@intentic/iq-engine 1.16.0 → 1.17.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@intentic/iq-engine",
-  "version": "1.16.0",
+  "version": "1.17.0",
   "description": "Search engine behind the iq CLI — disk index (node:sqlite), lexical/structural/semantic/git engines, rank fusion, budget renderer",
   "license": "MIT",
   "type": "module",
@@ -47,11 +47,12 @@
     "@ast-grep/lang-python": "0.0.6",
     "@ast-grep/lang-rust": "0.0.7",
     "@ast-grep/napi": "0.44.1",
+    "@huggingface/hub": "2.13.3",
     "@huggingface/transformers": "4.2.0",
     "tslib": "2.8.1",
     "zod": "4.4.3",
-    "@intentic/sandbox-contract": "1.16.0",
-    "@intentic/workspace-ignore": "1.16.0"
+    "@intentic/sandbox-contract": "1.17.0",
+    "@intentic/workspace-ignore": "1.17.0"
   },
   "devDependencies": {
     "@types/node": "24.13.2",

package/scripts/fetch-model.mjs

@@ -1,7 +1,15 @@
 #!/usr/bin/env node
-// Downloads the iq models into <dest> at Docker build time, so `iq ask` runs fully offline: the bge embedder
-// (semantic retrieval) and the ms-marco cross-encoder (rerank stage).
-// Usage: node scripts/fetch-model.mjs /opt/iq-models
+// Bakes the iq models into <dest> at Docker build time, so `iq ask` runs fully offline: the bge embedder
+// (semantic retrieval) and the ms-marco cross-encoder (rerank stage). Downloads go through Hugging Face's Xet
+// protocol (@huggingface/hub) — the CAS bridge denies anonymous plain-HTTP fetches of these artifacts — then
+// Transformers.js re-instantiates everything to validate the exact offline layout the runtime consumes.
+// Usage: node scripts/fetch-model.mjs <dest-dir>
+import { createWriteStream } from "node:fs";
+import { mkdir } from "node:fs/promises";
+import { dirname, join } from "node:path";
+import { Readable } from "node:stream";
+import { pipeline as streamToFile } from "node:stream/promises";
+import { downloadFile } from "@huggingface/hub";
 import { AutoModelForSequenceClassification, AutoTokenizer, env, pipeline } from "@huggingface/transformers";
 
 const dest = process.argv[2];
@@ -9,9 +17,32 @@ if (dest === undefined) {
     console.error("usage: fetch-model.mjs <dest-dir>");
     process.exit(2);
 }
+
+// Same model ids + dtypes as src/embed/{embedder,reranker}.ts, pinned to exact HF revisions.
+const MODELS = {
+    "Xenova/bge-small-en-v1.5": "ea104dacec62c0de699686887e3f920caeb4f3e3",
+    "Xenova/ms-marco-MiniLM-L-6-v2": "a09144355adeed5f58c8ed011d209bf8ee5a1fec",
+};
+// Everything a q8 pipeline reads: tokenizer + config + the int8 onnx graph.
+const FILES = ["config.json", "tokenizer.json", "tokenizer_config.json", "onnx/model_quantized.onnx"];
+
+for (const [repo, revision] of Object.entries(MODELS)) {
+    for (const file of FILES) {
+        const blob = await downloadFile({ repo, path: file, revision });
+        if (blob === null) {
+            throw new Error(`${repo}@${revision} has no ${file}`);
+        }
+        const path = join(dest, repo, file);
+        await mkdir(dirname(path), { recursive: true });
+        await streamToFile(Readable.fromWeb(blob.stream()), createWriteStream(path));
+    }
+}
+
+// Validate strictly offline — instantiating pulls every artifact the runtime needs, so a gap in FILES fails
+// here with a clear local-file error instead of at first boot.
 env.cacheDir = dest;
 env.localModelPath = dest;
-// Same model ids + dtypes as src/embed/{embedder,reranker}.ts — instantiating pulls every needed artifact.
+env.allowRemoteModels = false;
 await pipeline("feature-extraction", "Xenova/bge-small-en-v1.5", { dtype: "q8" });
 await AutoTokenizer.from_pretrained("Xenova/ms-marco-MiniLM-L-6-v2");
 await AutoModelForSequenceClassification.from_pretrained("Xenova/ms-marco-MiniLM-L-6-v2", { dtype: "q8" });