Commit Graph

  • 5d950c4b8d whisper : use ggml_backend_sched (#2239) Georgi Gerganov 2024-06-18 09:37:20 +03:00
  • 820446e230 fix : remove extra files Georgi Gerganov 2024-06-16 19:23:55 +03:00
  • 54d5823ebe scripts : sync ggml-blas Georgi Gerganov 2024-06-16 19:23:32 +03:00
  • 5181494e9f build : update make / cmake Georgi Gerganov 2024-06-16 19:10:20 +03:00
  • 4a6e6e8b30 sync : ggml Georgi Gerganov 2024-06-16 18:40:07 +03:00
  • de29b193f6 move BLAS to a separate backend (cont) (llama/6210) slaren 2024-06-16 13:57:37 +03:00
  • 922971041b Vulkan Shader Refactor, Memory Debugging Option (llama/7947) 0cc4m 2024-06-16 07:17:31 +02:00
  • 63a767a134 scripts : stop sync whisper example from ggml Georgi Gerganov 2024-06-16 18:38:46 +03:00
  • 30841fa786 cmake : fix sycl build (#0) Georgi Gerganov 2024-06-16 17:57:35 +03:00
  • 3b1ac03828 ggml : remove OpenCL (#0) Georgi Gerganov 2024-06-16 13:46:12 +03:00
  • 990de617b5 sycl : sync (#0) Georgi Gerganov 2024-06-16 13:24:17 +03:00
  • 6975600b4b cuda : enable CUDA graphs (#0) Georgi Gerganov 2024-06-16 13:20:19 +03:00
  • 061eeb9f61 talk-llama : sync llama.cpp Georgi Gerganov 2024-06-16 13:10:54 +03:00
  • 4942b1b428 cmake : fix CUDA build (#0) Georgi Gerganov 2024-06-16 13:07:43 +03:00
  • 3c7cc5c437 sync : ggml Georgi Gerganov 2024-06-16 12:43:14 +03:00
  • 5cd42ee2cc ggml : fix and optimize ppc64le (ggml/849) Hong Bo PENG 2024-06-16 16:53:11 +08:00
  • ee718f3da6 ggml : remove duplicate include of ggml-common.h (ggml/853) Daniel Bevenius 2024-06-16 10:51:18 +02:00
  • 63eac1f608 remove global variables (llama/7710) Meng, Hengyu 2024-06-15 14:05:10 +08:00
  • b17ba2815b CUDA: faster q2_K, q3_K MMQ + int8 tensor cores (llama/7921) Johannes Gäßler 2024-06-14 18:41:49 +02:00
  • 7a489af2f3 metal : utilize max shared memory for mul_mat_id (llama/7935) Georgi Gerganov 2024-06-14 17:14:09 +03:00
  • 4a4ea13d6d rpc : fix ggml_backend_rpc_supports_buft() (llama/7918) Radoslav Gerganov 2024-06-13 15:18:44 +03:00
  • 174a461fc6 move BLAS to a separate backend (llama/6210) slaren 2024-06-13 03:11:35 +02:00
  • d8b7a24bc9 CUDA: fix broken oob check for FA vec f32 kernel (llama/7904) Johannes Gäßler 2024-06-12 17:41:51 +02:00
  • acf3832c9c tests : add non-cont unary tests (llama/7857) Georgi Gerganov 2024-06-12 16:00:22 +03:00
  • d29ac44303 ggml : improve ggml_is_contiguous logic (llama/7856) Georgi Gerganov 2024-06-12 15:24:20 +03:00
  • 12638dfef0 vulkan: select only one device for single gpu with multiple drivers (llama/7582) k.h.lai 2024-06-12 03:26:05 +08:00
  • f100b3b523 Update Vulkan RoPE implementation (llama/7818) 0cc4m 2024-06-11 21:20:29 +02:00
  • a99e213a82 CUDA: int8 tensor cores for MMQ (q4_K, q5_K, q6_K) (llama/7860) Johannes Gäßler 2024-06-11 08:26:07 +02:00
  • 7483d2b61c CUDA: use tensor cores for MMQ (llama/7676) Johannes Gäßler 2024-06-10 11:45:13 +02:00
  • 1fe5948227 use the correct SYCL context for host USM allocations (llama/7777) Ben Ashbaugh 2024-06-10 02:21:31 -07:00
  • 760497e1ab CUDA: revise q8_1 data layout for mul_mat_q (llama/7824) Johannes Gäßler 2024-06-09 09:42:25 +02:00
  • b172e7714c vulkan : reuse parent extra for views (llama/7806) slaren 2024-06-07 19:47:49 +02:00
  • dc01aadb18 fix softmax r2r result wrong issue (llama/7811) pengxin99 2024-06-07 14:28:26 +08:00
  • e08c62149b CUDA: refactor mmq, dmmv, mmvq (llama/7716) Johannes Gäßler 2024-06-05 16:53:00 +02:00
  • abab4500fa ggml : refactor rope norm/neox (llama/7634) Georgi Gerganov 2024-06-05 11:29:20 +03:00
  • e666315fa8 Allow number of nodes in CUDA graph to change (llama/7738) agray3 2024-06-04 21:06:49 +01:00
  • 3f869af14c ggml : remove OpenCL (llama/7735) Georgi Gerganov 2024-06-04 21:23:20 +03:00
  • cbacb7634c ggml : prevent builds with -ffinite-math-only (llama/7726) Georgi Gerganov 2024-06-04 10:01:09 +03:00
  • 6cc3b022ee llama : offload to RPC in addition to other backends (llama/7640) Radoslav Gerganov 2024-06-03 20:03:26 +03:00
  • e5e38d4920 ggml : use OpenMP as a thread pool (llama/7606) Masaya, Kato 2024-06-04 00:14:15 +09:00
  • 2a6bab5655 Vulkan Mixture of Experts (MoE) support (llama/7628) 0cc4m 2024-06-03 10:59:14 +02:00
  • 8c01c9b85c kompute : implement op_getrows_f32 (llama/6403) woachk 2024-06-03 07:32:16 +02:00
  • d1123d795e fix bug introduced in using calloc (llama/7701) Dave Airlie 2024-06-03 07:59:54 +10:00
  • 9b3d784020 Fix FlashAttention debug test, FP32 assert (llama/7684) Johannes Gäßler 2024-06-01 23:26:10 +02:00
  • a16137d13d CUDA: fix Pascal FA, deq. KV to FP16 for batch > 8 (llama/7681) Johannes Gäßler 2024-06-01 15:47:04 +02:00
  • 5582039d0a CUDA: quantized KV support for FA vec (llama/7527) Johannes Gäßler 2024-06-01 08:44:14 +02:00
  • 9a16c643e2 ggml : fix loongson compile warnings (llama/7537) Georgi Gerganov 2024-05-31 14:17:10 +03:00
  • 10a8a23100 faster avx512 exp implementation (llama/7551) Chris Elrod 2024-05-30 07:32:55 -04:00
  • 29cfeef77f ggml : fix loongarch build (O2 issue) (llama/7636) junchao-loongson 2024-05-30 17:30:10 +08:00
  • e66e9ea25b metal : remove invalid asserts (llama/7617) Georgi Gerganov 2024-05-29 22:20:40 +03:00
  • 276779a849 metal : add missing asserts (llama/7617) Georgi Gerganov 2024-05-29 20:45:25 +03:00
  • 1f35ce61c1 ggml : fix YARN + add tests + add asserts (llama/7617) Georgi Gerganov 2024-05-29 20:17:31 +03:00
  • 4b19cc3ed4 cuda : non-cont concat support (llama/7610) Georgi Gerganov 2024-05-29 15:38:26 +03:00
  • a535d348dd llama-bench : add support for the RPC backend (llama/7435) Radoslav Gerganov 2024-05-29 14:45:44 +03:00
  • 8f5dc729d9 ggml : use atomic_flag for critical section (llama/7598) slaren 2024-05-29 13:36:39 +02:00
  • 02fc147a0b examples : adapt to new ggml_concat (ggml/0) Georgi Gerganov 2024-05-29 12:58:00 +03:00
  • 109148ac84 ggml : fix typo in ggml.c (llama/7603) zhouwg 2024-05-29 10:09:31 +08:00
  • 3563473d2c Align GEMM dispatch (llama/7566) Meng, Hengyu 2024-05-29 07:00:24 +08:00
  • 046834198d sycl : fix assert (llama/7563) Georgi Gerganov 2024-05-28 22:22:50 +03:00
  • 0a2ad9de06 vulkan: properly initialize vulkan devices for LLAMA_SPLIT_MODE_NONE (llama/7552) k.h.lai 2024-05-29 01:25:08 +08:00
  • 39b0640b09 rpc : resource management rework (llama/7562) Radoslav Gerganov 2024-05-28 18:13:36 +03:00
  • 8dca71de64 fix ggml_sycl_mul_mat_id() to match the change of api (llama/7436) Neo Zhang 2024-05-28 17:53:37 +08:00
  • 812787cbc5 ggml : generalize GGML_OP_CONCAT (llama/7563) Georgi Gerganov 2024-05-28 11:04:19 +03:00
  • 68ef10805e update HIP_UMA #7399 (llama/7414) Djip007 2024-05-28 01:40:47 +02:00
  • 96fdb90f5f Allow multiple copy function pointers for CUDA graph kernel param updates (llama/7565) agray3 2024-05-27 18:33:42 +01:00
  • e98f9ac554 Fix q_xxs using mul_mat_q (llama/7459) AidanBeltonS 2024-05-27 17:34:51 +01:00
  • 02d481595b Add freq factors (llama/7495) AidanBeltonS 2024-05-27 13:34:09 +01:00
  • 7091c7ab5a metal : add GGML_OP_REPEAT kernels (llama/7557) Georgi Gerganov 2024-05-27 12:10:19 +03:00
  • d70ccb75f5 metal : disable FA kernel for HS=256 (llama/7556) Georgi Gerganov 2024-05-27 10:38:39 +03:00
  • 5ee048eb67 ggml : restore ggml_rope_xpos_inplace (ggml/0) Georgi Gerganov 2024-05-26 18:35:23 +03:00
  • 37ed71c964 ggml: aarch64: SVE kernels for q8_0_q8_0, q4_0_q8_0 vector dot (llama/7433) Masaya, Kato 2024-05-25 17:42:31 +09:00
  • 8cd7a3df37 ggml : silence UB sanitizer error during iq2_xxs quantization (llama/0) Georgi Gerganov 2024-05-23 17:17:43 +03:00
  • 04a3279320 ggml : remove ggml_flash_attn and ggml_flash_ff (llama/7463) Georgi Gerganov 2024-05-23 10:00:44 +03:00
  • 45ddda8e0c ggml : drop support for QK_K=64 (llama/7473) Georgi Gerganov 2024-05-23 10:00:21 +03:00
  • c41317fd66 Update vulkan rope implementation to support frequency factors (llama/7475) 0cc4m 2024-05-23 08:59:59 +02:00
  • 96b8419b27 CUDA: fix FA out-of-bounds reads (llama/7479) Johannes Gäßler 2024-05-23 00:31:20 +02:00
  • 3c63f4cf35 CUDA: fix FA out-of-bounds writes (llama/7465) Johannes Gäßler 2024-05-22 17:58:25 +02:00
  • 5848dfd9c8 cuda : fix compile warning (llama/7454) Georgi Gerganov 2024-05-22 12:36:37 +03:00
  • 29ab5d0326 CUDA: remove incorrect precision check (llama/7454) Johannes Gäßler 2024-05-22 10:24:29 +02:00
  • c4d6958b3e cuda : fix rope + add tests (llama/7452) Georgi Gerganov 2024-05-22 11:01:35 +03:00
  • c9dcb75118 llama : add phi3 128K model support (llama/7225) liuwei-git 2024-05-22 04:28:32 +08:00
  • bbdbc3fc62 metal : handle F16 inf values, fix FA partial offload (llama/7434) Georgi Gerganov 2024-05-21 23:03:42 +03:00
  • 28c207a541 CUDA: fix unused warning in mmq.cu (llama/7442) Johannes Gäßler 2024-05-21 19:27:12 +02:00
  • c23f830983 CUDA: deduplicate mmq code (llama/7397) Johannes Gäßler 2024-05-21 16:02:12 +02:00
  • caeeb32b41 rpc : track allocated buffers (llama/7411) Radoslav Gerganov 2024-05-20 16:36:55 +03:00
  • 584cc1177a Update SYCL upscale operation (llama/7321) AidanBeltonS 2024-05-20 12:08:23 +01:00
  • cc1ae10989 ggml-opencl, llama: using reserve() if count already known (llama/7272) Herman Semenov 2024-05-20 07:33:21 +00:00
  • eb26f55b40 ggml : add loongarch lsx and lasx support (llama/6454) junchao-loongson 2024-05-20 15:19:21 +08:00
  • eb2b086584 Add provisions for windows support for BF16 code including CMake provision for enabling AVX512_BF16 (llama/7258) Srihari-mcw 2024-05-19 19:18:39 -07:00
  • 67919cfe11 Vulkan Embedding Fix (llama/7360) 0cc4m 2024-05-19 17:19:53 +02:00
  • bf5fc81a8a ggml : fix another case of quants nans (llama/7387) slaren 2024-05-19 17:08:46 +02:00
  • 2b07dc3186 ggml: implement quantized KV cache for FA (llama/7372) Johannes Gäßler 2024-05-19 16:46:13 +02:00
  • 951c463d39 cuda : clear error after buffer allocation failure (llama/7376) slaren 2024-05-19 14:19:37 +02:00
  • 7f257b210f Capture CUDA logging output (llama/7298) fraxy-v 2024-05-19 01:44:42 +03:00
  • 705fe30a02 android : use "ci-android" branch for CI (llama/7341) Georgi Gerganov 2024-05-18 13:40:39 +03:00
  • 45b5b95e29 CUDA: deduplicate FlashAttention code (llama/7352) Johannes Gäßler 2024-05-18 12:36:25 +02:00
  • f2c47d1e6a cuda : add half2 __shfl_xor() for ROCm 5.5 (llama/7263) Engininja2 2024-05-18 02:05:17 -06:00
  • b4bb9b9036 Update and fix Vulkan soft_max and argsort implementations (llama/7237) 0cc4m 2024-05-18 08:10:58 +02:00
  • 2bc6483299 ggml : fix quants nans when all the group weights are very close to zero (llama/7313) slaren 2024-05-18 02:39:54 +02:00
  • ec52f900e4 CUDA: faster large batch FA without tensor cores (llama/7314) Johannes Gäßler 2024-05-17 18:54:52 +02:00