{"full_name":"ggml-org/whisper.cpp","owner":"ggml-org","name":"whisper.cpp","url":"https://github.com/ggml-org/whisper.cpp","description":"Port of OpenAI's Whisper model in C/C++","homepage":"","topics":[],"language":"C++","license":"MIT","avatar":"https://avatars.githubusercontent.com/u/134263123?v=4","created_at":"2022-09-25T18:26:37Z","pushed_at":"2026-11-10T08:39:36Z","metrics":{"stars":55058,"forks":6295,"watchers":398,"issues":2004,"prs":1554,"releases":45,"issues_open":null,"issues_closed":null,"prs_open":null,"prs_closed":null,"prs_merged":null,"discussions":539},"growth":{"day":{"stars":23,"forks":7,"base_day":"2026-11-09"},"week":{"stars":190,"forks":29,"base_day":"2026-11-03"},"month":{"stars":906,"forks":91,"base_day":"2026-10-11"}},"history":{"dates":["2026-10-11","2026-10-12","2026-10-13","2026-10-14","2026-10-15","2026-10-16","2026-10-17","2026-10-18","2026-10-19","2026-10-20","2026-10-21","2026-10-22","2026-10-23","2026-10-24","2026-10-25","2026-10-26","2026-10-27","2026-10-28","2026-10-29","2026-10-30","2026-10-31","2026-11-01","2026-11-02","2026-11-03","2026-11-04","2026-11-09","2026-11-10"],"stars":[54152,54173,54200,54229,54249,54271,54317,54350,54385,54417,54439,54452,54490,54517,54542,54568,54597,54626,54687,54706,54736,54770,54803,54868,54887,55035,55058],"forks":[6204,6205,6206,6211,6215,6221,6235,6246,6249,6255,6262,6264,6270,6260,6229,6233,6237,6237,6241,6243,6249,6254,6257,6266,6270,6288,6295],"watchers":[413,412,409,409,409,409,409,409,409,409,409,409,409,406,399,399,400,395,395,395,399,400,399,399,399,398,398],"issues_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null],"prs_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null]},"boards":{},"same_owner":[{"full_name":"ggml-org/llama.cpp","name":"llama.cpp","boards":{"total":86}},{"full_name":"ggml-org/ggml","name":"ggml","boards":{}}],"issues":[{"number":89,"title":"Benchmark results","comments":187,"created_at":"2022-10-25T17:05:10Z"},{"number":64,"title":"whisper : mark speakers/voices (diarization)","comments":49,"created_at":"2022-10-18T09:33:00Z"},{"number":1724,"title":"Hallucination on silence","comments":29,"created_at":"2024-01-04T15:18:48Z"},{"number":2454,"title":"tests : add WER benchmarks","comments":26,"created_at":"2024-10-05T09:35:23Z"},{"number":1502,"title":"CUDA an illegal memory access was encountered","comments":22,"created_at":"2023-11-16T21:16:48Z"},{"number":1557,"title":"NPU support in whisper.cpp","comments":19,"created_at":"2023-11-27T00:09:35Z"},{"number":3708,"title":"OpenCL for Adreno support","comments":15,"created_at":"2026-03-15T11:12:38Z"},{"number":1507,"title":"Whisper large v3 model repeats a lot","comments":15,"created_at":"2023-11-17T06:49:54Z"}],"discussions":[{"number":4153,"title":"hark: macOS CLI that captures any app's audio and transcribes it with whisper.cpp","comments":0,"category":"Show and tell","created_at":"2026-11-05T01:55:50Z"},{"number":4145,"title":"interviewr-ai: a Linux interview and meeting copilot using whisper.cpp on Vulkan for live transcription","comments":0,"category":"Show and tell","created_at":"2026-10-31T13:43:46Z"},{"number":4131,"title":"Show whisper.cpp: Integrating offline Whisper transcription into an infinite 2D canvas AI workspace (Knovum, Tauri 2 + Rust)","comments":1,"category":"Show and tell","created_at":"2026-10-24T12:44:18Z"},{"number":4118,"title":"A tiny offline macOS menu-bar app for using whisper.cpp","comments":0,"category":"Show and tell","created_at":"2026-10-19T07:41:18Z"},{"number":4112,"title":"voicememotext — free web app for voice memo transcription (Whisper-powered)","comments":0,"category":"General","created_at":"2026-10-15T09:44:24Z"},{"number":4110,"title":"Utter – Native macOS on-device voice dictation app powered by whisper.cpp and Metal","comments":0,"category":"Show and tell","created_at":"2026-10-13T06:42:38Z"},{"number":4104,"title":"How to add Whisper headers and source to new Visual Studio project?","comments":2,"category":"Q&A","created_at":"2026-10-09T02:53:29Z"},{"number":4084,"title":"Add other models?","comments":1,"category":"Q&A","created_at":"2026-09-25T13:51:09Z"}],"latest_release":{"name":"v1.9.5","tag":"v1.9.5","published_at":"2026-10-06T17:39:17Z","body":"## Overview\r\n\r\nMaintenance release. Updates to [ggml v0.26.0](https://github.com/ggml-org/ggml/releases/tag/v0.26.0).\r\n\r\n**Nightly build:** [b5454](https://github.com/ggml-org/whisper.cpp/releases/tag/b5454)\r\n**More info:** [dist : releases and versioning of ggml-org projects](https://github.com/ggml-org/ggml/discussions/1579)\r\n\r\n## Changelog since v1.9.4\r\n\r\nd1be6fde whisper : bump version to 1.9.5 (#4102)\r\n4afec37b talk-llama : update llama.cpp to v0.6.0\r\n3d451abe sync : ggml\r\n47f1c7dc ggml : bump version to 0.26.0 (ggml/1652)\r\ne71b7844 CUDA: make the alloc_deps check batch independent (llama/29986)\r\ncdc87302 vulkan: fix Flash Attention shmem write out of bounds (llama/29988)\r\n563a9f7a vulkan: revert mul_mat_id tile selection PR #29182 (llama/29936)\r\nebb8a170 cuda: use the vector lightning indexer kernel on MUSA (llama/29990)\r\n8c6ca4ab CUDA: Optimize accumulation in mmq for NVFP4 type (llama/29857)\r\n82fa9df7 vulkan: fix stale prealloc_y reuse across flash attention and soft_max (llama/29591)\r\nf6915422 vulkan: sparse flash attention for quantized K/V (llama/29639)\r\nd477de3c llama : fix unexpected graph reallocation in the k-pool models (llama/29958)\r\n95283ca7 ggml-cpu : add Q8_0 IME1 matrix kernel for SpacemiT X60 (llama/28479)\r\n155c5c28 vulkan : Fix undeclared identifiers when -DGGML_VULKAN_RUN_TESTS=ON (llama/29912)\r\n7091d4a2 webgpu: add MMVQ support for Q1_0/Q5_0/Q5_1/Q3_K/Q5_K/Q6_K/MXFP4 (llama/29483)\r\n4cc62b81 cuda: tile the lightning indexer over keys and tokens for 4 heads (llama/29901)\r\n7e576cbd metal : few-row MMA mat-mul (llama/29869)\r\n91da4702 CUDA: use MMVF for thin f16/bf16 mul_mat at small batch size (llama/29633)\r\n8bb85d96 CUDA: prefer whole-tile FlashAttention scheduling for efficient two-stage kernels (llama/29435)\r\nce6552ef CUDA: refactor swizzling code (llama/29612)\r\nbfe3dc63 ggml-cpu: support BF16/FP16/FP32 K tails in tinyBLAS on x86 (llama/29806)\r\n25453c0a cuda : move neu_padded to where it is used (llama/29940)\r\n7914776a cuda : move blocks_per_col to where it is used (llama/29939)\r\n6421d2c1 CUDA: fix MMQ memory fault if n_expert >> n_ubatch (llama/29941)\r\n05d4d89a vulkan: fix rdna4 mat_vec tuning (llama/29934)\r\n51bfbd61 webgpu: add f16 support to fill/set_rows (llama/29897)\r\nf2aa80ad ggml-openvino: update to 2026.4.1, optimize performance, expand ops, improve device listing. (llama/29852)\r\n7f4a67a5 qwen4exp : halve the indexer score memory (llama/29825)\r\ncf5d9e30 CUDA: fuse shared experts into MMVQ (llama/29184)\r\nd7682b58 ggml-cuda : fix cpy transposed path corrupting non-contiguous dst (llama/27663)\r\n6b704713 ggml-quants : avoid invalid rounding in qkx3 scale search (llama/29817)\r\n0b35d1f3 ggml-cpu : fix soft_max_back wrong output when dst aliases src1 (llama/27096)\r\nd0dc4363 metal : add tensor API flash attention kernel for F16 KV (llama/29570)\r\ncca8f72b opencl: use sigmoid f16 for bf16 (llama/29787)\r\ncd1bee52 SYCL: Q8_0 DMMV ESIMD and MMVQ wide load (llama/29186)\r\n95a05ed4 vulkan: disable large matmul tile on Samsung GPUs with 32KB shared memory (llama/28531)\r\nc4051a33 sycl: large register file for D=512 FA vec kernels (llama/29062)\r\nae92605f sycl : do not use slow oneDNN reference matmul and fattn (llama/28985)\r\n9c968f7e qwen4exp : optimize mask constructions (llama/29824)\r\n4435763c ggml : add `alloc_buffer_n` to buffer type interface (llama/23671)\r\n8291ab84 vulkan: add logging to pipeline compile issues (llama/29794)\r\neaadab3d hexagon: install rebuilt HTP skels (llama/29828)\r\n0295ef68 hexagon: add q2_k and q3_k quant type support (llama/29717)\r\n396f68d3 CUDA: fix 2 broken Volta FA cases (llama/29803)\r\n3a3598cc llama: refer to segment documentation [no ci] (llama/29074)\r\n15229ac9 hexagon: shared strided DMA copy for CPY and CONCAT, any-dim CONCAT via DMA (llama/29685)\r\naa538005 cuda : route sm70 to the Turing MMVQ nwarps table (llama/29753)\r\n81ca4f81 metal : release temporary private transfer buffers (llama/29777)\r\n88948bdb webgpu: add bfloat16 support for MUL_MAT/MUL_MAT_ID/GET_ROWS- #29358 (llama/29"},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-10T09:35:25Z","history_start":"2026-09-06"}