Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-19 17:24:57 +02:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • eb5667a169 convert : fix DeepseekV4 rope parameters with transformers 5.x (#26673) Sigbjørn Skjæret 2026-08-06 15:06:52 +02:00
  • 3db4ff877d model-loader : fix quantized reshaped tensor strides (#26672) b10295 Georgi Gerganov 2026-08-06 15:21:44 +03:00
  • d14bb68e5f mtmd: carry the remaining pocket-tts per-pack settings Pascal 2026-08-06 13:50:09 +02:00
  • 1a0e158115 fix it xsn/mmid_dup_idx_fix Xuan Son Nguyen 2026-08-06 13:30:01 +02:00
  • e700bfb37f convert : accept "ExaoneMoeForCausalLM" arch spelling (#26660) Csaba Kecskemeti 2026-08-06 03:56:04 -07:00
  • a1f96d4fc2 ci : onboard AMD ROCm CI with gfx1151 fixes (#26544) b10293 Jim Wu 2026-08-06 01:43:26 -07:00
  • 9de0fcf2b3 model-conversion : add --model-name to conversion scripts (#26665) Daniel Bevenius 2026-08-06 09:38:06 +02:00
  • 803b7fcae8 vulkan: fix submission batching size, add debug tools for diagnosing causes of DeviceLost drivers errors (#26371) b10291 Ruben Ortlam 2026-08-06 09:24:13 +02:00
  • 6751477adf cont : avoid copy_state pr/25532-fix Georgi Gerganov 2026-08-06 10:19:52 +03:00
  • f8fa4df82a test: add duplicated index case for ggml_mul_mat_id xsn/mmid_dup_idx Xuan Son Nguyen 2026-08-06 01:16:17 +02:00
  • f7655b451d server: fix empty response for /cors-proxy xsn/server_proxy_fix_204 Xuan Son Nguyen 2026-08-06 01:09:24 +02:00
  • c8e03ce812 mtmd/ggml: add ggml_build_forward_order (#26649) b10290 Pascal 2026-08-06 00:47:59 +02:00
  • 91f82eb1b4 mtmd: fix granite 4v grid assembly xsn/grante4v-fix-assembly Xuan Son Nguyen 2026-08-06 00:42:06 +02:00
  • c898cb9093 chunking Xuan Son Nguyen 2026-08-06 00:27:24 +02:00
  • b282950725 flow_temp + frames_after_eos Xuan Son Nguyen 2026-08-06 00:13:26 +02:00
  • f9e832c10e server: harden the file_glob_search directory walk (#26626) b10289 Pascal 2026-08-05 21:31:54 +02:00
  • 9ff97e3157 mtmd: build the pocket-tts transposed convolutions as GEMM + col2im Pascal 2026-08-05 21:05:50 +02:00
  • 9fcc9117cf Address review comments Gaurav Garg 2026-08-06 00:25:11 +05:30
  • 8580cc0b81 Rebase and address review comments Gaurav Garg 2026-08-04 20:54:57 +05:30
  • fea0c3d410 Rebase changes Gaurav Garg 2026-08-02 12:18:06 +05:30
  • 483169a47e More test fixes Gaurav Garg 2026-07-15 22:56:35 +05:30
  • d9fe17bdad Fix tests on Vulkan Gaurav Garg 2026-07-15 16:15:07 +05:30
  • 9c22762603 Simpify some of the changes Gaurav Garg 2026-07-10 21:48:33 +05:30
  • 3e753aa27d Fix CPU and backend sampling mismatches Gaurav Garg 2026-07-10 18:15:30 +05:30
  • a24097dc57 Match dist between CPU and GPU Gaurav Garg 2026-07-07 20:54:05 +05:30
  • 813c39b225 Don't reuse memory for output views. Gaurav Garg 2026-07-06 17:28:14 +05:30
  • f1f008b2ee More fixes Gaurav Garg 2026-07-05 23:15:39 +05:30
  • eab4e8444f Add a numeric context parameter declaring the maximum outputs one sequence Gaurav Garg 2026-07-05 02:48:13 +05:30
  • 0eddfa0b4c Clamp the mask sum before converting it into the sampled index Gaurav Garg 2026-07-05 02:06:52 +05:30
  • 167d5e141b Enable backend sampling with token speculation Gaurav Garg 2026-07-01 18:21:47 +05:30
  • 360e1349f0 tests: re-enable MiniMax M3 in test-llama-archs (#26633) b10288 Niklas Wenzel 2026-08-05 17:58:34 +02:00
  • f1dffb70d9 mtmd: fix longest_edge ignoring min/max pixels Xuan Son Nguyen 2026-08-05 15:56:21 +02:00
  • b06aa774c0 mtmd: Unlimited-OCR fix max_tiles, setting in converter (#25614) b10287 Saba Fallah 2026-08-05 15:30:14 +02:00
  • 915b3a1da7 working impl, need verify and clean up Xuan Son Nguyen 2026-08-05 15:06:40 +02:00
  • cd0fa6051a grammar : degrade max repetition >= 2000 to unbounded (#26613) b10286 Aldehir Rojas 2026-08-05 07:39:10 -05:00
  • 0377fc17ad sampler : add llama_sampler_copy gg/llama-sampler-copy Georgi Gerganov 2026-08-05 14:40:53 +03:00
  • 717dad5c8e mtmd: support multi-row batching for deepseek-ocr (#26154) b10285 Xuan-Son Nguyen 2026-08-05 13:34:52 +02:00
  • 9a688e51e6 fit: Fix memory allocation for MTP layers (#26605) b10284 Sergey Malinin 2026-08-05 14:29:45 +03:00
  • 9303cdd8d3 security : clarify about AI-generated reports (#26579) Xuan-Son Nguyen 2026-08-05 13:27:06 +02:00
  • eeed56f725 sampler : add llama_sampler_copy() Georgi Gerganov 2026-08-05 14:27:01 +03:00
  • e77293d4cc mtmd: weave deepseek-ocr rows in one shot instead of per row (#26615) xsn/ds_ocr_batch Saba Fallah 2026-08-05 13:07:53 +02:00
  • a035a88878 server: Adding spec-decode counters to /metrics endpoint (#26389) b10282 Bhavik Sharda 2026-08-05 16:06:01 +05:30
  • 020760adfc convert: Add endianness conversion for Q1 and TQ2 quantizations (#26618) Andreas Krebbel 2026-08-05 12:06:09 +02:00
  • 61881b1f7f vendor : apply patches for subprocess.h (#26606) b10280 Xuan-Son Nguyen 2026-08-05 11:26:20 +02:00
  • 3e3a7a416d ui: show generation statistics by default in chat settings (#26624) Aleksander Grygier 2026-08-05 11:03:23 +02:00
  • d52ec04a66 build : remove GGML_METAL_USE_BF16 from all build scripts (#26604) b10278 Niklas Wenzel 2026-08-05 10:44:34 +02:00
  • 8045779cff cleanup stack/metal-tuner forforever73 2026-08-03 11:44:22 +08:00
  • 9be171f5fa abort on unknown KV type, single-source fa_vec_legal_ne forforever73 2026-08-02 12:02:07 +08:00
  • ff024685a3 tools : document the Metal tuner, point the table comment at it forforever73 2026-08-02 00:15:20 +08:00
  • 26edf081d8 test-backend-ops : replace the FA vec tune mode with a bounded (Q,NE) slice forforever73 2026-07-31 19:37:12 +08:00
  • 35c08cd022 cool down and re-measure the dirty window on thermal drift forforever73 2026-07-31 16:58:56 +08:00
  • f44d2ee27c tools : add FA-vec (Q,NE) sweep, compression and table emit forforever73 2026-07-31 16:33:06 +08:00
  • 8e7831fd3b add FA-vec graph build + tensor init to the tuner forforever73 2026-07-31 15:41:18 +08:00
  • cf80ea68c1 add op-agnostic perf cell + median timing for the tuner forforever73 2026-07-31 14:56:55 +08:00
  • 55318e3b01 add ggml-metal-tuning skeleton forforever73 2026-07-31 14:26:04 +08:00
  • eb052925e1 add device id -> token lookup for the offline tuning tool forforever73 2026-07-31 11:25:03 +08:00
  • 0f2793504e metal : per-device tuned (Q, NE) for flash-attn vec (#25750) YiChen Lv 2026-07-29 21:46:01 +08:00
  • 5522498343 metal : port new kernels into the split sources forforever73 2026-08-04 14:50:51 +08:00
  • db0a8ce844 ggml-metal: FWHT kernel for metal backend (#25924) Georgi Gerganov 2026-07-28 19:05:42 +03:00
  • b2d79a7e7f metal: fuse snake activation (mul, sin, sqr, mul, add) (#25459) Georgi Gerganov 2026-07-28 19:02:22 +03:00
  • 0b6cbb572c metal : add Q2_0 support (#25419) Georgi Gerganov 2026-07-28 19:00:04 +03:00
  • 7649856ea2 metal : add CONV_2D_DW (depthwise convolution) support (#21565) Georgi Gerganov 2026-07-10 14:51:31 +03:00
  • f348d51391 metal : add set_rows with src0 f16 (#25434) Georgi Gerganov 2026-07-08 16:51:31 +03:00
  • d2112aa30e metal: add col2im_1d op (f32/f16/bf16) (#25176) Georgi Gerganov 2026-07-08 16:50:17 +03:00
  • ff8169ce01 metal : per-op source split + parallel compile (#24021) YiChen Lv 2026-06-20 18:36:32 +08:00
  • e031d95679 ui: Update vulnerable packages + cleanup Storybook config (#26607) Aleksander Grygier 2026-08-05 08:06:37 +02:00
  • a0c869f043 text model ok Xuan Son Nguyen 2026-08-05 02:41:43 +02:00
  • f55a236349 adapt the api Xuan Son Nguyen 2026-08-05 02:00:52 +02:00
  • ac6c66699d vendor : apply patches for subprocess.h xsn/subproc_patches Xuan Son Nguyen 2026-08-05 01:34:32 +02:00
  • cf5d754e11 upload speaker_ref via form-data Xuan Son Nguyen 2026-08-05 01:09:32 +02:00
  • ae5217d834 wire up mtmd_helper_model_can_chat Xuan Son Nguyen 2026-08-05 00:52:07 +02:00
  • 83880383ce add docs Xuan Son Nguyen 2026-08-05 00:51:58 +02:00
  • 6ea215d171 Prefer npm ci over install for security (#26601) b10276 Evan Huus 2026-08-04 18:14:22 -04:00
  • 4308a4f035 server: decode Windows OEM output to UTF-8 in built-in tools (#26597) b10275 Pascal 2026-08-04 22:24:55 +02:00
  • 6ac1a33adc add /tts endpoint Xuan Son Nguyen 2026-08-04 22:13:56 +02:00
  • 474c92e722 mtmd: correcting duplicate empty audio chunks for short inputs (#26536) b10274 Abhinay Krishna 2026-08-04 16:05:56 -04:00
  • a6aa6f5450 sampler : remove "full-context windows" from history-based samplers (#26524) b10273 Oliver Simons 2026-08-04 20:28:55 +02:00
  • 76c956c137 gguf-split: Add option to delete split parts during merge (#26538) Guilherme Quintino 2026-08-04 19:27:47 +01:00
  • 2f56fc3431 ui: CWD for agent (#26518) b10271 Aleksander Grygier 2026-08-04 19:05:48 +02:00
  • 499e2417a1 mtmd: add audio out stream api Xuan Son Nguyen 2026-08-04 18:10:04 +02:00
  • 0713275082 mtmd: support Qwen3-TTS (note: breaking change to llama-tts binary) (#26254) b10270 Xuan-Son Nguyen 2026-08-04 17:26:15 +02:00
  • 58137f0f47 quant : allow quantization of ffn_gate_inp tensors gg/quant-allow-ffn-gate-inp Georgi Gerganov 2026-08-04 18:11:42 +03:00
  • 285941de20 nits 2 xsn/security_policy_ai Xuan Son Nguyen 2026-08-04 16:49:43 +02:00
  • 7e611e89e2 nits Xuan Son Nguyen 2026-08-04 16:28:43 +02:00
  • 6341756e28 security : clarify about AI-generated reports Xuan Son Nguyen 2026-08-04 16:26:20 +02:00
  • 1c3c9674de models : fix dflash wo_a reshape on load (#26577) b10269 Georgi Gerganov 2026-08-04 16:56:49 +03:00
  • 56fe6a8187 models : fix dflash wo_a reshape on load gg/dsv4-fix-dflash-load Georgi Gerganov 2026-08-04 16:41:42 +03:00
  • 201e50cc20 clean up comments xsn/qwen3-tts Xuan Son Nguyen 2026-08-04 15:18:19 +02:00
  • 6b5224cfcc ci: fix pre-built binaries no longer working on macOS 15 and below (#26375) b10268 Niklas Wenzel 2026-08-04 15:03:38 +02:00
  • 7bd8282c37 speculative : refactor enabled configs common_speculative_init (#26510) b10267 Daniel Bevenius 2026-08-04 13:17:15 +02:00
  • 5788b510a1 gguf-py: validate n_dims and guard against uint64 overflow in reader (#25401) hcl 2026-08-04 17:12:48 +08:00
  • 2e17f69ef4 sync : ggml b10265 Georgi Gerganov 2026-08-04 11:54:13 +03:00
  • 15831f579a ggml : bump version to 0.18.1 (ggml/1578) Georgi Gerganov 2026-08-04 11:44:52 +03:00
  • b5746d28ce convert : add missing return after setting tekken vocab (#25947) Angel Galindo 2026-08-04 01:41:18 -07:00
  • f26efa02a7 vulkan backend ops: implemented GATED_LINEAR_ATTN (#25601) b10262 Pranav Uttarkar 2026-08-04 03:40:54 -05:00
  • cf06ad7dfe vocab : validate plamo2 byte tokens (#26511) b10261 Sigbjørn Skjæret 2026-08-04 10:40:02 +02:00
  • b06fbc968b convert : import bytes_to_unicode from convert_slow_tokenizer (#26217) Caleb DeLeeuw 2026-08-04 00:34:30 -07:00
  • 1269cb1ff1 model : allow reshape of tensors during load (#26531) b10259 Georgi Gerganov 2026-08-04 09:06:44 +03:00
  • eb75c420f0 Shared default of 64 for history-based samplers, remove context_size osimons/penalty_sampler_folllow_up_no2 Oliver Simons 2026-08-03 17:45:49 +02:00
  • 748eca633a Resolve -1 to 1024 instead of ctx-len for samplers Oliver Simons 2026-08-03 16:13:07 +02:00