Commit Graph

  • 2979e5f95f ggml: gguf_init_from_callback and gguf_init_from_buffer (llama/22341) Gilad S 2026-05-25 11:33:29 +02:00
  • 44a50ca41a readme : add AMD ROCm/HIP GPU build instructions (#3823) Kaihui-AMD 2026-05-25 17:27:42 +08:00
  • 865ec171aa talk-llama : sync llama.cpp Georgi Gerganov 2026-05-25 12:18:31 +03:00
  • 0a62a579cc sync : ggml Georgi Gerganov 2026-05-25 12:14:40 +03:00
  • 946d6813b9 ggml : bump version to 0.12.1 (ggml/1508) Georgi Gerganov 2026-05-25 12:13:21 +03:00
  • a369b3949c ggml : Parallelize quant LUT init (llama/23595) Jeff Bolz 2026-05-25 02:15:46 -05:00
  • 3306af62b1 TP: fix entirely zero-sized slices per device (llama/23525) Johannes Gäßler 2026-05-24 08:19:33 +02:00
  • 1435988ab3 opencl: batch profiling to improve speed and prevent memory leaks (llama/23495) shaofeiqi 2026-05-23 23:11:43 -07:00
  • b84d03487c hexagon: apply repl optimization in flash attn softmax as #22993 (llama/23455) Yiwei Shao 2026-05-23 19:56:59 -07:00
  • 511f8602b1 ggml : Check the right iface method before using the fallback 2d get (llama/23514) dskwe 2026-05-23 18:49:24 +08:00
  • 6b85d73b33 vulkan: fix windows find_package of SPIRV-Headers (llama/23215) Jeff Bolz 2026-05-23 02:44:46 -05:00
  • aefffa1fa5 opencl: generalize Adreno MoE kernels on M (llama/23449) Shawn Gu 2026-05-22 17:08:41 -07:00
  • b0c9f90059 SYCL: improve MoE prefill throughput (llama/23142) Alexey Kopytko 2026-05-22 21:50:17 +09:00
  • 21c65a78a3 sycl : Level Zero detection in ggml_sycl_init (llama/23097) Alexey Kopytko 2026-05-22 21:49:45 +09:00
  • 0416feecfc SYCL : gated_delta_net K>1 (llama/23174) karavayev 2026-05-22 08:48:56 -04:00
  • 6fb7f1af2c SYCL: add BF16 to DMMV kernel path (~4x tg speedup on Intel Arc) (llama/21580) Katostrofik 2026-05-22 08:48:24 -04:00
  • 2d629533a5 ggml-zendnn : add Q8_0 quantization support (llama/23414) Sachin Sharma 2026-05-22 16:46:55 +05:30
  • ec183556c6 CUDA: fix PDL CC check for JIT compilation (llama/23471) Johannes Gäßler 2026-05-21 23:35:29 +02:00
  • 8402c36039 vulkan: fuse snake activation (mul, sin, sqr, mul, add) (llama/22855) Pascal 2026-05-21 19:39:42 +02:00
  • c436f1419f fix(flash-attn): replace f32 with kv_type and q_type (llama/23372) Chen Yuan 2026-05-21 10:58:49 -04:00
  • 158d93c836 metal : optimize concat kernel and fix set kernel threads (llama/23411) Georgi Gerganov 2026-05-21 13:34:08 +03:00
  • 03da9f17f4 ggml : Check the right iface method before using the fallback 2d get (llama/23306) Matt Corallo 2026-05-21 06:24:40 +00:00
  • 6d1d66de40 hexagon: ssm-conv fix for large prompts (llama/23307) Todor Boinovski 2026-05-20 22:14:13 -07:00
  • 896718eacf opencl: refactor backend initilization (llama/23318) lhez 2026-05-20 09:57:36 -07:00
  • ad717a6de0 vulkan: optimize operations in the IM2COL shader (llama/22685) Daniele 2026-05-20 17:15:13 +02:00
  • b93a5ba605 hexagon: HMX quantized matmul rework (llama/23368) Max Krasnyansky 2026-05-20 07:39:01 -07:00
  • 3fa19558f2 Programmatic Dependent Launch (PDL) for more performance on newer NVIDIA GPUs (Hopper+) (llama/22522) Andreas Kieslinger 2026-05-20 13:59:02 +02:00
  • c58fc465df metal : optimize pad + cpy (llama/23354) Georgi Gerganov 2026-05-20 09:42:00 +03:00
  • 0a0a34287e ggml-cuda: tune RDNA3 Q6_K MMVQ nwarps (llama/23349) ravel7524 2026-05-20 03:52:21 +02:00
  • 37f17208c2 opencl: add MoE support for q4_k, q5_k, q6_k on Adreno (llama/23303) shaofeiqi 2026-05-19 14:29:00 -07:00
  • aca63e7638 hexagon: add MROPE and IMROPE support in HTP rope op (llama/23317) Aparna M P 2026-05-20 02:40:13 +05:30
  • 459ff0707b hexagon: enable support for NORM op (llama/23319) Aparna M P 2026-05-19 22:18:21 +05:30
  • 6090f39f36 ggml-webgpu : extend GDN for K>1 (llama/23299) Reese Levine 2026-05-18 23:45:41 -07:00
  • 28edd0cb36 sycl: add GGML_SYCL_USE_ASYNC_MEM_OP env toggle (llama/22153) Intel AI Get-to Market Customer Success and Solutions 2026-05-18 23:44:02 -07:00
  • 3477fdb2e3 rpc : keep last_graph_uid in the device context (llama/23273) Radoslav Gerganov 2026-05-19 09:42:36 +03:00
  • eb558f23cb hexagon: add support for TRI op (llama/22822) Pranav Dhinakar 2026-05-18 14:04:57 -07:00
  • 0a11c9fe83 ggml-hexagon: add PAD op HVX kernel (llama/23078) Pranav Dhinakar 2026-05-18 13:39:36 -07:00
  • c65b082c94 sycl: scalar SWAR byte-subtract in Q6_K MMVQ dot product (llama/22156) Intel AI Get-to Market Customer Success and Solutions 2026-05-17 22:12:21 -07:00
  • 619262ad24 sycl: route small f32 matmuls to oneMKL, bypass oneDNN (llama/22150) Intel AI Get-to Market Customer Success and Solutions 2026-05-17 22:11:51 -07:00
  • 4fb3ccabd3 feat: Support d_conv=15 for ssm-conv.cu (llama/23017) Gabe Goodhart 2026-05-17 15:05:11 -06:00
  • 53736a3f0e CUDA: Continue directly including cuda/iterator (llama/23102) Oliver Simons 2026-05-17 18:00:10 +02:00
  • 9e96e0eaf1 ggml-vulkan/CMakeLists: add a check for SPIRV-Headers (llama/22009) Jan Ekström 2026-05-17 14:12:11 +03:00
  • 50482cbd22 vulkan: add cpy bf16 -> f32 pipelines (llama/22677) Pascal 2026-05-17 11:31:20 +02:00
  • e417ce7aeb vulkan: Support unaligned tensors for ROPE (llama/22637) Jeff Bolz 2026-05-17 04:30:16 -05:00
  • c7dd64c606 vulkan: fuse SSM_CONV + BIAS + SILU (llama/22653) Jeff Bolz 2026-05-17 03:25:50 -05:00
  • ef5ddecff9 vulkan: removed duplicate #include <memory> in headers (llama/23144) Winston Ma 2026-05-17 01:57:35 +08:00
  • e78e693017 ggml.h: correct ggml_silu_back arg docstring (a=dy, b=x) (ggml/1500) Ori Pekelman 2026-05-21 12:00:16 +00:00
  • 3583e35e0d ggml-alloc: fix out-of-bounds read in ggml_dyn_tallocr_remove_block (ggml/1492) Dev-X25874 2026-05-21 17:28:08 +05:30
  • 587dca0eda ggml : bump version to 0.12.0 (ggml/1494) Georgi Gerganov 2026-05-16 15:59:09 +03:00
  • 23f956de33 llama + spec: MTP Support (llama/22673) Aman Gupta 2026-05-16 20:06:23 +08:00
  • 18a61f44b6 ggml-hexagon: cpy: add contiguous fast-path in reshape copy (llama/23076) Pranav Dhinakar 2026-05-14 16:55:54 -07:00
  • e62d5893f4 HIP: RDNA3 mma FA, faster AMD transpose, tune AMD (llama/22880) Johannes Gäßler 2026-05-14 22:58:58 +02:00
  • 13133ab299 ggml-webgpu: makes the flash attn vec path subgroup-aware (llama/23040) Zheyuan Chen 2026-05-14 09:31:36 -07:00
  • 592a8cd15d logs : reduce (llama/23021) Georgi Gerganov 2026-05-14 13:05:52 +03:00
  • f0223903aa ggml-cpu: Add IME2 Instruction Support for the SpacemiT Backend (llama/22863) alex-spacemit 2026-05-14 17:39:30 +08:00
  • 97ba44338f vulkan: fix matmul integer pipeline selection (llama/23005) Ruben Ortlam 2026-05-14 10:36:54 +02:00
  • 06cfc3653b SYCL: fix multi-GPU system RAM exhaustion by using Level Zero allocations (llama/21597) Katostrofik 2026-05-14 01:39:14 -04:00
  • e414ecf674 cmake : add CMakePresets.json [no ci] (#3808) Daniel Bevenius 2026-05-25 11:25:15 +02:00
  • b3877e10c0 fix: in bindings/ruby/test/jfk_reader/jfk_reader in jfk_reader.c (#3756) OrbisAI Security 2026-05-25 11:49:23 +05:30
  • 0ccd896f5b common : fix server /inference fails to decode in-memory audio (regression) (#3818) Pascal 2026-05-22 08:27:35 +02:00
  • 8443cf05e3 ci : use github ubuntu-22.04-arm runner instead of qemu (#3815) Daniel Bevenius 2026-05-21 10:59:58 +02:00
  • afa2ea544f whisper : set bench data for each iteration (#3812) Daniel Bevenius 2026-05-19 08:58:43 +02:00
  • 47b9eb37a3 examples : fix memory leak in read_audio_data (#3810) petterreinholdtsen 2026-05-18 12:16:39 +02:00
  • 6227a0ef73 server : Return speaker information in JSON (#3782) Andreas Lubbe 2026-05-18 09:18:04 +02:00
  • 968eebe772 server: add support for carry_initial_prompt (#3781) Andreas Lubbe 2026-05-15 14:03:17 +02:00
  • 46ca43d639 talk-llama : sync llama.cpp Georgi Gerganov 2026-05-14 11:53:43 +03:00
  • 69500f5502 sync : ggml Georgi Gerganov 2026-05-14 11:53:30 +03:00
  • e4ce42e55f ggml-webgpu: only use subgroup-matrix path when head dims are divisible by sg_mat_k / sg_mat_n (llama/23020) Zheyuan Chen 2026-05-13 15:12:40 -07:00
  • 97371e9285 Fix for issue #22974. Cast intermediate results to float before adding and casting the result to the destination type. Avoids half+half operator ambiguity. (llama/22994) scutler-nv 2026-05-13 13:36:14 -07:00
  • d4a4d87f0e opencl: add q5_0 and q5_1 MoE for Adreno (llama/22985) shaofeiqi 2026-05-13 11:57:31 -07:00
  • b19beb6027 opencl: fix crash when warming up MoE on Adreno (llama/22876) lhez 2026-05-13 11:24:33 -07:00
  • 1cbbd0b6d0 flush the gpu profile timestamp before the queryset is overflowed (llama/22995) Masashi Yoshimura 2026-05-14 02:22:44 +09:00
  • cb7d38bf18 hexagon: add unary tanh op (llama/22999) Max Krasnyansky 2026-05-13 06:59:28 -07:00
  • 8b288f5d96 ggml-zendnn : adaptive fallback to CPU backend for small batch sizes (llama/22681) Sachin Sharma 2026-05-13 11:43:47 +05:30
  • bcaf449826 hexagon: eliminate scalar VTCM loads via HVX splat helpers (llama/22993) Trivikram Reddy 2026-05-12 19:28:02 -05:00
  • 1caed1d2ba opencl: add opt-in Adreno xmem F16xF32 GEMM for prefill (llama/22755) yzyyzyhhh 2026-05-13 04:10:37 +08:00
  • e8a7cd314f ggml-webgpu: Enables running gpt-oss-20b (llama/22906) Masashi Yoshimura 2026-05-12 23:27:40 +09:00
  • a9bcbf5595 ggml-webgpu: address precision issues for multimodal (llama/22808) Chen Yuan 2026-05-12 10:27:04 -04:00
  • be5a35ccee vulkan: Fix Windows performance regression on Intel GPU BF16 workloads for Xe2 and newer (llama/22461) Masato Nakasaka 2026-05-12 03:15:34 -07:00
  • 20895abdbd vulkan: Check shared memory size for mmq shaders (llama/22693) Jeff Bolz 2026-05-12 04:41:58 -05:00
  • 8ec91c91e1 metal : promote mul_mv/mul_mm batch divisors to function constants (llama/22711) guyfischman 2026-05-12 07:15:02 +02:00
  • ea4652c427 opencl: add q4_1 MoE for Adreno (llama/22856) Shawn Gu 2026-05-11 11:57:26 -07:00
  • 287f637fb1 CUDA: handle OW > 65535 in im2col (2D and 3D) (llama/22944) CrispStrobe 2026-05-11 19:48:29 +02:00
  • 449b33fc8f Ggml/cuda snake fusion hardening (llama/22912) Pascal 2026-05-11 18:42:08 +02:00
  • c0c1f994b7 vulkan: Support asymmetric FA in scalar/mmq/coopmat1 paths (llama/22589) Jeff Bolz 2026-05-11 05:49:03 -05:00
  • 0077a6d332 CUDA: directly include cuda/iterator (llama/22936) Oliver Simons 2026-05-11 12:16:38 +02:00
  • 4db2f45075 Add OP im2col_3d (llama/22903) Neo Zhang 2026-05-11 13:01:47 +08:00
  • 73f63f5295 ggml-virtgpu : include missing mutex header (llama/22810) Oliver Walsh 2026-05-10 16:32:41 +01:00
  • 5f08683bb6 metal : tighten input-position loop in kernel_conv_transpose_1d (ggml/1477) CrispStrobe 2026-05-10 16:45:00 +02:00
  • dd706793cc ggml: install ggml.pc in <libdir>/pkgconfig (ggml/1480) Steve Lhomme 2026-05-10 16:35:38 +02:00
  • 5a24c7538f Revert "ggml-cuda : add explicit casts to -INFINITY for float and half2 types" Daniel Bevenius 2026-05-14 05:28:56 +02:00
  • db7bcdb791 Revert "ggml-cuda : add ar_add() to avoid ambiguous operator+ for half/bfloat16 in CUDA 11.8" Daniel Bevenius 2026-05-14 05:27:13 +02:00
  • ea29be532e squash! ci : update ONEAPI version to 2025.3.3-0-devel-ubuntu24.04 Daniel Bevenius 2026-05-12 11:15:56 +02:00
  • d04a1faaec ci : update ONEAPI version to 2025.3.3-0-devel-ubuntu24.04 Daniel Bevenius 2026-05-12 08:36:14 +02:00
  • b6a4b32a88 ggml-cuda : add ar_add() to avoid ambiguous operator+ for half/bfloat16 in CUDA 11.8 Daniel Bevenius 2026-05-12 08:30:00 +02:00
  • b1ebddf154 ggml-cuda : add explicit casts to -INFINITY for float and half2 types Daniel Bevenius 2026-05-12 07:59:24 +02:00
  • 633de7f99e devops : add spirv-headers to vulkan dockerfile Daniel Bevenius 2026-05-12 06:38:12 +02:00
  • 5b2d4af850 Revert "try using CCCL 12.4.127 with cuda 11.8.0 to fix CI failure" Daniel Bevenius 2026-05-11 15:17:13 +02:00
  • e0bfd3ae4d try using CCCL 12.4.127 with cuda 11.8.0 to fix CI failure Daniel Bevenius 2026-05-11 14:44:23 +02:00
  • 1665885f76 Revert "try to fix window cublas CI failure" Daniel Bevenius 2026-05-11 14:39:16 +02:00