upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-28 13:41:01 -05:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • 2adf8d83ac parallel : add option for different RNG seeds (#14757) b5935 Georgi Gerganov 2025-07-18 17:33:41 +03:00
  • 021cc28bef cuda : Fix Gemma3n not executed as CUDA_GRAPH on NVGPUs (#14741) b5934 Oliver Simons 2025-07-18 13:35:32 +02:00
  • d498af3d5a graph : avoid huge warm-up graphs for MoE models (#14753) b5933 Georgi Gerganov 2025-07-18 14:31:15 +03:00
  • a856a5665d tests : add non-cont K,V FA tests Georgi Gerganov 2025-07-18 13:36:27 +03:00
  • eacdeb5bfc model : fix build after merge conflict (#14754) b5932 Georgi Gerganov 2025-07-18 11:53:55 +03:00
  • 9106d7595d model : fix build after merge conflict gg/fix-build-gf Georgi Gerganov 2025-07-18 11:50:59 +03:00
  • e0cb5c5cb8 model : add EXAONE 4.0 support (#14630) lgai-exaone 2025-07-18 17:45:49 +09:00
  • f9a31eea06 CUDA: set_rows + cpy.cu refactor (#14712) b5930 Aman Gupta 2025-07-18 14:54:18 +08:00
  • 8f974bc1e9 graph : refactor context to not pass gf explicitly (#14629) b5929 Georgi Gerganov 2025-07-18 08:29:28 +03:00
  • 09651d09ff graph : Pass the graph placeholder message in debug mode (#14748) b5928 Nexes the Elder 2025-07-18 06:25:54 +02:00
  • 349ea79fce use max work group size for device to replace the magic number (#14732) b5927 Neo Zhang Jianyu 2025-07-18 10:23:14 +08:00
  • 670e1360cd convert : fix Ernie4.5 MoE without shared experts (#14746) Piotr Wilkin (ilintar) 2025-07-18 01:17:16 +02:00
  • 760b4484e3 nix : use optionalAttrs for env mkDerivation attrset argument (#14726) Wroclaw 2025-07-18 00:18:16 +02:00
  • cb887f1bc1 model: add Ernie 4.5 MoE support (#14658) b5924 Piotr Wilkin (ilintar) 2025-07-17 23:15:32 +02:00
  • d6fb3f6b49 kv-cache : fix k-shift for multiple streams (#14742) b5923 Georgi Gerganov 2025-07-17 20:52:33 +03:00
  • 05baa62a73 kv-cache : fix k-shift for multiple streams gg/unified-fix-k-shift Georgi Gerganov 2025-07-17 20:18:36 +03:00
  • 01612b7409 llama : reuse compute graphs (#14482) b5922 Georgi Gerganov 2025-07-17 19:08:33 +03:00
  • 086cf81e88 llama : fix parallel processing for lfm2 (#14705) b5921 Tarek Dakhran 2025-07-17 09:22:11 +02:00
  • d9b691081c kv-cache : opt mask set input (#14600) b5920 Georgi Gerganov 2025-07-17 09:49:15 +03:00
  • ad57d3edd2 batch : fix uninitialized has_cpl flag (#14733) b5919 Georgi Gerganov 2025-07-17 09:45:54 +03:00
  • 1ba45d4982 ci : disable failing vulkan crossbuilds (#14723) Sigbjørn Skjæret 2025-07-17 01:52:08 +02:00
  • 19e5943d9e convert : make hf token optional (#14717) Sigbjørn Skjæret 2025-07-16 23:17:43 +02:00
  • 496957e1cb llama : fix parameter order for hybrid memory initialization (#14725) b5916 Diner Burger 2025-07-16 15:17:25 -04:00
  • 21c021745d ggml: Add initial WebGPU backend (#14521) b5915 Reese Levine 2025-07-16 08:18:51 -07:00
  • b0f0ecc3dc model : support output bias for qwen2 (#14711) b5914 tempstudio 2025-07-16 10:02:06 -05:00
  • 225e7a1438 llama : add high-throughput mode (#14363) b5913 Georgi Gerganov 2025-07-16 16:35:42 +03:00
  • ab14019821 Support diffusion models: Add Dream 7B (#14644) b5912 Aman Gupta 2025-07-16 20:03:51 +08:00
  • 64978340b0 ggml : add asserts (#14720) b5911 Georgi Gerganov 2025-07-16 14:43:32 +03:00
  • 6ffd4e9c44 server : pre-calculate EOG logit biases (#14721) b5910 Georgi Gerganov 2025-07-16 14:04:12 +03:00
  • 07908a824a server : pre-calculate EOG logit biases gg/server-eos-pre-calc Georgi Gerganov 2025-07-16 13:47:05 +03:00
  • e4841d24d3 llama : fix parallel processing for plamo2 (#14716) b5909 Shunta Saito 2025-07-16 19:12:22 +09:00
  • 538cc77f7f server : fix handling of the ignore_eos flag (#14710) b5908 Georgi Gerganov 2025-07-16 12:13:57 +03:00
  • 5cae766541 scripts: synthetic prompt mode for server-bench.py (#14695) Johannes Gäßler 2025-07-16 09:33:28 +02:00
  • 4b91d6f71f convert : only check for tokenizer folder if we need it (#14704) Sigbjørn Skjæret 2025-07-16 08:52:04 +02:00
  • cf91f217f1 convert : add pre-computed hashes first to prevent order mishaps (#14701) Sigbjørn Skjæret 2025-07-16 08:51:12 +02:00
  • 9f8d285901 server : fix handling of the ignore_eos flag gg/server-fix-ignore-eos Georgi Gerganov 2025-07-16 07:37:18 +03:00
  • 79e0b68c17 llama: add LLAMA_API to deprecated llama_kv_self_seq_div (#14708) b5904 Min-Hua 2025-07-16 12:00:42 +08:00
  • c81f4192f9 gguf-py : dump bpw per layer and model in markdown mode (#14703) Ed Addario 2025-07-15 23:04:42 +01:00
  • 4a4f426944 model : add Kimi-K2 support (#14654) b5902 Gabriel Larson 2025-07-15 14:54:22 -05:00
  • ba1ceb3456 vulkan: fix noncontig check for mat_mul_id splitting (#14683) b5901 Jeff Bolz 2025-07-15 14:51:09 -05:00
  • 10a0351a97 vulkan: add RTE variants for glu/add/sub/mul/div (#14653) b5900 Jeff Bolz 2025-07-15 14:32:11 -05:00
  • 68e37a61a7 model : add PLaMo-2 support (#14560) b5899 Shunta Saito 2025-07-16 01:11:42 +09:00
  • f68669d50f fix and opt kernel launch opencl-add-mul-mat-f16-f32-image rmatif 2025-07-15 11:28:26 +00:00
  • cbc68be51d cuda: fix build warnings in set-rows.cu (unused variable) (#14687) b5898 R0CKSTAR 2025-07-15 15:28:53 +08:00
  • bdca38376f sycl: Hotfix for non dnnl codepath (#14677) b5897 Anton Mitkov 2025-07-14 18:12:42 +01:00
  • 55c509daf5 ggml : refactor llamafile_sgemm PPC code (#14673) b5896 shalinib-ibm 2025-07-14 18:46:42 +05:30
  • 9c9e4fc635 llama-context: add ability to get logits (#14672) b5895 Aman Gupta 2025-07-14 21:01:41 +08:00
  • 494c5899cb scripts: benchmark for HTTP server throughput (#14668) b5894 Johannes Gäßler 2025-07-14 13:14:30 +02:00
  • 0f4c6ec0f1 SYCL: use 1D kernel for set_rows (#14618) b5893 Akarshan Biswas 2025-07-14 15:07:55 +05:30
  • 65a3ebb0aa sycl: Batched mulmat rework for oneDNN dispatch (#14617) b5892 Anton Mitkov 2025-07-14 10:37:35 +01:00
  • 0d9226763c llama : add jinja template for rwkv-world (#14665) b5891 Molly Sophia 2025-07-14 07:43:43 +08:00
  • 982e347255 quantize : fix minor logic flaw in --tensor-type (#14572) b5890 Ed Addario 2025-07-13 17:02:17 +01:00
  • 923e3ea2e3 cuda : add set rows for bf16 (#14664) b5889 Sigbjørn Skjæret 2025-07-13 15:01:24 +02:00
  • e743cddb60 cuda : add ELU support (#14657) b5888 Yavor Ivanov 2025-07-13 02:33:16 -07:00
  • 05fec5bd29 ggml : add build-time message to remind about ggml_set_rows (#14661) b5887 Georgi Gerganov 2025-07-13 10:36:33 +03:00
  • dcf7f2ea3c metal : Add missing unary ops Metal support (#14660) b5886 Yavor Ivanov 2025-07-12 22:38:13 -07:00
  • 84b396e051 cmake : Add CMake presets for Linux and GCC (#14656) Yavor Ivanov 2025-07-12 22:12:36 -07:00
  • 942c55cd57 imatrix : avoid using imatrix.dat in README compilade/imatrix-batched-chunks Francis Couture-Harpin 2025-07-12 14:56:18 -04:00
  • 183eeb5518 imatrix : avoid loading model to convert or combine imatrix Francis Couture-Harpin 2025-07-12 14:54:33 -04:00
  • 50f53b3e40 imatrix : warn when writing partial data, to help guess dataset coverage Francis Couture-Harpin 2025-07-12 14:09:28 -04:00
  • 42423ec4d3 imatrix : add warning when legacy format is written Francis Couture-Harpin 2025-07-12 13:42:35 -04:00
  • 0ee322cd0f Merge branch 'master' into compilade/imatrix-batched-chunks Francis Couture-Harpin 2025-07-12 13:31:19 -04:00
  • c31e60647d tests : cover lfm2 cases in test_ssm_conv (#14651) b5884 Tarek Dakhran 2025-07-12 19:10:14 +02:00
  • 67eade1bf9 docs : add LFM2 to models section (#14650) Tarek Dakhran 2025-07-12 19:07:08 +02:00
  • 7de5c7cab6 CUDA: add set rows for f32 and f16 (#14551) b5882 Aman Gupta 2025-07-12 21:31:38 +08:00
  • 8eff95544e sync : ggml Georgi Gerganov 2025-07-12 16:06:12 +03:00
  • 3120413ccd vulkan : remove unused vars (#0) b5880 Georgi Gerganov 2025-07-12 12:39:32 +03:00
  • 215535701d sync : ggml Georgi Gerganov 2025-07-12 12:39:27 +03:00
  • 74bb294591 vulkan : implement bilinear interpolation (ggml/1291) Acly 2025-07-12 12:37:37 +03:00
  • 3e303b1107 vulkan : implement ggml_roll (ggml/1290) Acly 2025-07-12 12:32:32 +03:00
  • 0c1df14b5f server : fix pooled embedding output (#14645) b5876 Douglas Hanley 2025-07-12 06:21:02 -04:00
  • b3ad3a0191 vulkan: support SET_ROWS (#14587) b5875 Jeff Bolz 2025-07-12 05:12:26 -05:00
  • 98197e5c98 vulkan: optimizations for deepseek prompt processing (#14555) b5874 Jeff Bolz 2025-07-12 04:51:58 -05:00
  • f5e96b368f model : support LiquidAI LFM2 hybrid family (#14620) b5873 Tarek Dakhran 2025-07-11 20:27:01 +02:00
  • 756aa1020a HIP : Add HIP 7.0+ compatibility for hipBLAS compute types (#14634) b5872 Slobodan Josic 2025-07-11 18:55:00 +02:00
  • 18310bf202 fix trailing whitespace rmatif 2025-07-11 13:13:54 +00:00
  • aaa088d87f readme : add hot PRs (#14636) Georgi Gerganov 2025-07-11 16:07:55 +03:00
  • 418606e8f2 add mul_mat_f16_f32_image kernel rmatif 2025-07-11 10:50:01 +00:00
  • 0d5375d54b llama : move enum llama_vocab_pre_type to implementation (#14631) b5870 Georgi Gerganov 2025-07-11 13:46:07 +03:00
  • 576c82eda2 vocab : add midm-2.0 model pre-tokenizer (#14626) b5869 Dowon 2025-07-11 16:36:04 +09:00
  • 0aedae00e6 model : Granite Four (#13550) b5868 Gabe Goodhart 2025-07-10 18:20:13 -06:00
  • 6bdda13981 opencl: add tiled mul_mat_f16_f32 (#14535) b5867 rmatif 2025-07-10 23:58:12 +02:00
  • 0b8855775c opencl: add set_rows for f16 and f32 (#14547) b5866 lhez 2025-07-10 11:48:52 -07:00
  • 4bb625b713 Smoldocling support (#14597) b5865 Ryan Mangeno 2025-07-10 13:41:00 -04:00
  • 11ee0fea2a Docs: script to auto-generate ggml operations docs (#14598) b5864 Aman Gupta 2025-07-10 23:29:01 +08:00
  • a457551332 cmake : do not search for curl libraries by ourselves (#14613) b5863 Eric Zhang 2025-07-10 20:29:05 +08:00
  • 704bb7a71c SYCL: Initial set_rows kernel implementation (#14562) b5862 Akarshan Biswas 2025-07-10 13:59:38 +05:30
  • 435a6d10d6 llama : minor coding style fix for smollm3 (#14605) b5861 Xuan-Son Nguyen 2025-07-10 09:00:20 +02:00
  • f9a867f592 cmake : bump llguidance version to v1.0.1 (#14609) b5860 Eric Zhang 2025-07-10 13:19:37 +08:00
  • ac44eb6c80 cmake : llguidance build parser library only (#14608) b5859 Eric Zhang 2025-07-10 13:19:13 +08:00
  • a57d1bcb3c cuda : support Falcon-H1 state size for SSM_SCAN (#14602) b5858 compilade 2025-07-09 23:54:38 -04:00
  • cb9178f885 llama : remove llm_graph_input_one (#14603) b5857 Xuan-Son Nguyen 2025-07-09 23:09:28 +02:00
  • 4a5686da22 llama : support Jamba hybrid Transformer-Mamba models (#7531) b5856 compilade 2025-07-09 14:59:57 -04:00
  • 1180752835 cuda : support Falcon-H1 state size for SSM_SCAN compilade/cuda-falcon-h1 Francis Couture-Harpin 2025-07-09 12:18:37 -04:00
  • 98bab638fb ggml : add ggml_scale_bias (#14417) b5855 Xuan-Son Nguyen 2025-07-09 18:16:12 +02:00
  • 4d6a179c68 gguf-py : avoid adding duplicate tensor mappings for Jamba compilade/refactor-kv-cache Francis Couture-Harpin 2025-07-09 11:58:35 -04:00
  • 452207f318 memory : avoid referring to KV in recurrent cache logs Francis Couture-Harpin 2025-07-09 10:05:35 -04:00
  • 7f3955a068 model : make falcon-h1 use shared mamba2 layer builder Francis Couture-Harpin 2025-07-09 09:44:37 -04:00
  • a60a24beed Merge branch 'master' into compilade/refactor-kv-cache Francis Couture-Harpin 2025-07-09 09:38:48 -04:00
  • 26a48ad699 ggml : prevent integer overflow in gguf tensor size calculation (#14595) b5854 Miaoqian Lin 2025-07-09 20:33:53 +08:00