1562e85fc cuda: optimize SOLVE_TRI using registers and FMAF (llama/17703) wsbagnsv1 2025-12-08 10:41:08 +01:00
  • c8d0ee2f9f ggml-cpu: add ggml_thread_cpu_relax with Zihintpause support (llama/17784) ixgbe 2025-12-08 16:41:34 +08:00
  • d6d44fac69 Vulkan: improve mul_mat_vec_iq1_m (llama/16907) lovedheart 2025-12-07 18:40:42 +01:00
  • 447ef8633b sycl: add missing BF16 conversion support for Intel oneAPI (llama/17780) Law Po Ying 2025-12-07 09:18:18 +08:00
  • 898f876fe2 vulkan: perf_logger improvements (llama/17672) Jeff Bolz 2025-12-06 11:46:46 -06:00
  • ebff8f9db9 ggml-zendnn : add ZenDNN backend for AMD CPUs (llama/17690) Vishal Singh 2025-12-06 21:43:33 +05:30
  • c5e1807071 ggml : add circular tiling support to pad, for Vulkan, CUDA, and CPU (used for making seamless textures) (llama/16985) Phylliida Dev 2025-12-06 06:07:02 -08:00
  • 94be71911f HIP: fix RDNA3 FP16/BF16 matrix multiplication (llama/17817) Johannes Gäßler 2025-12-06 13:45:36 +01:00
  • b67e3abdb2 ggml : improve error handling for search path existence checks (llama/17653) Sky 2025-12-06 19:28:16 +08:00
  • c66c71e9f4 vulkan: Use one row per workgroup for f32 mmv (llama/17711) Jeff Bolz 2025-12-06 04:12:26 -06:00
  • 875d861473 vulkan: support solve_tri with larger N/K values (llama/17781) Jeff Bolz 2025-12-06 01:56:45 -06:00
  • 41cf229d72 metal : fix build(#17799) Georgi Gerganov 2025-12-06 09:33:59 +02:00
  • a8d02735f7 vulkan: Replace deprecated VK_EXT_validation_features (llama/17637) Masato Nakasaka 2025-12-06 14:39:42 +09:00
  • 191e5f46a2 vulkan: Fix mismatch in TOPK_MOE unit test (llama/17541) Masato Nakasaka 2025-12-06 14:23:30 +09:00
  • 64a3f573e0 vulkan: add more num_blocks instantiations in rms_norm (llama/17701) Jeff Bolz 2025-12-05 15:08:56 -06:00
  • 0484147ab2 vulkan: fix top_k bug when there are ties in the input (llama/17659) Jeff Bolz 2025-12-05 15:03:19 -06:00
  • 0b53759b29 vulkan : support conv-2d with large output size (llama/17685) Acly 2025-12-05 21:46:39 +01:00
  • 23984be4da ggml webgpu: unary op suppport, code refactoring, ops support (llama/17764) Reese Levine 2025-12-05 12:25:51 -08:00
  • 7e97d3b069 vulkan: enable mmvq for q2_k on NVIDIA (llama/17675) Jeff Bolz 2025-12-05 14:21:57 -06:00
  • 32ba1ec8e0 vulkan: set all memory allocations to high priority (llama/17624) Jeff Bolz 2025-12-05 14:21:04 -06:00
  • aefcd75f4f rpc : fix alloc size logic (llama/17116) Georgi Gerganov 2025-12-05 19:39:04 +02:00
  • 322903fa67 metal : add residency sets keep-alive heartbeat (llama/17766) Georgi Gerganov 2025-12-05 19:38:54 +02:00
  • 4170159dcd HIP : fix RDNA4 build (llama/17792) Johannes Gäßler 2025-12-05 13:47:52 +01:00
  • d30b744047 Q4/Q8 Tiled Gemm Optimization. (llama/16999) shalinib-ibm 2025-12-05 17:11:51 +05:30