saeeol 1.2.5 → 1.2.7

package/src/provider/local/embedder.ts

@@ -1,4 +1,4 @@
-/** 로컬 임베딩 서버 — Ollama 기반 임베딩 모델 실행 관리 */
+/** Local embedding server — Ollama-based embedding model lifecycle management */
 
 import { Effect } from "effect"
 import * as Log from "@saeeol/core/util/log"
@@ -17,7 +17,7 @@ const log = Log.create({ service: "local/embedder" })
 
 let server: EmbeddingServer | undefined
 
-// ── Ollama 상태 확인 ──
+// ── Ollama status checks ──
 
 async function isOllamaRunning(endpoint: string): Promise<boolean> {
   try {
@@ -39,7 +39,7 @@ async function isModelLoaded(endpoint: string, model: string): Promise<boolean>
   }
 }
 
-/** Ollama에 임베딩 모델이 설치되어 있는지 확인 */
+/** Check if an embedding model is installed in Ollama */
 async function isModelInstalled(endpoint: string, model: string): Promise<boolean> {
   try {
     const res = await fetch(`${endpoint}/api/tags`, { signal: AbortSignal.timeout(5000) })
@@ -53,7 +53,7 @@ async function isModelInstalled(endpoint: string, model: string): Promise<boolea
   }
 }
 
-// ── Ollama에 모델 pull ──
+// ── Pull model from Ollama ──
 
 async function pullModel(endpoint: string, model: string): Promise<void> {
   log.info("pulling embedding model", { model })
@@ -64,7 +64,7 @@ async function pullModel(endpoint: string, model: string): Promise<void> {
       method: "POST",
       headers: { "Content-Type": "application/json" },
       body: JSON.stringify({ name: ollamaModel, stream: false }),
-      signal: AbortSignal.timeout(300_000), // 5분 타임아웃
+      signal: AbortSignal.timeout(300_000), // 5min timeout
     })
     if (!res.ok) {
       const err = await res.text().catch(() => "unknown error")
@@ -77,10 +77,10 @@ async function pullModel(endpoint: string, model: string): Promise<void> {
   }
 }
 
-/** 내부 모델 ID → Ollama 모델명 매핑 */
+/** Map internal model ID to Ollama model name */
 function getOllamaModelName(modelId: string): string {
   const map: Record<string, string> = {
-    "bge-small-en": "nomic-embed-text", // Ollama에 가장 가벼운 임베딩
+    "bge-small-en": "nomic-embed-text", // lightest embedding model in Ollama
     "bge-base-en": "nomic-embed-text",
     "bge-large-en": "nomic-embed-text",
     "bge-m3": "bge-m3",
@@ -91,7 +91,7 @@ function getOllamaModelName(modelId: string): string {
   return map[modelId] ?? modelId
 }
 
-// ── 임베딩 API 호출 ──
+// ── Embedding API call ──
 
 async function embedViaOllama(endpoint: string, model: string, texts: string[]): Promise<number[][]> {
   const ollamaModel = getOllamaModelName(model)
@@ -114,7 +114,7 @@ async function embedViaOllama(endpoint: string, model: string, texts: string[]):
   return data.embeddings
 }
 
-// ── VRAM 추정 ──
+// ── VRAM estimation ──
 
 function estimateEmbeddingVRAM(modelId: string): number {
   const model = RAG.EMBEDDING_MODELS.find((m) => m.id === modelId)
@@ -122,9 +122,9 @@ function estimateEmbeddingVRAM(modelId: string): number {
   return Math.ceil(model.sizeBytes * 1.2 / (1024 * 1024))
 }
 
-// ── 공개 API ──
+// ── Public API ──
 
-/** 임베딩 서버 시작 */
+/** Start the embedding server */
 export async function start(bake: LLMBakeParams): Promise<EmbeddingServer> {
   if (server && server.status === "running") {
     log.info("embedder already running", { model: server.model })
@@ -147,7 +147,7 @@ export async function start(bake: LLMBakeParams): Promise<EmbeddingServer> {
 
   void Bus.publish(LTMEvent.EmbedderStatusChanged, { status: "starting", model: bake.embeddingModel })
 
-  // Ollama 실행 확인
+  // Check if Ollama is running
   const running = await isOllamaRunning(endpoint)
   if (!running) {
     log.warn("Ollama not running, embedding server unavailable", { endpoint })
@@ -156,7 +156,7 @@ export async function start(bake: LLMBakeParams): Promise<EmbeddingServer> {
     return server
   }
 
-  // 모델 설치 확인
+  // Check if model is installed
   const ollamaModel = getOllamaModelName(bake.embeddingModel)
   const installed = await isModelInstalled(endpoint, ollamaModel)
   if (!installed) {
@@ -164,10 +164,10 @@ export async function start(bake: LLMBakeParams): Promise<EmbeddingServer> {
     await pullModel(endpoint, ollamaModel)
   }
 
-  // 모델 로드 확인 (첫 호출 시 자동 로드됨)
+  // Check if model is loaded (auto-loaded on first call)
   const loaded = await isModelLoaded(endpoint, ollamaModel)
   if (!loaded) {
-    // 웜업 호출 — 모델을 메모리에 로드
+    // Warmup call — load model into memory
     try {
       await embedViaOllama(endpoint, bake.embeddingModel, ["warmup"])
       log.info("model warmed up", { model: ollamaModel })
@@ -186,21 +186,21 @@ export async function start(bake: LLMBakeParams): Promise<EmbeddingServer> {
   return server
 }
 
-/** 임베딩 서버 중지 */
+/** Stop the embedding server */
 export async function stop(): Promise<void> {
   if (!server) return
-  server.status = "stopped"
+  server = { ...server!, status: "stopped" }
   void Bus.publish(LTMEvent.EmbedderStatusChanged, { status: "stopped" })
   log.info("embedding server stopped")
   server = undefined
 }
 
-/** 현재 상태 */
+/** Current server status */
 export function status(): EmbeddingServer | undefined {
   return server
 }
 
-/** 텍스트 배열 → 임베딩 벡터 */
+/** Embed an array of texts → embedding vectors */
 export async function embed(texts: string[]): Promise<number[][]> {
   if (!server || server.status !== "running") {
     throw new Error("Embedding server not running")
@@ -208,13 +208,13 @@ export async function embed(texts: string[]): Promise<number[][]> {
   return embedViaOllama(server.endpoint, server.model, texts)
 }
 
-/** 단일 텍스트 → 임베딩 벡터 */
+/** Embed a single text → embedding vector */
 export async function embedOne(text: string): Promise<number[]> {
   const vectors = await embed([text])
   return vectors[0]!
 }
 
-/** VRAM 사용량 (MB) */
+/** VRAM usage in MB */
 export function vramUsage(): number {
   return server?.vramMB ?? 0
 }

package/src/provider/models-snapshot.d.ts

@@ -0,0 +1,2 @@
+// Auto-generated by build.ts - do not edit
+export declare const snapshot: Record<string, unknown>