{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "a-short-history-of-llama-cpp",
 "title": "A short history of llama.cpp",
 "dek": "The free program that put a large language model on a laptop: one evening's hack in March 2023, the library and the company under it, the GGUF file format it gave the world, how it reached graphics cards and grew a server, each step dated, and its team's move to Hugging Face in 2026.",
 "published": "2026-11-14",
 "series": [
  "notes"
 ],
 "licence": "CC BY 4.0",
 "status": "pending-judge",
 "pack_note": "no judge is seated, so every chip is held",
 "notice": "items with published:false are held and are not this page's published words",
 "source": {
  "url": "https://research.strata2signal.com/a-short-history-of-llama-cpp/",
  "md_url": "https://research.strata2signal.com/a-short-history-of-llama-cpp/index.md",
  "md_sha": "4939af6e1f11c5b63fe47d1df12641a33854a9f49a1ae05467eff4deabf1c2a1",
  "html_sha": "a2394a2fd536d6e619ab61bb1277274951e24a5316a0d58ed79ee3f4c32afc10",
  "anchors_sha": "8fdf5a9e7b595e7cc8282619369ff6fbc71cca3dceabb7d853a6550d08f85992"
 },
 "short": {
  "paragraph": "llama.cpp is free, open-source software that runs a large language model, the kind of AI inside a chatbot, on an ordinary computer: a laptop's processor, a graphics card, a phone. Georgi Gerganov wrote the first version in what his README called an evening and published it on 2023-03-10, two weeks after Meta released its LLaMA models to researchers. It is the engine under Ollama, LM Studio and the Strata engine, the project that wrote GGUF, the file format most of Ollama's library ships in, and the program beneath most of the language-model numbers on this hub. (Despite the similar name, the Strata engine, github.com/Niko1221/Strata, is an unrelated project; strata2signal has no affiliation with it.) Its company, ggml.ai, joined Hugging Face on 2026-02-20, and its first major version, v1.0.0, was published on 2026-11-13. It sits beside our other short histories, among them Mistral, Ollama and vLLM.",
  "from_dek": false,
  "counts": {
   "words": 3461,
   "minutes": 16,
   "tables": 1,
   "kit": false
  },
  "bullets": []
 },
 "sections": [
  {
   "id": "the-engine-under-most-of-our-numbers",
   "heading": "The engine under most of our numbers",
   "level": 2,
   "span": [
    2937,
    3680
   ],
   "chunks": [
    [
     2937,
     3680
    ]
   ],
   "chars": 743,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "one-evening-in-march-2023",
   "heading": "One evening in March 2023",
   "level": 2,
   "span": [
    3680,
    5146
   ],
   "chunks": [
    [
     3680,
     5146
    ]
   ],
   "chars": 1466,
   "digest": "on 2023-02-24, Meta announced LLaMA in four sizes from 7 to 65 billion parameters. on 2023-03-10, Georgi Gerganov made a commit titled \"Initial release\" and put it on GitHub as `llama.cpp`. that same evening, it was on Hacker News, drawing 989 points and 284 comments.",
   "digest_skipped": null
  },
  {
   "id": "ggml-and-ggml-ai",
   "heading": "ggml, the library underneath, and ggml.ai, the company",
   "level": 2,
   "span": [
    5146,
    6757
   ],
   "chunks": [
    [
     5146,
     6757
    ]
   ],
   "chars": 1611,
   "digest": "llama.cpp is built on top of the ggml library. ggml.ai is a company founded in 2023 by Georgi Gerganov to support the development of ggml. the ggml team continues to lead, maintain and support the libraries, and the company was acquired by Hugging Face in 2026.",
   "digest_skipped": null
  },
  {
   "id": "from-ggml-files-to-gguf",
   "heading": "From GGML files to GGUF",
   "level": 2,
   "span": [
    6757,
    7612
   ],
   "chunks": [
    [
     6757,
     7612
    ]
   ],
   "chars": 855,
   "digest": "the first format was a `ggml-model-f16.bin` or a `q4_0` copy. on 2023-06-05, the project merged k-quants, a series of 2-6 bit quantization methods. on 2023-07-26, Georgi Gerganov opened the GGUF pull request, which merged on 2023-08-21.",
   "digest_skipped": null
  },
  {
   "id": "the-backends-dated",
   "heading": "The backends, dated",
   "level": 2,
   "span": [
    7612,
    8937
   ],
   "chunks": [
    [
     7612,
     8937
    ]
   ],
   "chars": 1325,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "the-server",
   "heading": "The server",
   "level": 2,
   "span": [
    8937,
    10586
   ],
   "chunks": [
    [
     8937,
     10586
    ]
   ],
   "chars": 1649,
   "digest": "a contributor added the \"Server example with API Rest\" on 2023-05-21. today's `llama-server` README lists OpenAI-compatible routes, an Anthropic Messages route, reranking, parallel decoding, multimodal input, and a web UI. the v0.8.0 release of 2026-10-28 wires realtime audio in and out.",
   "digest_skipped": null
  },
  {
   "id": "what-was-built-on-it",
   "heading": "What was built on it",
   "level": 2,
   "span": [
    10586,
    11750
   ],
   "chunks": [
    [
     10586,
     11750
    ]
   ],
   "chars": 1164,
   "digest": "Ollama's first README began \"Run large language models with `llama.cpp`\" on 2023-07-08. LM Studio is powered by the LM Studio runtime, with MLX and llama.cpp under the hood. The Strata engine is built with parts of llama.cpp / ggml (MIT).",
   "digest_skipped": null
  },
  {
   "id": "how-it-is-numbered",
   "heading": "How it is numbered",
   "level": 2,
   "span": [
    11750,
    13614
   ],
   "chunks": [
    [
     11750,
     13614
    ]
   ],
   "chars": 1864,
   "digest": "a llama.cpp build is named by a count of commits on master. the project added semantic versioning in August 2026. as of 2026-11-13, the newest stable release was v1.0.0, published that day and tagged on the same commit as build b12359.",
   "digest_skipped": null
  },
  {
   "id": "where-it-stands-today",
   "heading": "Where it stands today",
   "level": 2,
   "span": [
    13614,
    14277
   ],
   "chunks": [
    [
     13614,
     14277
    ]
   ],
   "chars": 663,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "what-llama-cpp-did-on-our-cards",
   "heading": "What llama.cpp did on our cards",
   "level": 2,
   "span": [
    14277,
    16003
   ],
   "chunks": [
    [
     14277,
     16003
    ]
   ],
   "chars": 1726,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "what-to-take-with-you",
   "heading": "What to take with you",
   "level": 2,
   "span": [
    16003,
    16890
   ],
   "chunks": [
    [
     16003,
     16890
    ]
   ],
   "chars": 887,
   "digest": "local AI was made ordinary by a C/C++ program, 4-bit weights and a laptop. GGUF has been llama.cpp's format since 2023-08-21. a build number is a commit count, and the ggml team has been at Hugging Face since 2026-02-20.",
   "digest_skipped": null
  },
  {
   "id": "how-to-check-our-work-and-see-it-live",
   "heading": "How to check our work — and see it live",
   "level": 2,
   "span": [
    16890,
    17858
   ],
   "chunks": [
    [
     16890,
     17858
    ]
   ],
   "chars": 968,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "the-rest-of-the-seminar",
   "heading": "The rest of the seminar",
   "level": 2,
   "span": [
    17858,
    19608
   ],
   "chunks": [
    [
     17858,
     19608
    ]
   ],
   "chars": 1750,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "who-ran-this-and-thanks",
   "heading": "Who ran this, and thanks",
   "level": 2,
   "span": [
    19608,
    20539
   ],
   "chunks": [
    [
     19608,
     20539
    ]
   ],
   "chars": 931,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "sources",
   "heading": "Sources",
   "level": 2,
   "span": [
    20539,
    28744
   ],
   "chunks": [
    [
     20539,
     28744
    ]
   ],
   "chars": 8205,
   "digest": null,
   "digest_skipped": null
  }
 ],
 "chips": [
  {
   "id": "c-4e3983f1",
   "q": "how does llama.cpp relate to other local ai software?",
   "a": "llama.cpp serves as the underlying engine for several programs, including Ollama, LM Studio, and the Strata engine. It is also used by Llama Windows.",
   "cites": [
    "the-engine-under-most-of-our-numbers",
    "what-was-built-on-it"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-a7007a93",
   "q": "when did the initial release of llama.cpp occur?",
   "a": "The initial release of llama.cpp was made by Georgi Gerganov on 2023-03-10.",
   "cites": [
    "one-evening-in-march-2023"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-b821bb29",
   "q": "what role does the ggml library play in llama.cpp?",
   "a": "llama.cpp is built on top of the ggml library, which is a tensor library for machine learning used for its arithmetic.",
   "cites": [
    "ggml-and-ggml-ai"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-bb824307",
   "q": "how did the gguf file format change the project?",
   "a": "The GGUF format, merged on 2023-08-21, introduced a file that describes itself with key-value metadata beside the weights, representing a breaking change that required existing models to be converted again.",
   "cites": [
    "from-ggml-files-to-gguf"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-b0767725",
   "q": "what features are available in the llama-server?",
   "a": "The server includes OpenAI-compatible chat, responses, and embeddings routes, an Anthropic Messages route, reranking, parallel decoding with multi-user support, multimodal input, and a web UI.",
   "cites": [
    "the-server"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-d46d0e95",
   "q": "how are llama.cpp build numbers determined?",
   "a": "A build number is determined by a count of the number of commits on the master branch using a git command.",
   "cites": [
    "how-it-is-numbered"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-a610a32e",
   "q": "what is the current licensing for llama.cpp?",
   "a": "The project is licensed under the MIT licence.",
   "cites": [
    "where-it-stands-today"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "the-llama-weights-that-got-out",
   "why": "shares ground with § One evening in March 2023 · § February 2023: a model for researchers, by application"
  },
  {
   "slug": "a-short-history-of-ollama",
   "why": "shares ground with § What was built on it · § The engine underneath: llama.cpp"
  },
  {
   "slug": "inside-a-gguf-file",
   "why": "shares ground with § From GGML files to GGUF · § Where it came from"
  }
 ],
 "thanks": "## Who ran this, and thanks {#who-ran-this-and-thanks}\n\nThanks to **Georgi Gerganov** and **the ggml authors** for **llama.cpp** and **ggml** (both MIT), the engine under most of this shelf's language-model numbers, and for a repository whose first evening is still readable in its history; to the contributors named in the backends table, whose pull requests are the record; to **Hugging Face** for its GGUF documentation and the `ggml-org` blog; to **Meta** for the LLaMA announcement this story starts from; to **Ollama** (MIT), **LM Studio** and **the Strata engine**'s author (MIT) for saying what runs under their programs; and to **GitHub** and **Hacker News** for the dated record. None of them owed us anything. A small human team asked for this page, chose what it would and would not claim, and signed off on it; a fleet of AI agents fetched and read every source listed below and drafted it under that team's rulings.",
 "kit": null,
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 0,
  "digests_written": 12,
  "digests_kept": 7,
  "chips_written": 8,
  "chips_kept": 7,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 0,
   "figure": 0,
   "length": 0,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 0,
   "profanity": 0,
   "identifier": 1,
   "bare_figure": 0,
   "house_voice": 0,
   "editorial": 0,
   "hand-edit": 8
  },
  "new_noun_tokens_checked": [
   "108",
   "2",
   "2-6",
   "2023",
   "2023-02-24",
   "2023-03-10",
   "2023-05-21",
   "2023-06-05",
   "2023-07-08",
   "2023-07-26",
   "2023-08-21",
   "2026",
   "2026-02-20",
   "2026-03-24",
   "2026-10-28",
   "2026-11-12",
   "2026-11-13",
   "22.657",
   "24",
   "24.4",
   "242",
   "284",
   "34",
   "4-bit",
   "5090",
   "65",
   "7",
   "989",
   "AI",
   "API",
   "Anthropic",
   "August",
   "C/C",
   "CLBlast",
   "CUDA",
   "Face",
   "GB",
   "GGUF",
   "GPU",
   "GeForce",
   "Georgi",
   "Gerganov",
   "GitHub",
   "HIP",
   "Hacker",
   "Hugging",
   "Initial",
   "LLaMA",
   "LM",
   "Laptop",
   "Llama",
   "MIT",
   "MLX",
   "Messages",
   "Meta",
   "Metal",
   "NVIDIA",
   "News",
   "Ollama",
   "OpenAI-compatible",
   "OpenVINO",
   "README",
   "README's",
   "RTX",
   "Rest",
   "RuleSage's",
   "Run",
   "SYCL",
   "Server",
   "Strata",
   "Studio",
   "UI",
   "Vulkan",
   "WebGPU",
   "Windows",
   "b11514",
   "b12359",
   "ggml-model-f16.bin",
   "q40",
   "v0.8.0",
   "v1.0.0"
  ],
  "new_noun_tokens_withheld": 0,
  "figure_definition": "v3",
  "figure_boundary": "guarded",
  "drops": [
   {
    "kind": "bullet",
    "reason": "identifier",
    "item": "the ggml-org organization on Hugging Face held 242",
    "detail": "ggml-org is in the cited section only as code or in a table cell"
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 53,
    "text_sha": "81adb0dda564",
    "figure": "2026-11-13",
    "withheld": "one bullet removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "As of 2026-11-13 (UTC), the day we last read, the newest stable release was v1.0.0, published that day and tagged on the same commit as build b12359.",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: As of 2026-11-13 (UTC), the day we last read, the newest stable release was v1.0.0, published that day and tagged on the same commit as build b12359. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 41,
    "text_sha": "565daa2a112d",
    "figure": "7",
    "withheld": "one bullet removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "On 2023-02-24 Meta announced LLaMA, \"a state-of-the-art foundational large language model\", in four sizes from 7 to 65 billion parameters",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: On 2023-02-24 Meta announced LLaMA, \"a state-of-the-art foundational large language model\", in four sizes from 7 to 65 billion parameters. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 273,
    "text_sha": "8d66bb4cc583",
    "section": "the-engine-under-most-of-our-numbers",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "Our two pages on the Strata engine benched llama.cpp itself on three cards.",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: Our two pages on the Strata engine benched llama.cpp itself on three cards. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 248,
    "text_sha": "131dd2022bbd",
    "section": "the-backends-dated",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "The README's \"Supported backends\" table at master, read 2026-11-12, has seventeen rows, among them CANN, MUSA, Hexagon, WebGPU, an RPC backend and OpenVINO (marked in progress).",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: The README's \"Supported backends\" table at master, read 2026-11-12, has seventeen rows, among them CANN, MUSA, Hexagon, WebGPU, an RPC backend and OpenVINO (marked in progress). Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 221,
    "text_sha": "d205bd6f711c",
    "section": "where-it-stands-today",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "an update to the same post restored the maintainers' write access and raised the approvals each pull request needs from one to two.",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: an update to the same post restored the maintainers' write access and raised the approvals each pull request needs from one to two. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 212,
    "text_sha": "46c5632f0e44",
    "section": "what-llama-cpp-did-on-our-cards",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "24.4, 102.889 and 22.490 tokens a second writing the answer to a 4,038-token prompt, under the conditions that page states.",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: 24.4, 102.889 and 22.490 tokens a second writing the answer to a 4,038-token prompt, under the conditions that page states. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "digest",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 240,
    "text_sha": "efe0b17351c8",
    "section": "sources",
    "withheld": "one section digest removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "a date printed on the page credited, or a sentence quoted from it",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: a date printed on the page credited, or a sentence quoted from it. Not a gate drop, and recorded here so the ledger adds up."
   },
   {
    "kind": "chip",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 193,
    "text_sha": "65556e06decf",
    "chip_id": "c-534f0226",
    "withheld": "one chip removed from this pack by hand",
    "removed_utc": "2026-11-13",
    "why": "The README's \"Supported backends\" table at master, read 2026-11-12, has seventeen rows, among them CANN, MUSA, Hexagon, WebGPU, an RPC backend and OpenVINO (marked in progress).",
    "detail": "removed from this pack by hand on 2026-11-13 (UTC), after the writer ran and after the gates had counted: The README's \"Supported backends\" table at master, read 2026-11-12, has seventeen rows, among them CANN, MUSA, Hexagon, WebGPU, an RPC backend and OpenVINO (marked in progress). Not a gate drop, and recorded here so the ledger adds up."
   }
  ]
 },
 "generated_utc": "2026-11-13T17:41:24Z"
}
