@averyyy/pi-coding-agent 0.87.1-piclient.2 → 1.0.0-piclient.1

package/dist/extensions/llama/huggingface.js.map

@@ -1 +1 @@
-{"version":3,"file":"huggingface.js","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,MAAM,wBAAwB,GAAG,wBAAwB,CAAC;AAC1D,MAAM,oBAAoB,GACzB,uGAAuG,CAAC;AACzG,MAAM,oBAAoB,GAAG,mBAAmB,CAAC;AAkBjD,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB,EAAU;IACjE,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC7D;AAED,SAAS,mBAAmB,CAAC,KAAoB,EAAsB;IACtE,MAAM,KAAK,GAAG,KAAK,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC5C;AAED,KAAK,UAAU,SAAS,CAAC,IAAY,EAA+B;IACnE,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACpD,OAAO,KAAK,IAAI,SAAS,CAAC;IAC3B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AAAA,CACD;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,GAAG,GAAsB,OAAO,CAAC,GAAG,EAA+B;IAC7G,MAAM,eAAe,GAAG,GAAG,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC7C,IAAI,eAAe;QAAE,OAAO,eAAe,CAAC;IAE5C,MAAM,KAAK,GAAG;QACb,GAAG,CAAC,aAAa;QACjB,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACpD,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACjF,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC;KACjD,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;IAClD,KAAK,MAAM,IAAI,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,CAAC;QACpC,IAAI,KAAK;YAAE,OAAO,KAAK,CAAC;IACzB,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,MAAM,OAAO,iBAAiB;IACZ,KAAK,CAAqB;IAC1B,OAAO,CAAS;IAEjC,YAAY,KAAc,EAAE,OAAO,GAAG,wBAAwB,EAAE;QAC/D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,MAAoB,EAAoB;QAC3E,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,KAAK;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;QACrE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE;YACtD,OAAO;YACP,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO;SAC7D,CAAC,CAAC;QACH,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YAClB,MAAM,QAAQ,GAAG,8BAA8B,QAAQ,CAAC,MAAM,EAAE,CAAC;YACjE,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBAC7B,MAAM,KAAK,GACV,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,IAAI,mBAAmB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;gBACvG,MAAM,IAAI,KAAK,CACd,KAAK,CAAC,CAAC,CAAC,6CAA6C,KAAK,GAAG,CAAC,CAAC,CAAC,iCAAiC,CACjG,CAAC;YACH,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC;QAClD,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB,EAA+B;QAC9E,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC;YAClC,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,WAAW;YACjB,SAAS,EAAE,IAAI;YACf,KAAK,EAAE,IAAI;SACX,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC;QACpE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAC7F,OAAO,OAAO,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YACjC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,OAAQ,KAA0B,CAAC,EAAE,KAAK,QAAQ;gBACpG,OAAO,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,KAA4C,CAAC;YAC3D,OAAO,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAAA,CAChG,CAAC,CAAC;IAAA,CACH;IAED,KAAK,CAAC,OAAO,CAAC,EAAU,EAAE,MAAoB,EAAoC;QACjF,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC;QAClF,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;YACrD,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,KAAK,GAAG,OAAgE,CAAC;QAC/E,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgD,CAAC;QACtE,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACnC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;gBACpC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,KAAgD,CAAC;gBAC9D,IAAI,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;oBAAE,SAAS;gBACpG,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAE,CAAC;gBACnD,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;gBACrE,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;gBAC1E,IAAI,CAAC,YAAY;oBAAE,SAAS;gBAC5B,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;gBACxE,IAAI,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;oBAAE,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC;;oBACzD,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;gBAC9B,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;YAClC,CAAC;QACF,CAAC;QACD,MAAM,aAAa,GAAG,CAAC,GAAG,KAAK,CAAC;aAC9B,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;aAC/E,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACtB,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC,CAAC;YACtC,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC;YACtC,OAAO,CACN,CAAC,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC;gBAChF,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CACnC,CAAC;QAAA,CACF,CAAC,CAAC;QACJ,OAAO;YACN,EAAE,EAAE,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;YAChD,KAAK,EAAE,KAAK,CAAC,KAAK,KAAK,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK;YAC/E,aAAa;SACb,CAAC;IAAA,CACF;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}
+{"version":3,"file":"huggingface.js","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,MAAM,wBAAwB,GAAG,wBAAwB,CAAC;AAC1D,MAAM,oBAAoB,GACzB,uGAAuG,CAAC;AACzG,MAAM,oBAAoB,GAAG,mBAAmB,CAAC;AAkBjD,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB;IACvD,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAC9D,CAAC;AAED,SAAS,mBAAmB,CAAC,KAAoB;IAChD,MAAM,KAAK,GAAG,KAAK,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC7C,CAAC;AAED,KAAK,UAAU,SAAS,CAAC,IAAY;IACpC,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACpD,OAAO,KAAK,IAAI,SAAS,CAAC;IAC3B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AACF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,GAAG,GAAsB,OAAO,CAAC,GAAG;IAC9E,MAAM,eAAe,GAAG,GAAG,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC7C,IAAI,eAAe;QAAE,OAAO,eAAe,CAAC;IAE5C,MAAM,KAAK,GAAG;QACb,GAAG,CAAC,aAAa;QACjB,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACpD,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACjF,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC;KACjD,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;IAClD,KAAK,MAAM,IAAI,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,CAAC;QACpC,IAAI,KAAK;YAAE,OAAO,KAAK,CAAC;IACzB,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED,MAAM,OAAO,iBAAiB;IACZ,KAAK,CAAqB;IAC1B,OAAO,CAAS;IAEjC,YAAY,KAAc,EAAE,OAAO,GAAG,wBAAwB;QAC7D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAC7C,CAAC;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,MAAoB;QACvD,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,KAAK;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;QACrE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE;YACtD,OAAO;YACP,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO;SAC7D,CAAC,CAAC;QACH,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YAClB,MAAM,QAAQ,GAAG,8BAA8B,QAAQ,CAAC,MAAM,EAAE,CAAC;YACjE,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBAC7B,MAAM,KAAK,GACV,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,IAAI,mBAAmB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;gBACvG,MAAM,IAAI,KAAK,CACd,KAAK,CAAC,CAAC,CAAC,6CAA6C,KAAK,GAAG,CAAC,CAAC,CAAC,iCAAiC,CACjG,CAAC;YACH,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC;QAClD,CAAC;QACD,OAAO,OAAO,CAAC;IAChB,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB;QAC/C,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC;YAClC,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,WAAW;YACjB,SAAS,EAAE,IAAI;YACf,KAAK,EAAE,IAAI;SACX,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC;QACpE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAC7F,OAAO,OAAO,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE;YAChC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,OAAQ,KAA0B,CAAC,EAAE,KAAK,QAAQ;gBACpG,OAAO,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,KAA4C,CAAC;YAC3D,OAAO,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACjG,CAAC,CAAC,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,EAAU,EAAE,MAAoB;QAC7C,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC;QAClF,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;YACrD,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,KAAK,GAAG,OAAgE,CAAC;QAC/E,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgD,CAAC;QACtE,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACnC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;gBACpC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,KAAgD,CAAC;gBAC9D,IAAI,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;oBAAE,SAAS;gBACpG,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAE,CAAC;gBACnD,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;gBACrE,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;gBAC1E,IAAI,CAAC,YAAY;oBAAE,SAAS;gBAC5B,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;gBACxE,IAAI,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;oBAAE,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC;;oBACzD,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;gBAC9B,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;YAClC,CAAC;QACF,CAAC;QACD,MAAM,aAAa,GAAG,CAAC,GAAG,KAAK,CAAC;aAC9B,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;aAC/E,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;YACrB,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC,CAAC;YACtC,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC;YACtC,OAAO,CACN,CAAC,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC;gBAChF,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CACnC,CAAC;QACH,CAAC,CAAC,CAAC;QACJ,OAAO;YACN,EAAE,EAAE,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;YAChD,KAAK,EAAE,KAAK,CAAC,KAAK,KAAK,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK;YAC/E,aAAa;SACb,CAAC;IACH,CAAC;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}

package/dist/extensions/llama/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAA2B,MAAM,gCAAgC,CAAC;AAyC5F,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CA4L7D","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\t// /llama already contacted the configured llama.cpp server, so keep this refresh live even in PI_OFFLINE.\n\t\t\tallowNetwork: true,\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAA2B,MAAM,gCAAgC,CAAC;AAyC5F,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CA4L7D"}

package/dist/extensions/llama/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,WAAW,EAAuB,uBAAuB,EAAE,MAAM,aAAa,CAAC;AACrG,OAAO,EAAE,oBAAoB,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC3E,OAAO,EAAE,mBAAmB,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACvE,OAAO,EAAgB,eAAe,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAErE,SAAS,aAAa,CAAC,KAAqB,EAAW;IACtD,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,CAAC;AAAA,CAC5E;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAW;IACnD,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,MAAM,OAAO,GAAG,GAAG,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,OAAO,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AAAA,CACtG;AAED,SAAS,sBAAsB,CAAC,KAAc,EAAU;IACvD,IAAI,iBAAiB,CAAC,KAAK,CAAC;QAAE,OAAO,kCAAkC,CAAC;IACxE,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAAA,CAC9D;AAED,SAAS,qBAAqB,CAAC,KAAa,EAAiD;IAC5F,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACzD,OAAO,KAAK,GAAG,CAAC;QACf,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE;QACvB,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;AAAA,CAC/E;AAED,KAAK,UAAU,gBAAgB,CAAC,GAA4B,EAAoC;IAC/F,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,eAAe,CAAC,iBAAiB,CAAC,CAAC;IAC1E,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,mCAAmC,iBAAiB,EAAE,EAAE,SAAS,CAAC,CAAC;QACjF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,EAAE,cAAc,CAAC;IACjD,MAAM,SAAS,GAAG,uBAAuB,CACxC,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAChG,CAAC;IACF,OAAO,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAAA,CACtD;AAED,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAgB,EAAQ;IAC9D,MAAM,QAAQ,GAAG,mBAAmB,EAAE,CAAC;IACvC,EAAE,CAAC,gBAAgB,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAEvC,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,MAAmB,EACnB,OAA0B,EACE,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC3C,MAAM,OAAO,GAAG,OAAO,IAAI,CAAC,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC;QAC3D,QAAQ,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,OAAO,CAAC;YAC9C,SAAS,EAAE,CAAC,iBAAiB,CAAC;YAC9B,0GAA0G;YAC1G,YAAY,EAAE,IAAI;YAClB,MAAM;SACN,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,OAAO;YAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;QACxE,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QAC1D,IAAI,YAAY;YAAE,MAAM,YAAY,CAAC;QACrC,OAAO,OAAO,CAAC;IAAA,CACf,CAAC;IAEF,MAAM,SAAS,GAAG,KAAK,EACtB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,OAAyB,EACzB,MAAsB,EACN,EAAE,CAAC;QACnB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,IAAI,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC;QACzF,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvB,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,SAAS,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,SAAS,EAAE;gBACvG,qBAAqB;gBACrB,sBAAsB;gBACtB,QAAQ;aACR,CAAC,CAAC;YACH,IAAI,CAAC,MAAM,IAAI,MAAM,KAAK,QAAQ;gBAAE,OAAO;YAC3C,OAAO,GAAG,MAAM,KAAK,qBAAqB,CAAC;QAC5C,CAAC;QAED,MAAM,aAAa,GAAG,KAAK,IAAmB,EAAE,CAAC;YAChD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,oCAAoC,CAAC,CAAC;YACpD,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;YACzE,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAAA,CAC/B,CAAC;QACF,IAAI,OAAO,EAAE,CAAC;YACb,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;gBACxC,KAAK,EAAE,eAAe;gBACtB,KAAK,EAAE,MAAM,CAAC,EAAE;gBAChB,cAAc,EAAE,aAAW;gBAC3B,WAAW,EAAE,eAAe;gBAC5B,aAAa,EAAE,MAAM,CAAC,EAAE;gBACxB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC;gBACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;aACtC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;gBACtB,IAAI,OAAO;oBAAE,MAAM,aAAa,EAAE,CAAC;gBACnC,OAAO;YACR,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACjD,MAAM,WAAW,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAC;YACtE,GAAG,CAAC,EAAE,CAAC,MAAM,CACZ,WAAW,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,oBAAoB,MAAM,CAAC,EAAE,EAAE,CAChG,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,EAAE,CAAC;gBACb,IAAI,CAAC;oBACJ,MAAM,aAAa,EAAE,CAAC;gBACvB,CAAC;gBAAC,MAAM,CAAC;oBACR,oCAAoC;gBACrC,CAAC;YACF,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,KAAqB,EACL,EAAE,CAAC;QACnB,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;YAAE,OAAO;QAC3D,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACrC,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC/B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,YAAY,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAAA,CACtC,CAAC;IAEF,MAAM,aAAa,GAAG,KAAK,EAAE,GAA4B,EAAE,EAAW,EAAE,MAAmB,EAAiB,EAAE,CAAC;QAC9G,MAAM,WAAW,GAAG,IAAI,iBAAiB,CAAC,MAAM,oBAAoB,EAAE,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,EAAE,CAAC,YAAY,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,WAAW,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QAC7F,IAAI,CAAC,QAAQ;YAAE,OAAO;QACtB,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QAC/C,EAAE,CAAC,UAAU,CAAC,uBAAuB,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;QAC1D,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAC7D,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,yBAAyB,CAAC;YACxG,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAC7B,iCAAiC,OAAO,CAAC,EAAE,OAAO,QAAQ,gCAAgC,OAAO,CAAC,EAAE,sDAAsD,EAC1J,CAAC,UAAU,EAAE,MAAM,CAAC,CACpB,CAAC;YACF,IAAI,MAAM,KAAK,UAAU;gBAAE,OAAO;QACnC,CAAC;QACD,IAAI,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACvC,IAAI,CAAC,YAAY,IAAI,OAAO,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,OAAO,GAAG,OAAO,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;gBACpD,MAAM,MAAM,GAAG;oBACd,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC;oBAC9D,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;iBACnD;qBACC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;qBAClD,IAAI,CAAC,MAAK,CAAC,CAAC;gBACd,OAAO,MAAM,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,OAAM,MAAM,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;YAAA,CACzD,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,wBAAwB,OAAO,CAAC,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,YAAY,GAAG,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC;YACpE,IAAI,CAAC,YAAY;gBAAE,OAAO;QAC3B,CAAC;QACD,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1E,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;YACxC,KAAK,EAAE,mBAAmB;YAC1B,KAAK;YACL,cAAc,EAAE,aAAW;YAC3B,WAAW,EAAE,gBAAgB;YAC7B,aAAa,EAAE,KAAK;YACpB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC;YACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC;SAClC,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,SAAS;YAAE,OAAO;QAC7B,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAC7C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,cAAc,KAAK,EAAE,CAAC,CAAC;IAAA,CACrC,CAAC;IAEF,EAAE,CAAC,eAAe,CAAC,OAAO,EAAE;QAC3B,WAAW,EAAE,gCAAgC;QAC7C,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC;YAC9B,IAAI,GAAG,CAAC,IAAI,KAAK,KAAK,EAAE,CAAC;gBACxB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,yCAAyC,EAAE,SAAS,CAAC,CAAC;gBACpE,OAAO;YACR,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,GAAG,CAAC,CAAC;YAC3C,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,MAAM,WAAW,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAAC;gBACpC,MAAM,WAAW,GAAG,KAAK,IAA2C,EAAE,CAAC;oBACtE,OAAO,IAAI,EAAE,CAAC;wBACb,IAAI,CAAC;4BACJ,OAAO,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;wBACvC,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BAChB,IAAI,CAAC,MAAM,EAAE,CAAC,eAAe,CAAC,MAAM,CAAC,SAAS,EAAE,sBAAsB,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,OAAO,EAAE,CAAC;gCAC7F,OAAO,SAAS,CAAC;4BAClB,CAAC;wBACF,CAAC;oBACF,CAAC;gBAAA,CACD,CAAC;gBAEF,IAAI,OAAO,GAAG,MAAM,WAAW,EAAE,CAAC;gBAClC,IAAI,CAAC,OAAO;oBAAE,OAAO;gBACrB,OAAO,IAAI,EAAE,CAAC;oBACb,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;oBAC9D,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO;wBAAE,OAAO;oBACpC,IAAI,WAAoB,CAAC;oBACzB,IAAI,CAAC;wBACJ,IAAI,MAAM,CAAC,IAAI,KAAK,UAAU;4BAAE,MAAM,aAAa,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;6BAChE,IAAI,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC;4BAAE,MAAM,WAAW,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;6BAClF,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;4BAChD,MAAM,SAAS,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;;4BACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,OAAO,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;oBACrF,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,WAAW,GAAG,KAAK,CAAC;oBACrB,CAAC;oBACD,MAAM,SAAS,GAAG,MAAM,WAAW,EAAE,CAAC;oBACtC,IAAI,CAAC,SAAS;wBAAE,OAAO;oBACvB,OAAO,GAAG,SAAS,CAAC;oBACpB,IAAI,WAAW,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,EAAE,CAAC;wBACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,YAAY,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,CAAC;oBAClG,CAAC;gBACF,CAAC;YAAA,CACD,CAAC,CAAC;QAAA,CACH;KACD,CAAC,CAAC;AAAA,CACH","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\t// /llama already contacted the configured llama.cpp server, so keep this refresh live even in PI_OFFLINE.\n\t\t\tallowNetwork: true,\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,WAAW,EAAuB,uBAAuB,EAAE,MAAM,aAAa,CAAC;AACrG,OAAO,EAAE,oBAAoB,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC3E,OAAO,EAAE,mBAAmB,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACvE,OAAO,EAAgB,eAAe,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAErE,SAAS,aAAa,CAAC,KAAqB;IAC3C,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,CAAC;AAC7E,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc;IACxC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,MAAM,OAAO,GAAG,GAAG,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,OAAO,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AACvG,CAAC;AAED,SAAS,sBAAsB,CAAC,KAAc;IAC7C,IAAI,iBAAiB,CAAC,KAAK,CAAC;QAAE,OAAO,kCAAkC,CAAC;IACxE,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAC/D,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAa;IAC3C,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACzD,OAAO,KAAK,GAAG,CAAC;QACf,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE;QACvB,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;AAChF,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,GAA4B;IAC3D,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,eAAe,CAAC,iBAAiB,CAAC,CAAC;IAC1E,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,mCAAmC,iBAAiB,EAAE,EAAE,SAAS,CAAC,CAAC;QACjF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,EAAE,cAAc,CAAC;IACjD,MAAM,SAAS,GAAG,uBAAuB,CACxC,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAChG,CAAC;IACF,OAAO,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AACvD,CAAC;AAED,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAgB;IACtD,MAAM,QAAQ,GAAG,mBAAmB,EAAE,CAAC;IACvC,EAAE,CAAC,gBAAgB,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAEvC,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,MAAmB,EACnB,OAA0B,EACE,EAAE;QAC9B,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC3C,MAAM,OAAO,GAAG,OAAO,IAAI,CAAC,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC;QAC3D,QAAQ,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,OAAO,CAAC;YAC9C,SAAS,EAAE,CAAC,iBAAiB,CAAC;YAC9B,0GAA0G;YAC1G,YAAY,EAAE,IAAI;YAClB,MAAM;SACN,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,OAAO;YAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;QACxE,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QAC1D,IAAI,YAAY;YAAE,MAAM,YAAY,CAAC;QACrC,OAAO,OAAO,CAAC;IAChB,CAAC,CAAC;IAEF,MAAM,SAAS,GAAG,KAAK,EACtB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,OAAyB,EACzB,MAAsB,EACN,EAAE;QAClB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,IAAI,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC;QACzF,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvB,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,SAAS,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,SAAS,EAAE;gBACvG,qBAAqB;gBACrB,sBAAsB;gBACtB,QAAQ;aACR,CAAC,CAAC;YACH,IAAI,CAAC,MAAM,IAAI,MAAM,KAAK,QAAQ;gBAAE,OAAO;YAC3C,OAAO,GAAG,MAAM,KAAK,qBAAqB,CAAC;QAC5C,CAAC;QAED,MAAM,aAAa,GAAG,KAAK,IAAmB,EAAE;YAC/C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,oCAAoC,CAAC,CAAC;YACpD,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,EAAE,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;YACzE,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAChC,CAAC,CAAC;QACF,IAAI,OAAO,EAAE,CAAC;YACb,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;gBACxC,KAAK,EAAE,eAAe;gBACtB,KAAK,EAAE,MAAM,CAAC,EAAE;gBAChB,cAAc,EAAE,WAAW;gBAC3B,WAAW,EAAE,eAAe;gBAC5B,aAAa,EAAE,MAAM,CAAC,EAAE;gBACxB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC;gBACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;aACtC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;gBACtB,IAAI,OAAO;oBAAE,MAAM,aAAa,EAAE,CAAC;gBACnC,OAAO;YACR,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACjD,MAAM,WAAW,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAC;YACtE,GAAG,CAAC,EAAE,CAAC,MAAM,CACZ,WAAW,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,oBAAoB,MAAM,CAAC,EAAE,EAAE,CAChG,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,EAAE,CAAC;gBACb,IAAI,CAAC;oBACJ,MAAM,aAAa,EAAE,CAAC;gBACvB,CAAC;gBAAC,MAAM,CAAC;oBACR,oCAAoC;gBACrC,CAAC;YACF,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;IACF,CAAC,CAAC;IAEF,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,KAAqB,EACL,EAAE;QAClB,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;YAAE,OAAO;QAC3D,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACrC,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC/B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,YAAY,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IACvC,CAAC,CAAC;IAEF,MAAM,aAAa,GAAG,KAAK,EAAE,GAA4B,EAAE,EAAW,EAAE,MAAmB,EAAiB,EAAE;QAC7G,MAAM,WAAW,GAAG,IAAI,iBAAiB,CAAC,MAAM,oBAAoB,EAAE,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,EAAE,CAAC,YAAY,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,WAAW,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QAC7F,IAAI,CAAC,QAAQ;YAAE,OAAO;QACtB,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QAC/C,EAAE,CAAC,UAAU,CAAC,uBAAuB,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;QAC1D,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAC7D,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,yBAAyB,CAAC;YACxG,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAC7B,iCAAiC,OAAO,CAAC,EAAE,OAAO,QAAQ,gCAAgC,OAAO,CAAC,EAAE,sDAAsD,EAC1J,CAAC,UAAU,EAAE,MAAM,CAAC,CACpB,CAAC;YACF,IAAI,MAAM,KAAK,UAAU;gBAAE,OAAO;QACnC,CAAC;QACD,IAAI,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACvC,IAAI,CAAC,YAAY,IAAI,OAAO,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,OAAO,GAAG,OAAO,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;gBACnD,MAAM,MAAM,GAAG;oBACd,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC;oBAC9D,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;iBACnD;qBACC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;qBAClD,IAAI,CAAC,KAAK,CAAC,CAAC;gBACd,OAAO,MAAM,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,MAAM,MAAM,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;YAC1D,CAAC,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,wBAAwB,OAAO,CAAC,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,YAAY,GAAG,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC;YACpE,IAAI,CAAC,YAAY;gBAAE,OAAO;QAC3B,CAAC;QACD,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1E,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;YACxC,KAAK,EAAE,mBAAmB;YAC1B,KAAK;YACL,cAAc,EAAE,WAAW;YAC3B,WAAW,EAAE,gBAAgB;YAC7B,aAAa,EAAE,KAAK;YACpB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC;YACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC;SAClC,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,SAAS;YAAE,OAAO;QAC7B,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAC7C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,cAAc,KAAK,EAAE,CAAC,CAAC;IACtC,CAAC,CAAC;IAEF,EAAE,CAAC,eAAe,CAAC,OAAO,EAAE;QAC3B,WAAW,EAAE,gCAAgC;QAC7C,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE;YAC7B,IAAI,GAAG,CAAC,IAAI,KAAK,KAAK,EAAE,CAAC;gBACxB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,yCAAyC,EAAE,SAAS,CAAC,CAAC;gBACpE,OAAO;YACR,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,GAAG,CAAC,CAAC;YAC3C,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,MAAM,WAAW,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE;gBACnC,MAAM,WAAW,GAAG,KAAK,IAA2C,EAAE;oBACrE,OAAO,IAAI,EAAE,CAAC;wBACb,IAAI,CAAC;4BACJ,OAAO,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;wBACvC,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BAChB,IAAI,CAAC,MAAM,EAAE,CAAC,eAAe,CAAC,MAAM,CAAC,SAAS,EAAE,sBAAsB,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,OAAO,EAAE,CAAC;gCAC7F,OAAO,SAAS,CAAC;4BAClB,CAAC;wBACF,CAAC;oBACF,CAAC;gBACF,CAAC,CAAC;gBAEF,IAAI,OAAO,GAAG,MAAM,WAAW,EAAE,CAAC;gBAClC,IAAI,CAAC,OAAO;oBAAE,OAAO;gBACrB,OAAO,IAAI,EAAE,CAAC;oBACb,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;oBAC9D,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO;wBAAE,OAAO;oBACpC,IAAI,WAAoB,CAAC;oBACzB,IAAI,CAAC;wBACJ,IAAI,MAAM,CAAC,IAAI,KAAK,UAAU;4BAAE,MAAM,aAAa,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;6BAChE,IAAI,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC;4BAAE,MAAM,WAAW,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;6BAClF,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;4BAChD,MAAM,SAAS,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;;4BACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,OAAO,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;oBACrF,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,WAAW,GAAG,KAAK,CAAC;oBACrB,CAAC;oBACD,MAAM,SAAS,GAAG,MAAM,WAAW,EAAE,CAAC;oBACtC,IAAI,CAAC,SAAS;wBAAE,OAAO;oBACvB,OAAO,GAAG,SAAS,CAAC;oBACpB,IAAI,WAAW,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,EAAE,CAAC;wBACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,YAAY,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,CAAC;oBAClG,CAAC;gBACF,CAAC;YACF,CAAC,CAAC,CAAC;QACJ,CAAC;KACD,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\t// /llama already contacted the configured llama.cpp server, so keep this refresh live even in PI_OFFLINE.\n\t\t\tallowNetwork: true,\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}

package/dist/extensions/llama/provider.d.ts

@@ -1,4 +1,4 @@
-import type { Provider } from "@earendil-works/pi-ai";
+import { type Provider } from "@earendil-works/pi-ai";
 import { type LlamaModelInfo } from "./client.ts";
 export declare const LLAMA_PROVIDER_ID = "llama.cpp";
 export declare const DEFAULT_LLAMA_SERVER_URL = "http://127.0.0.1:8080";

package/dist/extensions/llama/provider.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAKX,QAAQ,EAGR,MAAM,uBAAuB,CAAC;AAE/B,OAAO,EAEN,KAAK,cAAc,EAInB,MAAM,aAAa,CAAC;AAErB,eAAO,MAAM,iBAAiB,cAAc,CAAC;AAC7C,eAAO,MAAM,wBAAwB,0BAA0B,CAAC;AAiEhE,MAAM,WAAW,uBAAuB;IACvC,QAAQ,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACzC,UAAU,CAAC,MAAM,EAAE,SAAS,cAAc,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,cAAc,CAAC,EAAE,OAAO,CAAA;KAAE,GAAG,IAAI,CAAC;CAC/G;AAED,wBAAgB,mBAAmB,IAAI,uBAAuB,CA+G7D","sourcesContent":["import type {\n\tApiKeyCredential,\n\tAuthContext,\n\tAuthResult,\n\tModel,\n\tProvider,\n\tProviderStreamOptions,\n\tRefreshModelsContext,\n} from \"@earendil-works/pi-ai\";\nimport { stream, streamSimple } from \"@earendil-works/pi-ai/compat\";\nimport {\n\tLlamaClient,\n\ttype LlamaModelInfo,\n\ttype LlamaServerProps,\n\tllamaInferenceUrl,\n\tnormalizeLlamaServerUrl,\n} from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction modelIsSelectable(model: LlamaModelInfo, routerAutoload: boolean): boolean {\n\tif (model.status.value === \"loaded\") return true;\n\t// llama.cpp reports idle-slept models as \"sleeping\"; requests wake them automatically.\n\tif (model.status.value === \"sleeping\") return true;\n\t// Unloaded presets are routable only when llama.cpp router autoload can load them on first use.\n\treturn routerAutoload && model.status.value === \"unloaded\" && !model.status.failed && model.source === \"preset\";\n}\n\nasync function routerAutoloadEnabled(\n\tclient: LlamaClient,\n\tcatalog: readonly LlamaModelInfo[],\n\tsignal: AbortSignal,\n): Promise<boolean> {\n\tif (!catalog.some((model) => model.status.value === \"unloaded\" && model.source === \"preset\")) return false;\n\ttry {\n\t\treturn (await client.props({ signal })).models_autoload === true;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\nfunction toPiModel(model: LlamaModelInfo, serverUrl: string, props?: LlamaServerProps): Model<\"openai-completions\"> {\n\tconst reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;\n\tconst contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;\n\tconst reasoning = props?.chat_template?.includes(\"enable_thinking\") === true;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning,\n\t\t...(reasoning && {\n\t\t\tthinkingLevelMap: { off: \"off\", minimal: null, low: null, medium: \"medium\", high: null, xhigh: null },\n\t\t}),\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: contextWindow,\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: true,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t\t...(reasoning && { thinkingFormat: \"qwen-chat-template\" }),\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: { routerAutoload?: boolean }): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\n\tconst setCatalog = (\n\t\tcatalog: readonly LlamaModelInfo[],\n\t\tserverUrl: string,\n\t\toptions: { routerAutoload?: boolean } = {},\n\t): void => {\n\t\tmodels = catalog\n\t\t\t.filter((model) => modelIsSelectable(model, options.routerAutoload === true))\n\t\t\t.map((model) => toPiModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tif (context.stored) {\n\t\t\t\tconst restored = context.stored.models.filter(\n\t\t\t\t\t(model): model is Model<\"openai-completions\"> =>\n\t\t\t\t\t\tmodel.provider === LLAMA_PROVIDER_ID && model.api === \"openai-completions\",\n\t\t\t\t);\n\t\t\t\tif (\n\t\t\t\t\t!(await context.publish({\n\t\t\t\t\t\tupdate: () => {\n\t\t\t\t\t\t\tmodels = restored;\n\t\t\t\t\t\t},\n\t\t\t\t\t}))\n\t\t\t\t) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!context.allowNetwork || context.signal.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst client = new LlamaClient(serverUrl, context.credential.key);\n\t\t\tconst catalog = await client.list({ signal: context.signal });\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst refreshed = await Promise.all(\n\t\t\t\tcatalog\n\t\t\t\t\t.filter((model) => modelIsSelectable(model, routerAutoload))\n\t\t\t\t\t.map(async (model) => {\n\t\t\t\t\t\t// Only loaded models expose their template without side effects. Unloaded autoload presets\n\t\t\t\t\t\t// would need to be loaded, while querying sleeping models may wake them. Those models remain\n\t\t\t\t\t\t// unclassified until they are loaded or woken and a later catalog refresh discovers them.\n\t\t\t\t\t\tif (model.status.value !== \"loaded\") return toPiModel(model, serverUrl);\n\t\t\t\t\t\tconst props = await client.props({ model: model.id, signal: context.signal });\n\t\t\t\t\t\treturn toPiModel(model, serverUrl, props);\n\t\t\t\t\t}),\n\t\t\t);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tawait context.publish({\n\t\t\t\tpersist: { models: refreshed, checkedAt: Date.now() },\n\t\t\t\tupdate: () => {\n\t\t\t\t\tmodels = refreshed;\n\t\t\t\t},\n\t\t\t});\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}
+{"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,EAON,KAAK,QAAQ,EAGb,MAAM,uBAAuB,CAAC;AAG/B,OAAO,EAEN,KAAK,cAAc,EAInB,MAAM,aAAa,CAAC;AAErB,eAAO,MAAM,iBAAiB,cAAc,CAAC;AAC7C,eAAO,MAAM,wBAAwB,0BAA0B,CAAC;AA6GhE,MAAM,WAAW,uBAAuB;IACvC,QAAQ,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACzC,UAAU,CAAC,MAAM,EAAE,SAAS,cAAc,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,cAAc,CAAC,EAAE,OAAO,CAAA;KAAE,GAAG,IAAI,CAAC;CAC/G;AAED,wBAAgB,mBAAmB,IAAI,uBAAuB,CAiI7D"}

package/dist/extensions/llama/provider.js

@@ -1,3 +1,5 @@
+import { isModelType, } from "@earendil-works/pi-ai";
+import { llamaCppClassifyApi } from "@earendil-works/pi-ai/api/llama-cpp-classify.lazy";
 import { stream, streamSimple } from "@earendil-works/pi-ai/compat";
 import { LlamaClient, llamaInferenceUrl, normalizeLlamaServerUrl, } from "./client.js";
 export const LLAMA_PROVIDER_ID = "llama.cpp";
@@ -29,9 +31,46 @@ async function routerAutoloadEnabled(client, catalog, signal) {
         return false;
     }
 }
-function toPiModel(model, serverUrl, props) {
-    const reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;
-    const contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;
+function configuredContextWindow(model) {
+    const args = model.status.args ?? [];
+    for (let index = 0; index < args.length - 1; index++) {
+        const flag = args[index];
+        if (flag !== "--ctx-size" && flag !== "-c" && flag !== "-ctx")
+            continue;
+        const contextWindow = Number(args[index + 1]);
+        if (Number.isSafeInteger(contextWindow) && contextWindow > 0)
+            return contextWindow;
+    }
+    return undefined;
+}
+function contextWindowOf(model, cachedContextWindow) {
+    const runtimeContextWindow = model.meta?.n_ctx;
+    if (runtimeContextWindow && runtimeContextWindow > 0)
+        return runtimeContextWindow;
+    const configuredContext = configuredContextWindow(model);
+    if (configuredContext)
+        return configuredContext;
+    if (cachedContextWindow && cachedContextWindow > 0)
+        return cachedContextWindow;
+    const trainingContextWindow = model.meta?.n_ctx_train;
+    return trainingContextWindow && trainingContextWindow > 0 ? trainingContextWindow : 128000;
+}
+/** The same llama.cpp model used as a classifier: answers are read from next-token label probabilities. */
+function toPiClassifierModel(model, serverUrl, cachedContextWindow) {
+    return {
+        type: "classifier",
+        id: model.id,
+        name: model.id,
+        api: "llama-cpp-classify",
+        provider: LLAMA_PROVIDER_ID,
+        baseUrl: serverUrl,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: contextWindowOf(model, cachedContextWindow),
+    };
+}
+function toPiModel(model, serverUrl, props, cachedContextWindow) {
+    const contextWindow = contextWindowOf(model, cachedContextWindow);
     const reasoning = props?.chat_template?.includes("enable_thinking") === true;
     return {
         id: model.id,
@@ -60,10 +99,12 @@ function toPiModel(model, serverUrl, props) {
 }
 export function createLlamaProvider() {
     let models = [];
+    let classifiers = [];
+    const classifier = llamaCppClassifyApi();
     const setCatalog = (catalog, serverUrl, options = {}) => {
-        models = catalog
-            .filter((model) => modelIsSelectable(model, options.routerAutoload === true))
-            .map((model) => toPiModel(model, serverUrl));
+        const selectable = catalog.filter((model) => modelIsSelectable(model, options.routerAutoload === true));
+        models = selectable.map((model) => toPiModel(model, serverUrl));
+        classifiers = selectable.map((model) => toPiClassifierModel(model, serverUrl));
     };
     const provider = {
         id: LLAMA_PROVIDER_ID,
@@ -110,12 +151,20 @@ export function createLlamaProvider() {
             },
         },
         getModels: () => models,
+        getAllModels: () => [...models, ...classifiers],
         refreshModels: async (context) => {
+            const cachedContextWindows = new Map();
             if (context.stored) {
-                const restored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID && model.api === "openai-completions");
+                const stored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID);
+                const restored = stored.filter((model) => isModelType(model, "chat") && model.api === "openai-completions");
+                const restoredClassifiers = stored.filter((model) => isModelType(model, "classifier") && model.api === "llama-cpp-classify");
+                for (const model of [...restored, ...restoredClassifiers]) {
+                    cachedContextWindows.set(model.id, model.contextWindow);
+                }
                 if (!(await context.publish({
                     update: () => {
                         models = restored;
+                        classifiers = restoredClassifiers;
                     },
                 }))) {
                     return;
@@ -133,28 +182,31 @@ export function createLlamaProvider() {
             const routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);
             if (context.signal.aborted)
                 return;
-            const refreshed = await Promise.all(catalog
-                .filter((model) => modelIsSelectable(model, routerAutoload))
-                .map(async (model) => {
+            const selectable = catalog.filter((model) => modelIsSelectable(model, routerAutoload));
+            const refreshed = await Promise.all(selectable.map(async (model) => {
+                const cachedContextWindow = cachedContextWindows.get(model.id);
                 // Only loaded models expose their template without side effects. Unloaded autoload presets
                 // would need to be loaded, while querying sleeping models may wake them. Those models remain
                 // unclassified until they are loaded or woken and a later catalog refresh discovers them.
                 if (model.status.value !== "loaded")
-                    return toPiModel(model, serverUrl);
+                    return toPiModel(model, serverUrl, undefined, cachedContextWindow);
                 const props = await client.props({ model: model.id, signal: context.signal });
-                return toPiModel(model, serverUrl, props);
+                return toPiModel(model, serverUrl, props, cachedContextWindow);
             }));
+            const refreshedClassifiers = selectable.map((model) => toPiClassifierModel(model, serverUrl, cachedContextWindows.get(model.id)));
             if (context.signal.aborted)
                 return;
             await context.publish({
-                persist: { models: refreshed, checkedAt: Date.now() },
+                persist: { models: [...refreshed, ...refreshedClassifiers], checkedAt: Date.now() },
                 update: () => {
                     models = refreshed;
+                    classifiers = refreshedClassifiers;
                 },
             });
         },
         stream: (model, context, options) => stream(model, context, options),
         streamSimple: (model, context, options) => streamSimple(model, context, options),
+        classify: (model, context, options) => classifier.classify(model, context, options),
     };
     return { provider, setCatalog };
 }

package/dist/extensions/llama/provider.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,EACN,WAAW,EAGX,iBAAiB,EACjB,uBAAuB,GACvB,MAAM,aAAa,CAAC;AAErB,MAAM,CAAC,MAAM,iBAAiB,GAAG,WAAW,CAAC;AAC7C,MAAM,CAAC,MAAM,wBAAwB,GAAG,uBAAuB,CAAC;AAChE,SAAS,mBAAmB,CAAC,UAAwC,EAAsB;IAC1F,MAAM,KAAK,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,CAAC;IAC9C,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,uBAAuB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC9F;AAED,KAAK,UAAU,gBAAgB,CAC9B,GAAgB,EAChB,UAAwC,EACV;IAC9B,MAAM,UAAU,GAAG,mBAAmB,CAAC,UAAU,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;IAChG,OAAO,UAAU,CAAC,CAAC,CAAC,uBAAuB,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CACpE;AAED,SAAS,iBAAiB,CAAC,KAAqB,EAAE,cAAuB,EAAW;IACnF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IACjD,uFAAuF;IACvF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;QAAE,OAAO,IAAI,CAAC;IACnD,gGAAgG;IAChG,OAAO,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;AAAA,CAChH;AAED,KAAK,UAAU,qBAAqB,CACnC,MAAmB,EACnB,OAAkC,EAClC,MAAmB,EACA;IACnB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3G,IAAI,CAAC;QACJ,OAAO,CAAC,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,eAAe,KAAK,IAAI,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AAAA,CACD;AAED,SAAS,SAAS,CAAC,KAAqB,EAAE,SAAiB,EAAE,KAAwB,EAA+B;IACnH,MAAM,qBAAqB,GAAG,KAAK,CAAC,IAAI,EAAE,KAAK,IAAI,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;IAC3E,MAAM,aAAa,GAAG,qBAAqB,IAAI,qBAAqB,GAAG,CAAC,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,MAAM,CAAC;IAC1G,MAAM,SAAS,GAAG,KAAK,EAAE,aAAa,EAAE,QAAQ,CAAC,iBAAiB,CAAC,KAAK,IAAI,CAAC;IAC7E,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC;QACrC,SAAS;QACT,GAAG,CAAC,SAAS,IAAI;YAChB,gBAAgB,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE;SACrG,CAAC;QACF,KAAK,EAAE,KAAK,CAAC,YAAY,EAAE,gBAAgB,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7F,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa;QACb,SAAS,EAAE,aAAa;QACxB,MAAM,EAAE;YACP,aAAa,EAAE,KAAK;YACpB,qBAAqB,EAAE,KAAK;YAC5B,uBAAuB,EAAE,KAAK;YAC9B,wBAAwB,EAAE,IAAI;YAC9B,kBAAkB,EAAE,KAAK;YACzB,cAAc,EAAE,YAAY;YAC5B,GAAG,CAAC,SAAS,IAAI,EAAE,cAAc,EAAE,oBAAoB,EAAE,CAAC;SAC1D;KACD,CAAC;AAAA,CACF;AAOD,MAAM,UAAU,mBAAmB,GAA4B;IAC9D,IAAI,MAAM,GAA2C,EAAE,CAAC;IAExD,MAAM,UAAU,GAAG,CAClB,OAAkC,EAClC,SAAiB,EACjB,OAAO,GAAiC,EAAE,EACnC,EAAE,CAAC;QACV,MAAM,GAAG,OAAO;aACd,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC;aAC5E,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;IAAA,CAC9C,CAAC;IAEF,MAAM,QAAQ,GAAmC;QAChD,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,WAAW;QACjB,OAAO,EAAE,iBAAiB,CAAC,wBAAwB,CAAC;QACpD,IAAI,EAAE;YACL,MAAM,EAAE;gBACP,IAAI,EAAE,kBAAkB;gBACxB,KAAK,EAAE,KAAK,EAAE,WAAW,EAA6B,EAAE,CAAC;oBACxD,MAAM,UAAU,GAAG,MAAM,WAAW,CAAC,MAAM,CAAC;wBAC3C,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,sBAAsB;wBAC/B,WAAW,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB;qBACnE,CAAC,CAAC;oBACH,MAAM,SAAS,GAAG,uBAAuB,CACxC,UAAU,CAAC,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB,CAC3E,CAAC;oBACF,MAAM,MAAM,GAAG,CACd,MAAM,WAAW,CAAC,MAAM,CAAC;wBACxB,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,oBAAoB;qBAC7B,CAAC,CACF,CAAC,IAAI,EAAE,CAAC;oBACT,MAAM,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,IAAI,SAAS,CAAC,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;oBAC3F,OAAO;wBACN,IAAI,EAAE,SAAS;wBACf,GAAG,EAAE,MAAM,IAAI,SAAS;wBACxB,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;qBAClC,CAAC;gBAAA,CACF;gBACD,KAAK,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,EAAE,CAAC;oBACrC,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,OAAO,SAAS;wBACf,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB,EAAE;wBAClF,CAAC,CAAC,SAAS,CAAC;gBAAA,CACb;gBACD,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAmC,EAAE,CAAC;oBACxE,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,IAAI,CAAC,SAAS;wBAAE,OAAO,SAAS,CAAC;oBACjC,MAAM,MAAM,GAAG,UAAU,EAAE,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,IAAI,OAAO,CAAC;oBAC9E,OAAO;wBACN,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,EAAE;wBACvD,GAAG,EAAE,EAAE,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,EAAE,SAAS,EAAE;wBACtD,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB;qBAC3D,CAAC;gBAAA,CACF;aACD;SACD;QACD,SAAS,EAAE,GAAG,EAAE,CAAC,MAAM;QACvB,aAAa,EAAE,KAAK,EAAE,OAA6B,EAAiB,EAAE,CAAC;YACtE,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACpB,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAC5C,CAAC,KAAK,EAAwC,EAAE,CAC/C,KAAK,CAAC,QAAQ,KAAK,iBAAiB,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CAC3E,CAAC;gBACF,IACC,CAAC,CAAC,MAAM,OAAO,CAAC,OAAO,CAAC;oBACvB,MAAM,EAAE,GAAG,EAAE,CAAC;wBACb,MAAM,GAAG,QAAQ,CAAC;oBAAA,CAClB;iBACD,CAAC,CAAC,EACF,CAAC;oBACF,OAAO;gBACR,CAAC;YACF,CAAC;YAED,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,OAAO,CAAC,UAAU,EAAE,IAAI,KAAK,SAAS;gBAAE,OAAO;YACtG,MAAM,SAAS,GAAG,mBAAmB,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;YAC1D,IAAI,CAAC,SAAS;gBAAE,OAAO;YACvB,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;YAClE,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;YACpF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,OAAO;iBACL,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC;iBAC3D,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC;gBACrB,2FAA2F;gBAC3F,6FAA6F;gBAC7F,0FAA0F;gBAC1F,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAAE,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;gBACxE,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC9E,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;YAAA,CAC1C,CAAC,CACH,CAAC;YACF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,OAAO,CAAC,OAAO,CAAC;gBACrB,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE;gBACrD,MAAM,EAAE,GAAG,EAAE,CAAC;oBACb,MAAM,GAAG,SAAS,CAAC;gBAAA,CACnB;aACD,CAAC,CAAC;QAAA,CACH;QACD,MAAM,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAA4C,CAAC;QACzG,YAAY,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;KAChF,CAAC;IAEF,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC;AAAA,CAChC","sourcesContent":["import type {\n\tApiKeyCredential,\n\tAuthContext,\n\tAuthResult,\n\tModel,\n\tProvider,\n\tProviderStreamOptions,\n\tRefreshModelsContext,\n} from \"@earendil-works/pi-ai\";\nimport { stream, streamSimple } from \"@earendil-works/pi-ai/compat\";\nimport {\n\tLlamaClient,\n\ttype LlamaModelInfo,\n\ttype LlamaServerProps,\n\tllamaInferenceUrl,\n\tnormalizeLlamaServerUrl,\n} from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction modelIsSelectable(model: LlamaModelInfo, routerAutoload: boolean): boolean {\n\tif (model.status.value === \"loaded\") return true;\n\t// llama.cpp reports idle-slept models as \"sleeping\"; requests wake them automatically.\n\tif (model.status.value === \"sleeping\") return true;\n\t// Unloaded presets are routable only when llama.cpp router autoload can load them on first use.\n\treturn routerAutoload && model.status.value === \"unloaded\" && !model.status.failed && model.source === \"preset\";\n}\n\nasync function routerAutoloadEnabled(\n\tclient: LlamaClient,\n\tcatalog: readonly LlamaModelInfo[],\n\tsignal: AbortSignal,\n): Promise<boolean> {\n\tif (!catalog.some((model) => model.status.value === \"unloaded\" && model.source === \"preset\")) return false;\n\ttry {\n\t\treturn (await client.props({ signal })).models_autoload === true;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\nfunction toPiModel(model: LlamaModelInfo, serverUrl: string, props?: LlamaServerProps): Model<\"openai-completions\"> {\n\tconst reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;\n\tconst contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;\n\tconst reasoning = props?.chat_template?.includes(\"enable_thinking\") === true;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning,\n\t\t...(reasoning && {\n\t\t\tthinkingLevelMap: { off: \"off\", minimal: null, low: null, medium: \"medium\", high: null, xhigh: null },\n\t\t}),\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: contextWindow,\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: true,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t\t...(reasoning && { thinkingFormat: \"qwen-chat-template\" }),\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: { routerAutoload?: boolean }): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\n\tconst setCatalog = (\n\t\tcatalog: readonly LlamaModelInfo[],\n\t\tserverUrl: string,\n\t\toptions: { routerAutoload?: boolean } = {},\n\t): void => {\n\t\tmodels = catalog\n\t\t\t.filter((model) => modelIsSelectable(model, options.routerAutoload === true))\n\t\t\t.map((model) => toPiModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tif (context.stored) {\n\t\t\t\tconst restored = context.stored.models.filter(\n\t\t\t\t\t(model): model is Model<\"openai-completions\"> =>\n\t\t\t\t\t\tmodel.provider === LLAMA_PROVIDER_ID && model.api === \"openai-completions\",\n\t\t\t\t);\n\t\t\t\tif (\n\t\t\t\t\t!(await context.publish({\n\t\t\t\t\t\tupdate: () => {\n\t\t\t\t\t\t\tmodels = restored;\n\t\t\t\t\t\t},\n\t\t\t\t\t}))\n\t\t\t\t) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!context.allowNetwork || context.signal.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst client = new LlamaClient(serverUrl, context.credential.key);\n\t\t\tconst catalog = await client.list({ signal: context.signal });\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst refreshed = await Promise.all(\n\t\t\t\tcatalog\n\t\t\t\t\t.filter((model) => modelIsSelectable(model, routerAutoload))\n\t\t\t\t\t.map(async (model) => {\n\t\t\t\t\t\t// Only loaded models expose their template without side effects. Unloaded autoload presets\n\t\t\t\t\t\t// would need to be loaded, while querying sleeping models may wake them. Those models remain\n\t\t\t\t\t\t// unclassified until they are loaded or woken and a later catalog refresh discovers them.\n\t\t\t\t\t\tif (model.status.value !== \"loaded\") return toPiModel(model, serverUrl);\n\t\t\t\t\t\tconst props = await client.props({ model: model.id, signal: context.signal });\n\t\t\t\t\t\treturn toPiModel(model, serverUrl, props);\n\t\t\t\t\t}),\n\t\t\t);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tawait context.publish({\n\t\t\t\tpersist: { models: refreshed, checkedAt: Date.now() },\n\t\t\t\tupdate: () => {\n\t\t\t\t\tmodels = refreshed;\n\t\t\t\t},\n\t\t\t});\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}
+{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,EAKN,WAAW,GAKX,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAE,mBAAmB,EAAE,MAAM,mDAAmD,CAAC;AACxF,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,8BAA8B,CAAC;AACpE,OAAO,EACN,WAAW,EAGX,iBAAiB,EACjB,uBAAuB,GACvB,MAAM,aAAa,CAAC;AAErB,MAAM,CAAC,MAAM,iBAAiB,GAAG,WAAW,CAAC;AAC7C,MAAM,CAAC,MAAM,wBAAwB,GAAG,uBAAuB,CAAC;AAChE,SAAS,mBAAmB,CAAC,UAAwC;IACpE,MAAM,KAAK,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,CAAC;IAC9C,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,uBAAuB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/F,CAAC;AAED,KAAK,UAAU,gBAAgB,CAC9B,GAAgB,EAChB,UAAwC;IAExC,MAAM,UAAU,GAAG,mBAAmB,CAAC,UAAU,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;IAChG,OAAO,UAAU,CAAC,CAAC,CAAC,uBAAuB,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACrE,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAqB,EAAE,cAAuB;IACxE,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IACjD,uFAAuF;IACvF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;QAAE,OAAO,IAAI,CAAC;IACnD,gGAAgG;IAChG,OAAO,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;AACjH,CAAC;AAED,KAAK,UAAU,qBAAqB,CACnC,MAAmB,EACnB,OAAkC,EAClC,MAAmB;IAEnB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3G,IAAI,CAAC;QACJ,OAAO,CAAC,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,eAAe,KAAK,IAAI,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC;AAED,SAAS,uBAAuB,CAAC,KAAqB;IACrD,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;IACrC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QACzB,IAAI,IAAI,KAAK,YAAY,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,KAAK,MAAM;YAAE,SAAS;QACxE,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;QAC9C,IAAI,MAAM,CAAC,aAAa,CAAC,aAAa,CAAC,IAAI,aAAa,GAAG,CAAC;YAAE,OAAO,aAAa,CAAC;IACpF,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED,SAAS,eAAe,CAAC,KAAqB,EAAE,mBAA4B;IAC3E,MAAM,oBAAoB,GAAG,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC;IAC/C,IAAI,oBAAoB,IAAI,oBAAoB,GAAG,CAAC;QAAE,OAAO,oBAAoB,CAAC;IAClF,MAAM,iBAAiB,GAAG,uBAAuB,CAAC,KAAK,CAAC,CAAC;IACzD,IAAI,iBAAiB;QAAE,OAAO,iBAAiB,CAAC;IAChD,IAAI,mBAAmB,IAAI,mBAAmB,GAAG,CAAC;QAAE,OAAO,mBAAmB,CAAC;IAC/E,MAAM,qBAAqB,GAAG,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;IACtD,OAAO,qBAAqB,IAAI,qBAAqB,GAAG,CAAC,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,MAAM,CAAC;AAC5F,CAAC;AAED,2GAA2G;AAC3G,SAAS,mBAAmB,CAC3B,KAAqB,EACrB,SAAiB,EACjB,mBAA4B;IAE5B,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,SAAS;QAClB,KAAK,EAAE,CAAC,MAAM,CAAC;QACf,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa,EAAE,eAAe,CAAC,KAAK,EAAE,mBAAmB,CAAC;KAC1D,CAAC;AACH,CAAC;AAED,SAAS,SAAS,CACjB,KAAqB,EACrB,SAAiB,EACjB,KAAwB,EACxB,mBAA4B;IAE5B,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,EAAE,mBAAmB,CAAC,CAAC;IAClE,MAAM,SAAS,GAAG,KAAK,EAAE,aAAa,EAAE,QAAQ,CAAC,iBAAiB,CAAC,KAAK,IAAI,CAAC;IAC7E,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC;QACrC,SAAS;QACT,GAAG,CAAC,SAAS,IAAI;YAChB,gBAAgB,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE;SACrG,CAAC;QACF,KAAK,EAAE,KAAK,CAAC,YAAY,EAAE,gBAAgB,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7F,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa;QACb,SAAS,EAAE,aAAa;QACxB,MAAM,EAAE;YACP,aAAa,EAAE,KAAK;YACpB,qBAAqB,EAAE,KAAK;YAC5B,uBAAuB,EAAE,KAAK;YAC9B,wBAAwB,EAAE,IAAI;YAC9B,kBAAkB,EAAE,KAAK;YACzB,cAAc,EAAE,YAAY;YAC5B,GAAG,CAAC,SAAS,IAAI,EAAE,cAAc,EAAE,oBAAoB,EAAE,CAAC;SAC1D;KACD,CAAC;AACH,CAAC;AAOD,MAAM,UAAU,mBAAmB;IAClC,IAAI,MAAM,GAA2C,EAAE,CAAC;IACxD,IAAI,WAAW,GAAqD,EAAE,CAAC;IACvE,MAAM,UAAU,GAAG,mBAAmB,EAAE,CAAC;IAEzC,MAAM,UAAU,GAAG,CAClB,OAAkC,EAClC,SAAiB,EACjB,OAAO,GAAiC,EAAE,EACnC,EAAE;QACT,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,CAAC;QACxG,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;QAChE,WAAW,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;IAChF,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAmC;QAChD,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,WAAW;QACjB,OAAO,EAAE,iBAAiB,CAAC,wBAAwB,CAAC;QACpD,IAAI,EAAE;YACL,MAAM,EAAE;gBACP,IAAI,EAAE,kBAAkB;gBACxB,KAAK,EAAE,KAAK,EAAE,WAAW,EAA6B,EAAE;oBACvD,MAAM,UAAU,GAAG,MAAM,WAAW,CAAC,MAAM,CAAC;wBAC3C,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,sBAAsB;wBAC/B,WAAW,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB;qBACnE,CAAC,CAAC;oBACH,MAAM,SAAS,GAAG,uBAAuB,CACxC,UAAU,CAAC,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB,CAC3E,CAAC;oBACF,MAAM,MAAM,GAAG,CACd,MAAM,WAAW,CAAC,MAAM,CAAC;wBACxB,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,oBAAoB;qBAC7B,CAAC,CACF,CAAC,IAAI,EAAE,CAAC;oBACT,MAAM,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,IAAI,SAAS,CAAC,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;oBAC3F,OAAO;wBACN,IAAI,EAAE,SAAS;wBACf,GAAG,EAAE,MAAM,IAAI,SAAS;wBACxB,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;qBAClC,CAAC;gBACH,CAAC;gBACD,KAAK,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,EAAE;oBACpC,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,OAAO,SAAS;wBACf,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB,EAAE;wBAClF,CAAC,CAAC,SAAS,CAAC;gBACd,CAAC;gBACD,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAmC,EAAE;oBACvE,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,IAAI,CAAC,SAAS;wBAAE,OAAO,SAAS,CAAC;oBACjC,MAAM,MAAM,GAAG,UAAU,EAAE,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,IAAI,OAAO,CAAC;oBAC9E,OAAO;wBACN,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,EAAE;wBACvD,GAAG,EAAE,EAAE,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,EAAE,SAAS,EAAE;wBACtD,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB;qBAC3D,CAAC;gBACH,CAAC;aACD;SACD;QACD,SAAS,EAAE,GAAG,EAAE,CAAC,MAAM;QACvB,YAAY,EAAE,GAAG,EAAE,CAAC,CAAC,GAAG,MAAM,EAAE,GAAG,WAAW,CAAC;QAC/C,aAAa,EAAE,KAAK,EAAE,OAA6B,EAAiB,EAAE;YACrE,MAAM,oBAAoB,GAAG,IAAI,GAAG,EAAkB,CAAC;YACvD,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACpB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,iBAAiB,CAAC,CAAC;gBAC7F,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAC7B,CAAC,KAAK,EAAwC,EAAE,CAC/C,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CACjE,CAAC;gBACF,MAAM,mBAAmB,GAAG,MAAM,CAAC,MAAM,CACxC,CAAC,KAAK,EAAkD,EAAE,CACzD,WAAW,CAAC,KAAK,EAAE,YAAY,CAAC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CACvE,CAAC;gBACF,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,QAAQ,EAAE,GAAG,mBAAmB,CAAC,EAAE,CAAC;oBAC3D,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,aAAa,CAAC,CAAC;gBACzD,CAAC;gBACD,IACC,CAAC,CAAC,MAAM,OAAO,CAAC,OAAO,CAAC;oBACvB,MAAM,EAAE,GAAG,EAAE;wBACZ,MAAM,GAAG,QAAQ,CAAC;wBAClB,WAAW,GAAG,mBAAmB,CAAC;oBACnC,CAAC;iBACD,CAAC,CAAC,EACF,CAAC;oBACF,OAAO;gBACR,CAAC;YACF,CAAC;YAED,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,OAAO,CAAC,UAAU,EAAE,IAAI,KAAK,SAAS;gBAAE,OAAO;YACtG,MAAM,SAAS,GAAG,mBAAmB,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;YAC1D,IAAI,CAAC,SAAS;gBAAE,OAAO;YACvB,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;YAClE,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;YACpF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC,CAAC;YACvF,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,UAAU,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;gBAC9B,MAAM,mBAAmB,GAAG,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;gBAC/D,2FAA2F;gBAC3F,6FAA6F;gBAC7F,0FAA0F;gBAC1F,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAAE,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,mBAAmB,CAAC,CAAC;gBACxG,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC9E,OAAO,SAAS,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,mBAAmB,CAAC,CAAC;YAChE,CAAC,CAAC,CACF,CAAC;YACF,MAAM,oBAAoB,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACrD,mBAAmB,CAAC,KAAK,EAAE,SAAS,EAAE,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CACzE,CAAC;YACF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,OAAO,CAAC,OAAO,CAAC;gBACrB,OAAO,EAAE,EAAE,MAAM,EAAE,CAAC,GAAG,SAAS,EAAE,GAAG,oBAAoB,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE;gBACnF,MAAM,EAAE,GAAG,EAAE;oBACZ,MAAM,GAAG,SAAS,CAAC;oBACnB,WAAW,GAAG,oBAAoB,CAAC;gBACpC,CAAC;aACD,CAAC,CAAC;QACJ,CAAC;QACD,MAAM,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAA4C,CAAC;QACzG,YAAY,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;QAChF,QAAQ,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;KACnF,CAAC;IAEF,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC;AACjC,CAAC","sourcesContent":["import {\n\ttype ApiKeyCredential,\n\ttype AuthContext,\n\ttype AuthResult,\n\ttype ClassifierModel,\n\tisModelType,\n\ttype Model,\n\ttype Provider,\n\ttype ProviderStreamOptions,\n\ttype RefreshModelsContext,\n} from \"@earendil-works/pi-ai\";\nimport { llamaCppClassifyApi } from \"@earendil-works/pi-ai/api/llama-cpp-classify.lazy\";\nimport { stream, streamSimple } from \"@earendil-works/pi-ai/compat\";\nimport {\n\tLlamaClient,\n\ttype LlamaModelInfo,\n\ttype LlamaServerProps,\n\tllamaInferenceUrl,\n\tnormalizeLlamaServerUrl,\n} from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction modelIsSelectable(model: LlamaModelInfo, routerAutoload: boolean): boolean {\n\tif (model.status.value === \"loaded\") return true;\n\t// llama.cpp reports idle-slept models as \"sleeping\"; requests wake them automatically.\n\tif (model.status.value === \"sleeping\") return true;\n\t// Unloaded presets are routable only when llama.cpp router autoload can load them on first use.\n\treturn routerAutoload && model.status.value === \"unloaded\" && !model.status.failed && model.source === \"preset\";\n}\n\nasync function routerAutoloadEnabled(\n\tclient: LlamaClient,\n\tcatalog: readonly LlamaModelInfo[],\n\tsignal: AbortSignal,\n): Promise<boolean> {\n\tif (!catalog.some((model) => model.status.value === \"unloaded\" && model.source === \"preset\")) return false;\n\ttry {\n\t\treturn (await client.props({ signal })).models_autoload === true;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\nfunction configuredContextWindow(model: LlamaModelInfo): number | undefined {\n\tconst args = model.status.args ?? [];\n\tfor (let index = 0; index < args.length - 1; index++) {\n\t\tconst flag = args[index];\n\t\tif (flag !== \"--ctx-size\" && flag !== \"-c\" && flag !== \"-ctx\") continue;\n\t\tconst contextWindow = Number(args[index + 1]);\n\t\tif (Number.isSafeInteger(contextWindow) && contextWindow > 0) return contextWindow;\n\t}\n\treturn undefined;\n}\n\nfunction contextWindowOf(model: LlamaModelInfo, cachedContextWindow?: number): number {\n\tconst runtimeContextWindow = model.meta?.n_ctx;\n\tif (runtimeContextWindow && runtimeContextWindow > 0) return runtimeContextWindow;\n\tconst configuredContext = configuredContextWindow(model);\n\tif (configuredContext) return configuredContext;\n\tif (cachedContextWindow && cachedContextWindow > 0) return cachedContextWindow;\n\tconst trainingContextWindow = model.meta?.n_ctx_train;\n\treturn trainingContextWindow && trainingContextWindow > 0 ? trainingContextWindow : 128000;\n}\n\n/** The same llama.cpp model used as a classifier: answers are read from next-token label probabilities. */\nfunction toPiClassifierModel(\n\tmodel: LlamaModelInfo,\n\tserverUrl: string,\n\tcachedContextWindow?: number,\n): ClassifierModel<\"llama-cpp-classify\"> {\n\treturn {\n\t\ttype: \"classifier\",\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"llama-cpp-classify\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: serverUrl,\n\t\tinput: [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow: contextWindowOf(model, cachedContextWindow),\n\t};\n}\n\nfunction toPiModel(\n\tmodel: LlamaModelInfo,\n\tserverUrl: string,\n\tprops?: LlamaServerProps,\n\tcachedContextWindow?: number,\n): Model<\"openai-completions\"> {\n\tconst contextWindow = contextWindowOf(model, cachedContextWindow);\n\tconst reasoning = props?.chat_template?.includes(\"enable_thinking\") === true;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning,\n\t\t...(reasoning && {\n\t\t\tthinkingLevelMap: { off: \"off\", minimal: null, low: null, medium: \"medium\", high: null, xhigh: null },\n\t\t}),\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: contextWindow,\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: true,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t\t...(reasoning && { thinkingFormat: \"qwen-chat-template\" }),\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: { routerAutoload?: boolean }): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\tlet classifiers: readonly ClassifierModel<\"llama-cpp-classify\">[] = [];\n\tconst classifier = llamaCppClassifyApi();\n\n\tconst setCatalog = (\n\t\tcatalog: readonly LlamaModelInfo[],\n\t\tserverUrl: string,\n\t\toptions: { routerAutoload?: boolean } = {},\n\t): void => {\n\t\tconst selectable = catalog.filter((model) => modelIsSelectable(model, options.routerAutoload === true));\n\t\tmodels = selectable.map((model) => toPiModel(model, serverUrl));\n\t\tclassifiers = selectable.map((model) => toPiClassifierModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\tgetAllModels: () => [...models, ...classifiers],\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tconst cachedContextWindows = new Map<string, number>();\n\t\t\tif (context.stored) {\n\t\t\t\tconst stored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID);\n\t\t\t\tconst restored = stored.filter(\n\t\t\t\t\t(model): model is Model<\"openai-completions\"> =>\n\t\t\t\t\t\tisModelType(model, \"chat\") && model.api === \"openai-completions\",\n\t\t\t\t);\n\t\t\t\tconst restoredClassifiers = stored.filter(\n\t\t\t\t\t(model): model is ClassifierModel<\"llama-cpp-classify\"> =>\n\t\t\t\t\t\tisModelType(model, \"classifier\") && model.api === \"llama-cpp-classify\",\n\t\t\t\t);\n\t\t\t\tfor (const model of [...restored, ...restoredClassifiers]) {\n\t\t\t\t\tcachedContextWindows.set(model.id, model.contextWindow);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\t!(await context.publish({\n\t\t\t\t\t\tupdate: () => {\n\t\t\t\t\t\t\tmodels = restored;\n\t\t\t\t\t\t\tclassifiers = restoredClassifiers;\n\t\t\t\t\t\t},\n\t\t\t\t\t}))\n\t\t\t\t) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!context.allowNetwork || context.signal.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst client = new LlamaClient(serverUrl, context.credential.key);\n\t\t\tconst catalog = await client.list({ signal: context.signal });\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst selectable = catalog.filter((model) => modelIsSelectable(model, routerAutoload));\n\t\t\tconst refreshed = await Promise.all(\n\t\t\t\tselectable.map(async (model) => {\n\t\t\t\t\tconst cachedContextWindow = cachedContextWindows.get(model.id);\n\t\t\t\t\t// Only loaded models expose their template without side effects. Unloaded autoload presets\n\t\t\t\t\t// would need to be loaded, while querying sleeping models may wake them. Those models remain\n\t\t\t\t\t// unclassified until they are loaded or woken and a later catalog refresh discovers them.\n\t\t\t\t\tif (model.status.value !== \"loaded\") return toPiModel(model, serverUrl, undefined, cachedContextWindow);\n\t\t\t\t\tconst props = await client.props({ model: model.id, signal: context.signal });\n\t\t\t\t\treturn toPiModel(model, serverUrl, props, cachedContextWindow);\n\t\t\t\t}),\n\t\t\t);\n\t\t\tconst refreshedClassifiers = selectable.map((model) =>\n\t\t\t\ttoPiClassifierModel(model, serverUrl, cachedContextWindows.get(model.id)),\n\t\t\t);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tawait context.publish({\n\t\t\t\tpersist: { models: [...refreshed, ...refreshedClassifiers], checkedAt: Date.now() },\n\t\t\t\tupdate: () => {\n\t\t\t\t\tmodels = refreshed;\n\t\t\t\t\tclassifiers = refreshedClassifiers;\n\t\t\t\t},\n\t\t\t});\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t\tclassify: (model, context, options) => classifier.classify(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}

package/dist/extensions/llama/ui.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ui.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/ui.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,gCAAgC,CAAC;AAK9E,OAAO,KAAK,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AACjE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAIzD,MAAM,MAAM,kBAAkB,GAAG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,KAAK,EAAE,cAAc,CAAA;CAAE,GAAG;IAAE,IAAI,EAAE,UAAU,CAAA;CAAE,GAAG;IAAE,IAAI,EAAE,OAAO,CAAA;CAAE,CAAC;AAErH,UAAU,aAAc,SAAQ,aAAa;IAC5C,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACd;AA+CD,MAAM,WAAW,OAAO;IACvB,UAAU,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAAC;IACrF,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;IACtE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC1D,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,GAAG,OAAO,CAAC,CAAC;IAChF,YAAY,CACX,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,gBAAgB,EAAE,CAAC,GACzE,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;IAC/B,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACjD,QAAQ,CAAC,KAAK,EAAE,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC9C,cAAc,CAAC,KAAK,EAAE,aAAa,GAAG,IAAI,CAAC;CAC3C;AAwYD,wBAAsB,WAAW,CAAC,GAAG,EAAE,uBAAuB,EAAE,GAAG,EAAE,CAAC,EAAE,EAAE,OAAO,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAYlH;AAED,wBAAsB,eAAe,CAAC,CAAC,EACtC,EAAE,EAAE,OAAO,EACX,OAAO,EAAE;IACR,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,GAAG,CAAC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,CAAC,QAAQ,EAAE,aAAa,KAAK,IAAI,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAChF,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACxB,GACC,OAAO,CAAC;IAAE,SAAS,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,SAAS,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,CAAC,CAAA;CAAE,CAAC,CAqC/D","sourcesContent":["import {\n\ttype Component,\n\tContainer,\n\ttype Focusable,\n\tfuzzyFilter,\n\tInput,\n\ttype SelectItem,\n\tSelectList,\n\tSpacer,\n\tText,\n\ttype TUI,\n\ttruncateToWidth,\n\tvisibleWidth,\n} from \"@earendil-works/pi-tui\";\nimport type { ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport type { KeybindingsManager } from \"../../core/keybindings.ts\";\nimport { DynamicBorder } from \"../../modes/interactive/components/dynamic-border.ts\";\nimport { keyHint } from \"../../modes/interactive/components/keybinding-hints.ts\";\nimport type { Theme } from \"../../modes/interactive/theme/theme.ts\";\nimport type { LlamaModelInfo, LlamaProgress } from \"./client.ts\";\nimport type { HuggingFaceModel } from \"./huggingface.ts\";\n\nconst DOWNLOAD_VALUE = \"\\0download\";\n\nexport type LlamaManagerAction = { type: \"model\"; model: LlamaModelInfo } | { type: \"download\" } | { type: \"close\" };\n\ninterface ProgressState extends LlamaProgress {\n\ttitle: string;\n\tmodel: string;\n}\n\nfunction contextLabel(model: LlamaModelInfo): string | undefined {\n\tconst context = model.meta?.n_ctx ?? model.meta?.n_ctx_train;\n\tif (context) return context >= 1000 ? `${Math.round(context / 1000)}k` : String(context);\n\tconst args = model.status.args ?? [];\n\tfor (let index = 0; index < args.length - 1; index++) {\n\t\tif (args[index] !== \"--ctx-size\" && args[index] !== \"-c\" && args[index] !== \"-ctx\") continue;\n\t\tconst value = Number(args[index + 1]);\n\t\tif (Number.isFinite(value) && value > 0) return value >= 1000 ? `${Math.round(value / 1000)}k` : String(value);\n\t}\n\treturn undefined;\n}\n\nfunction modelDescription(model: LlamaModelInfo): string {\n\tconst details: string[] = [];\n\tconst loaded = model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n\tif (loaded) details.push(\"loaded\");\n\telse if (model.status.value !== \"unloaded\") details.push(model.status.value);\n\tconst context = loaded ? contextLabel(model) : undefined;\n\tif (context) details.push(`${context} context`);\n\treturn details.join(\" · \");\n}\n\nfunction selectTheme(theme: Theme) {\n\treturn {\n\t\tselectedPrefix: (text: string) => theme.fg(\"accent\", text),\n\t\tselectedText: (text: string) => theme.fg(\"accent\", text),\n\t\tdescription: (text: string) => theme.fg(\"muted\", text),\n\t\tscrollInfo: (text: string) => theme.fg(\"dim\", text),\n\t\tnoMatch: (text: string) => theme.fg(\"warning\", text),\n\t};\n}\n\nfunction frame(theme: Theme, title: string, body: Component[], footer?: string): Container {\n\tconst container = new Container();\n\tcontainer.addChild(new DynamicBorder((text: string) => theme.fg(\"accent\", text)));\n\tcontainer.addChild(new Text(theme.fg(\"accent\", theme.bold(title)), 1, 0));\n\tfor (const child of body) container.addChild(child);\n\tif (footer) {\n\t\tcontainer.addChild(new Spacer(1));\n\t\tcontainer.addChild(new Text(theme.fg(\"dim\", footer), 1, 0));\n\t}\n\tcontainer.addChild(new DynamicBorder((text: string) => theme.fg(\"accent\", text)));\n\treturn container;\n}\n\nexport interface LlamaUi {\n\tshowModels(serverUrl: string, models: LlamaModelInfo[]): Promise<LlamaManagerAction>;\n\tselect(title: string, options: string[]): Promise<string | undefined>;\n\tconfirm(title: string, message: string): Promise<boolean>;\n\tconnectionError(serverUrl: string, message: string): Promise<\"retry\" | \"close\">;\n\tsearchModels(\n\t\tsearch: (query: string, signal: AbortSignal) => Promise<HuggingFaceModel[]>,\n\t): Promise<string | undefined>;\n\tshowStatus(title: string, message: string): void;\n\tprogress(state: ProgressState): Promise<void>;\n\tupdateProgress(state: ProgressState): void;\n}\n\nfunction compactCount(value: number): string {\n\tif (value >= 1_000_000) return `${(value / 1_000_000).toFixed(value >= 10_000_000 ? 0 : 1)}M`;\n\tif (value >= 1_000) return `${(value / 1_000).toFixed(value >= 100_000 ? 0 : 1)}k`;\n\treturn String(value);\n}\n\nclass HuggingFaceSearch extends Container implements Focusable {\n\tprivate readonly tui: TUI;\n\tprivate readonly theme: Theme;\n\tprivate readonly keybindings: KeybindingsManager;\n\tprivate readonly search: (query: string, signal: AbortSignal) => Promise<HuggingFaceModel[]>;\n\tprivate readonly cache: Map<string, HuggingFaceModel[]>;\n\tprivate readonly onSelectModel: (model: string | undefined) => void;\n\tprivate readonly input = new Input();\n\tprivate readonly resultsContainer = new Container();\n\tprivate results: HuggingFaceModel[] = [];\n\tprivate filteredResults: HuggingFaceModel[] = [];\n\tprivate selectedIndex = 0;\n\tprivate query = \"\";\n\tprivate status = \"Type at least 2 characters\";\n\tprivate debounce: ReturnType<typeof setTimeout> | undefined;\n\tprivate request: AbortController | undefined;\n\tprivate closed = false;\n\tprivate _focused = false;\n\n\tconstructor(\n\t\ttui: TUI,\n\t\ttheme: Theme,\n\t\tkeybindings: KeybindingsManager,\n\t\tsearch: (query: string, signal: AbortSignal) => Promise<HuggingFaceModel[]>,\n\t\tcache: Map<string, HuggingFaceModel[]>,\n\t\tonSelectModel: (model: string | undefined) => void,\n\t) {\n\t\tsuper();\n\t\tthis.tui = tui;\n\t\tthis.theme = theme;\n\t\tthis.keybindings = keybindings;\n\t\tthis.search = search;\n\t\tthis.cache = cache;\n\t\tthis.onSelectModel = onSelectModel;\n\t\tthis.addChild(new Text(theme.fg(\"dim\", \"Model name or owner/repository[:quant]\"), 1, 0));\n\t\tthis.addChild(this.input);\n\t\tthis.addChild(new Spacer(1));\n\t\tthis.addChild(this.resultsContainer);\n\t\tthis.updateResults();\n\t}\n\n\tget focused(): boolean {\n\t\treturn this._focused;\n\t}\n\n\tset focused(value: boolean) {\n\t\tthis._focused = value;\n\t\tthis.input.focused = value;\n\t}\n\n\tprivate updateResults(): void {\n\t\tthis.resultsContainer.clear();\n\t\tconst maxVisible = 10;\n\t\tconst start = Math.max(\n\t\t\t0,\n\t\t\tMath.min(this.selectedIndex - Math.floor(maxVisible / 2), this.filteredResults.length - maxVisible),\n\t\t);\n\t\tconst end = Math.min(start + maxVisible, this.filteredResults.length);\n\t\tfor (let index = start; index < end; index++) {\n\t\t\tconst model = this.filteredResults[index];\n\t\t\tif (!model) continue;\n\t\t\tconst prefix = index === this.selectedIndex ? \"→ \" : \"  \";\n\t\t\tconst details = `${compactCount(model.downloads)} downloads`;\n\t\t\tthis.resultsContainer.addChild(\n\t\t\t\tnew Text(\n\t\t\t\t\tindex === this.selectedIndex\n\t\t\t\t\t\t? this.theme.fg(\"accent\", `${prefix}${model.id}  ${details}`)\n\t\t\t\t\t\t: `${prefix}${model.id}${this.theme.fg(\"muted\", `  ${details}`)}`,\n\t\t\t\t\t0,\n\t\t\t\t\t0,\n\t\t\t\t),\n\t\t\t);\n\t\t}\n\t\tif (start > 0 || end < this.filteredResults.length) {\n\t\t\tthis.resultsContainer.addChild(\n\t\t\t\tnew Text(this.theme.fg(\"dim\", `  (${this.selectedIndex + 1}/${this.filteredResults.length})`), 0, 0),\n\t\t\t);\n\t\t}\n\t\tif (this.filteredResults.length === 0) {\n\t\t\tthis.resultsContainer.addChild(new Text(this.theme.fg(\"dim\", `  ${this.status}`), 0, 0));\n\t\t} else if (this.status === \"Searching Hugging Face…\") {\n\t\t\tthis.resultsContainer.addChild(new Text(this.theme.fg(\"dim\", `  ${this.status}`), 0, 0));\n\t\t}\n\t\tthis.tui.requestRender();\n\t}\n\n\tprivate filterResults(): void {\n\t\tif (this.query) {\n\t\t\tconst matches = new Set(fuzzyFilter(this.results, this.query, (model) => model.id).map((model) => model.id));\n\t\t\tthis.filteredResults = this.results.filter((model) => matches.has(model.id));\n\t\t} else {\n\t\t\tthis.filteredResults = this.results;\n\t\t}\n\t\tthis.selectedIndex = Math.min(this.selectedIndex, Math.max(0, this.filteredResults.length - 1));\n\t\tthis.updateResults();\n\t}\n\n\tprivate scheduleSearch(): void {\n\t\tif (this.debounce) clearTimeout(this.debounce);\n\t\tthis.request?.abort();\n\t\tthis.request = undefined;\n\t\tif (this.query.length < 2) {\n\t\t\tthis.status = \"Type at least 2 characters\";\n\t\t\tthis.filterResults();\n\t\t\treturn;\n\t\t}\n\t\tconst cached = this.cache.get(this.query.toLowerCase());\n\t\tif (cached) {\n\t\t\tthis.results = cached;\n\t\t\tthis.status = cached.length === 0 ? \"No GGUF models found\" : \"\";\n\t\t\tthis.filterResults();\n\t\t\treturn;\n\t\t}\n\t\tthis.status = \"Searching Hugging Face…\";\n\t\tthis.filterResults();\n\t\tthis.debounce = setTimeout(() => void this.runSearch(this.query), 500);\n\t}\n\n\tprivate async runSearch(query: string): Promise<void> {\n\t\tconst request = new AbortController();\n\t\tthis.request = request;\n\t\ttry {\n\t\t\tconst results = await this.search(query, request.signal);\n\t\t\tthis.cache.set(query.toLowerCase(), results);\n\t\t\tif (this.closed || request.signal.aborted || this.query !== query) return;\n\t\t\tthis.results = results;\n\t\t\tthis.selectedIndex = 0;\n\t\t\tthis.status = results.length === 0 ? \"No GGUF models found\" : \"\";\n\t\t\tthis.filterResults();\n\t\t} catch (error) {\n\t\t\tif (this.closed || request.signal.aborted || this.query !== query) return;\n\t\t\tthis.results = [];\n\t\t\tthis.status = error instanceof Error ? error.message : String(error);\n\t\t\tthis.filterResults();\n\t\t} finally {\n\t\t\tif (this.request === request) this.request = undefined;\n\t\t}\n\t}\n\n\tprivate close(model: string | undefined): void {\n\t\tif (this.closed) return;\n\t\tthis.closed = true;\n\t\tif (this.debounce) clearTimeout(this.debounce);\n\t\tthis.request?.abort();\n\t\tthis.onSelectModel(model);\n\t}\n\n\thandleInput(data: string): void {\n\t\tif (this.keybindings.matches(data, \"tui.select.up\")) {\n\t\t\tif (this.filteredResults.length > 0) {\n\t\t\t\tthis.selectedIndex = this.selectedIndex === 0 ? this.filteredResults.length - 1 : this.selectedIndex - 1;\n\t\t\t\tthis.updateResults();\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (this.keybindings.matches(data, \"tui.select.down\")) {\n\t\t\tif (this.filteredResults.length > 0) {\n\t\t\t\tthis.selectedIndex = this.selectedIndex === this.filteredResults.length - 1 ? 0 : this.selectedIndex + 1;\n\t\t\t\tthis.updateResults();\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (this.keybindings.matches(data, \"tui.select.confirm\")) {\n\t\t\tconst exact = /^[^/\\s]+\\/[^:\\s]+(?::[^\\s:]+)?$/u.test(this.query) ? this.query : undefined;\n\t\t\tconst selected = exact ?? this.filteredResults[this.selectedIndex]?.id;\n\t\t\tif (selected) this.close(selected);\n\t\t\treturn;\n\t\t}\n\t\tif (this.keybindings.matches(data, \"tui.select.cancel\")) {\n\t\t\tthis.close(undefined);\n\t\t\treturn;\n\t\t}\n\t\tthis.input.handleInput(data);\n\t\tconst query = this.input.getValue().trim();\n\t\tif (query === this.query) return;\n\t\tthis.query = query;\n\t\tthis.scheduleSearch();\n\t}\n}\n\nclass LlamaView implements LlamaUi, Focusable {\n\tprivate readonly tui: TUI;\n\tprivate readonly theme: Theme;\n\tprivate readonly keybindings: KeybindingsManager;\n\tprivate readonly searchCache = new Map<string, HuggingFaceModel[]>();\n\tprivate content: Container;\n\tprivate inputHandler: { handleInput?(data: string): void } | undefined;\n\tprivate inputTarget: Focusable | undefined;\n\tprivate progressPromise: Promise<void> | undefined;\n\tprivate progressResolver: (() => void) | undefined;\n\tprivate showingProgress = false;\n\tprivate _focused = false;\n\n\tconstructor(tui: TUI, theme: Theme, keybindings: KeybindingsManager) {\n\t\tthis.tui = tui;\n\t\tthis.theme = theme;\n\t\tthis.keybindings = keybindings;\n\t\tthis.content = frame(theme, \"llama.cpp models\", [new Text(theme.fg(\"muted\", \"Loading…\"), 1, 1)]);\n\t}\n\n\tget focused(): boolean {\n\t\treturn this._focused;\n\t}\n\n\tset focused(value: boolean) {\n\t\tthis._focused = value;\n\t\tif (this.inputTarget) this.inputTarget.focused = value;\n\t}\n\n\tprivate setContent(\n\t\tcontent: Container,\n\t\tinputHandler?: { handleInput?(data: string): void },\n\t\tinputTarget?: Focusable,\n\t): void {\n\t\tif (this.inputTarget) this.inputTarget.focused = false;\n\t\tthis.progressPromise = undefined;\n\t\tthis.progressResolver = undefined;\n\t\tthis.showingProgress = false;\n\t\tthis.content = content;\n\t\tthis.inputHandler = inputHandler;\n\t\tthis.inputTarget = inputTarget;\n\t\tif (this.inputTarget) this.inputTarget.focused = this._focused;\n\t\tthis.tui.requestRender();\n\t}\n\n\tshowModels(serverUrl: string, models: LlamaModelInfo[]): Promise<LlamaManagerAction> {\n\t\tconst sorted = [...models].sort((left, right) => {\n\t\t\tconst loaded = Number(right.status.value === \"loaded\") - Number(left.status.value === \"loaded\");\n\t\t\treturn loaded || left.id.localeCompare(right.id);\n\t\t});\n\t\tconst byId = new Map(sorted.map((model) => [model.id, model]));\n\t\tconst items: SelectItem[] = [\n\t\t\t...sorted.map((model) => ({\n\t\t\t\tvalue: model.id,\n\t\t\t\tlabel: model.id,\n\t\t\t\tdescription: modelDescription(model),\n\t\t\t})),\n\t\t\t{ value: DOWNLOAD_VALUE, label: \"Download model…\", description: \"Hugging Face owner/repository[:quant]\" },\n\t\t];\n\t\treturn new Promise((resolve) => {\n\t\t\tconst list = new SelectList(items, Math.min(items.length, 12), selectTheme(this.theme), {\n\t\t\t\tminPrimaryColumnWidth: 36,\n\t\t\t\tmaxPrimaryColumnWidth: 56,\n\t\t\t});\n\t\t\tlist.onSelect = (item) => {\n\t\t\t\tif (item.value === DOWNLOAD_VALUE) resolve({ type: \"download\" });\n\t\t\t\telse {\n\t\t\t\t\tconst model = byId.get(item.value);\n\t\t\t\t\tif (model) resolve({ type: \"model\", model });\n\t\t\t\t}\n\t\t\t};\n\t\t\tlist.onCancel = () => resolve({ type: \"close\" });\n\t\t\tthis.setContent(\n\t\t\t\tframe(\n\t\t\t\t\tthis.theme,\n\t\t\t\t\t\"llama.cpp models\",\n\t\t\t\t\t[new Text(this.theme.fg(\"dim\", serverUrl), 1, 0), new Spacer(1), list],\n\t\t\t\t\t`${keyHint(\"tui.select.confirm\", \"load/unload/download\")} • ${keyHint(\"tui.select.cancel\", \"close\")}`,\n\t\t\t\t),\n\t\t\t\tlist,\n\t\t\t);\n\t\t});\n\t}\n\n\tselect(title: string, options: string[]): Promise<string | undefined> {\n\t\treturn new Promise((resolve) => {\n\t\t\tconst list = new SelectList(\n\t\t\t\toptions.map((option) => ({ value: option, label: option })),\n\t\t\t\tMath.min(options.length, 12),\n\t\t\t\tselectTheme(this.theme),\n\t\t\t);\n\t\t\tlist.onSelect = (item) => resolve(item.value);\n\t\t\tlist.onCancel = () => resolve(undefined);\n\t\t\tthis.setContent(\n\t\t\t\tframe(\n\t\t\t\t\tthis.theme,\n\t\t\t\t\ttitle,\n\t\t\t\t\t[new Spacer(1), list],\n\t\t\t\t\t`${keyHint(\"tui.select.confirm\", \"select\")} • ${keyHint(\"tui.select.cancel\", \"cancel\")}`,\n\t\t\t\t),\n\t\t\t\tlist,\n\t\t\t);\n\t\t});\n\t}\n\n\tasync confirm(title: string, message: string): Promise<boolean> {\n\t\treturn (await this.select(`${title}\\n${message}`, [\"Yes\", \"No\"])) === \"Yes\";\n\t}\n\n\tasync connectionError(serverUrl: string, message: string): Promise<\"retry\" | \"close\"> {\n\t\tconst choice = await this.select(`llama.cpp unavailable\\n${serverUrl}\\n\\n${message}`, [\"Retry\", \"Close\"]);\n\t\treturn choice === \"Retry\" ? \"retry\" : \"close\";\n\t}\n\n\tsearchModels(\n\t\tsearch: (query: string, signal: AbortSignal) => Promise<HuggingFaceModel[]>,\n\t): Promise<string | undefined> {\n\t\treturn new Promise((resolve) => {\n\t\t\tconst component = new HuggingFaceSearch(\n\t\t\t\tthis.tui,\n\t\t\t\tthis.theme,\n\t\t\t\tthis.keybindings,\n\t\t\t\tsearch,\n\t\t\t\tthis.searchCache,\n\t\t\t\tresolve,\n\t\t\t);\n\t\t\tthis.setContent(\n\t\t\t\tframe(\n\t\t\t\t\tthis.theme,\n\t\t\t\t\t\"Download model\",\n\t\t\t\t\t[new Spacer(1), component],\n\t\t\t\t\t`${keyHint(\"tui.select.confirm\", \"select\")} • ${keyHint(\"tui.select.cancel\", \"back\")}`,\n\t\t\t\t),\n\t\t\t\tcomponent,\n\t\t\t\tcomponent,\n\t\t\t);\n\t\t});\n\t}\n\n\tshowStatus(title: string, message: string): void {\n\t\tthis.setContent(frame(this.theme, title, [new Spacer(1), new Text(this.theme.fg(\"muted\", message), 1, 0)]));\n\t}\n\n\tprogress(state: ProgressState): Promise<void> {\n\t\tif (!this.progressPromise) {\n\t\t\tthis.progressPromise = new Promise((resolve) => {\n\t\t\t\tthis.progressResolver = resolve;\n\t\t\t});\n\t\t}\n\t\tthis.showingProgress = true;\n\t\tthis.updateProgress(state);\n\t\treturn this.progressPromise;\n\t}\n\n\tupdateProgress(state: ProgressState): void {\n\t\tif (!this.showingProgress) return;\n\t\tconst body = [\n\t\t\tnew Text(this.theme.fg(\"text\", state.model), 1, 0),\n\t\t\tnew Spacer(1),\n\t\t\tnew Text(this.theme.fg(\"muted\", state.message), 1, 0),\n\t\t];\n\t\tif (state.ratio !== undefined) {\n\t\t\tconst available = 40;\n\t\t\tconst filled = Math.round(Math.max(0, Math.min(1, state.ratio)) * available);\n\t\t\tbody.push(\n\t\t\t\tnew Text(\n\t\t\t\t\tthis.theme.fg(\n\t\t\t\t\t\t\"accent\",\n\t\t\t\t\t\t`${\"█\".repeat(filled)}${\"─\".repeat(available - filled)} ${Math.round(state.ratio * 100)}%`,\n\t\t\t\t\t),\n\t\t\t\t\t1,\n\t\t\t\t\t0,\n\t\t\t\t),\n\t\t\t);\n\t\t}\n\t\tif (state.detail) body.push(new Text(this.theme.fg(\"dim\", state.detail), 1, 0));\n\t\tthis.content = frame(this.theme, state.title, body, keyHint(\"tui.select.cancel\", \"stop\"));\n\t\tthis.inputHandler = undefined;\n\t\tthis.tui.requestRender();\n\t}\n\n\thandleInput(data: string): void {\n\t\tif (this.progressResolver && this.keybindings.matches(data, \"tui.select.cancel\")) {\n\t\t\tconst resolve = this.progressResolver;\n\t\t\tthis.progressPromise = undefined;\n\t\t\tthis.progressResolver = undefined;\n\t\t\tresolve();\n\t\t\treturn;\n\t\t}\n\t\tthis.inputHandler?.handleInput?.(data);\n\t\tthis.tui.requestRender();\n\t}\n\n\trender(width: number): string[] {\n\t\treturn this.content\n\t\t\t.render(width)\n\t\t\t.map((line) => (visibleWidth(line) > width ? truncateToWidth(line, width, \"\") : line));\n\t}\n\n\tinvalidate(): void {\n\t\tthis.content.invalidate();\n\t}\n}\n\nexport async function showLlamaUi(ctx: ExtensionCommandContext, run: (ui: LlamaUi) => Promise<void>): Promise<void> {\n\tawait ctx.ui.custom<void>((tui, theme, keybindings, done) => {\n\t\tconst view = new LlamaView(tui, theme, keybindings);\n\t\tvoid run(view).then(\n\t\t\t() => done(),\n\t\t\t(error: unknown) => {\n\t\t\t\tctx.ui.notify(error instanceof Error ? error.message : String(error), \"error\");\n\t\t\t\tdone();\n\t\t\t},\n\t\t);\n\t\treturn view;\n\t});\n}\n\nexport async function runWithProgress<T>(\n\tui: LlamaUi,\n\toptions: {\n\t\ttitle: string;\n\t\tmodel: string;\n\t\tinitialMessage: string;\n\t\tcancelTitle: string;\n\t\tcancelMessage: string;\n\t\trun(signal: AbortSignal, update: (progress: LlamaProgress) => void): Promise<T>;\n\t\tcancel(): Promise<void>;\n\t},\n): Promise<{ cancelled: true } | { cancelled: false; value: T }> {\n\tconst controller = new AbortController();\n\tconst state: ProgressState = { title: options.title, model: options.model, message: options.initialMessage };\n\tconst settled = options\n\t\t.run(controller.signal, (progress) => {\n\t\t\tObject.assign(state, progress);\n\t\t\tui.updateProgress(state);\n\t\t})\n\t\t.then(\n\t\t\t(value) => ({ ok: true as const, value }),\n\t\t\t(error: unknown) => ({ ok: false as const, error }),\n\t\t);\n\tlet completed = false;\n\tsettled.finally(() => {\n\t\tcompleted = true;\n\t});\n\n\twhile (!completed) {\n\t\tconst outcome = await Promise.race([\n\t\t\tsettled.then(() => \"settled\" as const),\n\t\t\tui.progress(state).then(() => \"stop\" as const),\n\t\t]);\n\t\tif (outcome === \"settled\") break;\n\t\tconst stop = await ui.confirm(options.cancelTitle, options.cancelMessage);\n\t\tif (!stop || completed) continue;\n\t\ttry {\n\t\t\tawait options.cancel();\n\t\t} finally {\n\t\t\tcontroller.abort(new Error(\"Cancelled\"));\n\t\t}\n\t\tawait settled;\n\t\treturn { cancelled: true };\n\t}\n\n\tconst result = await settled;\n\tif (!result.ok) throw result.error;\n\treturn { cancelled: false, value: result.value };\n}\n"]}
+{"version":3,"file":"ui.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/ui.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,gCAAgC,CAAC;AAK9E,OAAO,KAAK,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AACjE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAIzD,MAAM,MAAM,kBAAkB,GAAG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,KAAK,EAAE,cAAc,CAAA;CAAE,GAAG;IAAE,IAAI,EAAE,UAAU,CAAA;CAAE,GAAG;IAAE,IAAI,EAAE,OAAO,CAAA;CAAE,CAAC;AAErH,UAAU,aAAc,SAAQ,aAAa;IAC5C,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACd;AA+CD,MAAM,WAAW,OAAO;IACvB,UAAU,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAAC;IACrF,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;IACtE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC1D,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,GAAG,OAAO,CAAC,CAAC;IAChF,YAAY,CACX,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,gBAAgB,EAAE,CAAC,GACzE,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;IAC/B,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACjD,QAAQ,CAAC,KAAK,EAAE,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC9C,cAAc,CAAC,KAAK,EAAE,aAAa,GAAG,IAAI,CAAC;CAC3C;AAwYD,wBAAsB,WAAW,CAAC,GAAG,EAAE,uBAAuB,EAAE,GAAG,EAAE,CAAC,EAAE,EAAE,OAAO,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAYlH;AAED,wBAAsB,eAAe,CAAC,CAAC,EACtC,EAAE,EAAE,OAAO,EACX,OAAO,EAAE;IACR,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,GAAG,CAAC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,CAAC,QAAQ,EAAE,aAAa,KAAK,IAAI,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAChF,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACxB,GACC,OAAO,CAAC;IAAE,SAAS,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,SAAS,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,CAAC,CAAA;CAAE,CAAC,CAqC/D"}