QVAC-17265 feat: auto-download models in pivot and indictrans examples

packages/qvac-lib-infer-nmtcpp/CHANGELOG.md

@@ -5,6 +5,19 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.0.3] - 2026-04-15
+
+### Added
+
+- `lib/indictrans-model-fetcher.js` — downloads IndicTrans2 GGML models from the QVAC registry when not found locally
+- `@qvac/registry-client` as devDependency for IndicTrans model downloads
+- Package export for `./lib/indictrans-model-fetcher`
+
+### Changed
+
+- `examples/pivot.example.js` — auto-downloads Bergamot models (es-en, en-it) from Firefox CDN via `ensureBergamotModelFiles()`
+- `examples/indictrans.js` — auto-downloads IndicTrans model from QVAC registry via `ensureIndicTransModelFile()`
+
 ## [2.0.2] - 2026-04-14
 
 ### Fixed

packages/qvac-lib-infer-nmtcpp/examples/indictrans.js

@@ -6,7 +6,7 @@
  * This example demonstrates translation using the IndicTrans2 model
  * for English to Hindi translation (eng_Latn -> hin_Deva).
  *
- * Requires a local IndicTrans model file.
+ * The model file is downloaded automatically from the QVAC registry if not found locally.
  *
  * Usage:
  *   bare examples/indictrans.js
@@ -17,9 +17,14 @@
  */
 
 const TranslationNmtcpp = require('../index')
-const fs = require('bare-fs')
+const path = require('bare-path')
 const process = require('bare-process')
 
+const {
+  ensureIndicTransModelFile,
+  getIndicTransFileName
+} = require('../lib/indictrans-model-fetcher')
+
 // ============================================================
 // LOGGING CONFIGURATION
 // Set VERBOSE=1 environment variable to enable C++ debug logs
@@ -38,13 +43,12 @@ const logger = VERBOSE
 const text = 'How are you'
 
 async function main () {
-  const modelPath = process.env.INDICTRANS_MODEL_PATH || './models/ggml-indictrans2-en-indic-dist-200M.bin'
+  // Use local model path if provided, otherwise auto-download from QVAC registry
+  const defaultModelPath = path.join('./model/indictrans', getIndicTransFileName())
+  const modelPath = process.env.INDICTRANS_MODEL_PATH || defaultModelPath
 
-  if (!fs.existsSync(modelPath)) {
-    console.log('IndicTrans model not found at:', modelPath)
-    console.log('Set INDICTRANS_MODEL_PATH env var or place model at ./models/ggml-indictrans2-en-indic-dist-200M.bin')
-    return
-  }
+  // Ensure model file is present (downloads from QVAC registry if not)
+  await ensureIndicTransModelFile(modelPath)
 
   const model = new TranslationNmtcpp({
     files: { model: modelPath },

packages/qvac-lib-infer-nmtcpp/examples/pivot.example.js

@@ -8,7 +8,7 @@
  * - Second model: English -> Italian (en-it)
  * - Result: Spanish -> Italian translation via English pivot
  *
- * Requires local Bergamot model files for both language pairs.
+ * Model files are downloaded automatically from the Firefox CDN if not found locally.
  *
  * Usage:
  *   bare examples/pivot.example.js
@@ -19,10 +19,14 @@
  */
 
 const TranslationNmtcpp = require('../index')
-const fs = require('bare-fs')
 const path = require('bare-path')
 const process = require('bare-process')
 
+const {
+  ensureBergamotModelFiles,
+  getBergamotFileNames
+} = require('../lib/bergamot-model-fetcher')
+
 // ============================================================
 // LOGGING CONFIGURATION
 // Set VERBOSE=1 environment variable to enable C++ debug logs
@@ -54,30 +58,25 @@ async function main () {
   console.log(spanishText)
   console.log('-----------------------------------------------------------\n')
 
-  const esenPath = process.env.BERGAMOT_ESEN_PATH || './models/es-en'
-  const enitPath = process.env.BERGAMOT_ENIT_PATH || './models/en-it'
+  // Use local model paths if provided, otherwise auto-download from Firefox CDN
+  const esenPath = process.env.BERGAMOT_ESEN_PATH || './model/bergamot/esen'
+  const enitPath = process.env.BERGAMOT_ENIT_PATH || './model/bergamot/enit'
 
-  const primaryModel = path.join(esenPath, 'model.esen.intgemm.alphas.bin')
-  const primaryVocab = path.join(esenPath, 'vocab.esen.spm')
-  const pivotModel = path.join(enitPath, 'model.enit.intgemm.alphas.bin')
-  const pivotVocab = path.join(enitPath, 'vocab.enit.spm')
+  // Ensure model files are present (downloads from Firefox CDN if not)
+  const esenDir = await ensureBergamotModelFiles('es', 'en', esenPath)
+  const enitDir = await ensureBergamotModelFiles('en', 'it', enitPath)
 
-  for (const f of [primaryModel, primaryVocab, pivotModel, pivotVocab]) {
-    if (!fs.existsSync(f)) {
-      console.log('Missing model file:', f)
-      console.log('\nSet BERGAMOT_ESEN_PATH and BERGAMOT_ENIT_PATH env vars or place models in ./models/es-en and ./models/en-it')
-      return
-    }
-  }
+  const esenFiles = getBergamotFileNames('es', 'en')
+  const enitFiles = getBergamotFileNames('en', 'it')
 
   const model = new TranslationNmtcpp({
     files: {
-      model: primaryModel,
-      srcVocab: primaryVocab,
-      dstVocab: primaryVocab,
-      pivotModel,
-      pivotSrcVocab: pivotVocab,
-      pivotDstVocab: pivotVocab
+      model: path.join(esenDir, esenFiles.modelName),
+      srcVocab: path.join(esenDir, esenFiles.srcVocabName),
+      dstVocab: path.join(esenDir, esenFiles.dstVocabName),
+      pivotModel: path.join(enitDir, enitFiles.modelName),
+      pivotSrcVocab: path.join(enitDir, enitFiles.srcVocabName),
+      pivotDstVocab: path.join(enitDir, enitFiles.dstVocabName)
     },
     params: {
       srcLang: 'es',

packages/qvac-lib-infer-nmtcpp/lib/indictrans-model-fetcher.js

@@ -0,0 +1,136 @@
+'use strict'
+
+/**
+ * IndicTrans Model Fetcher
+ *
+ * Downloads IndicTrans2 GGML model files from the QVAC model registry.
+ *
+ * This module does NOT touch Bergamot or OPUS models.
+ */
+
+const fs = require('bare-fs')
+const path = require('bare-path')
+
+// ============================================================================
+// Model registry paths (from SDK models.ts)
+// ============================================================================
+
+const INDICTRANS_MODELS = {
+  'en-indic-200M-q4_0': {
+    registryPath: 'qvac_models_compiled/ggml/indictrans2/q4_0/ggml-indictrans2-en-indic-dist-200M/2026-01-01/ggml-indictrans2-en-indic-dist-200M-q4_0.bin',
+    registrySource: 's3',
+    filename: 'ggml-indictrans2-en-indic-dist-200M-q4_0.bin',
+    expectedMinSizeMB: 100
+  }
+}
+
+// ============================================================================
+// Helpers
+// ============================================================================
+
+/**
+ * Checks whether a file exists and meets minimum size requirements.
+ */
+function hasValidModelFile (filePath, minSizeMB) {
+  try {
+    const stats = fs.statSync(filePath)
+    return stats.size >= minSizeMB * 1024 * 1024
+  } catch {
+    return false
+  }
+}
+
+// ============================================================================
+// Download via QVAC Registry
+// ============================================================================
+
+/**
+ * Downloads an IndicTrans model file from the QVAC model registry.
+ */
+async function downloadIndicTransFromRegistry (modelKey, destPath) {
+  const { QVACRegistryClient } = require('@qvac/registry-client')
+
+  const modelInfo = INDICTRANS_MODELS[modelKey]
+  if (!modelInfo) {
+    throw new Error(`Unknown IndicTrans model key: ${modelKey}. Available: ${Object.keys(INDICTRANS_MODELS).join(', ')}`)
+  }
+
+  console.log(`[indictrans-fetcher] Downloading ${modelInfo.filename} from QVAC registry...`)
+
+  const client = new QVACRegistryClient()
+  await client.ready()
+
+  try {
+    const destDir = path.dirname(destPath)
+    fs.mkdirSync(destDir, { recursive: true })
+
+    const result = await client.downloadModel(
+      modelInfo.registryPath,
+      modelInfo.registrySource,
+      { outputFile: destPath }
+    )
+
+    console.log(`[indictrans-fetcher] Download complete → ${result.artifact.path}`)
+
+    if (!hasValidModelFile(destPath, modelInfo.expectedMinSizeMB)) {
+      throw new Error(`Downloaded file seems corrupted (expected >${modelInfo.expectedMinSizeMB}MB)`)
+    }
+
+    return destPath
+  } finally {
+    await client.close()
+  }
+}
+
+// ============================================================================
+// Public API
+// ============================================================================
+
+/**
+ * Ensures an IndicTrans model file is present at destPath.
+ *
+ *   1. If a valid model file already exists → returns immediately
+ *   2. Downloads from QVAC model registry
+ *
+ * @param {string} destPath  Full path where the model file should be stored
+ * @param {string} [modelKey='en-indic-200M-q4_0']  Model variant key
+ * @returns {Promise<string>} Resolved path to the model file
+ */
+async function ensureIndicTransModelFile (destPath, modelKey = 'en-indic-200M-q4_0') {
+  const modelInfo = INDICTRANS_MODELS[modelKey]
+  if (!modelInfo) {
+    throw new Error(`Unknown IndicTrans model key: ${modelKey}. Available: ${Object.keys(INDICTRANS_MODELS).join(', ')}`)
+  }
+
+  if (hasValidModelFile(destPath, modelInfo.expectedMinSizeMB)) {
+    console.log(`[indictrans-fetcher] Model already available at ${destPath}`)
+    return destPath
+  }
+
+  return downloadIndicTransFromRegistry(modelKey, destPath)
+}
+
+/**
+ * Returns the default filename for an IndicTrans model variant.
+ *
+ * @param {string} [modelKey='en-indic-200M-q4_0']  Model variant key
+ * @returns {string} Filename
+ */
+function getIndicTransFileName (modelKey = 'en-indic-200M-q4_0') {
+  const modelInfo = INDICTRANS_MODELS[modelKey]
+  if (!modelInfo) {
+    throw new Error(`Unknown IndicTrans model key: ${modelKey}. Available: ${Object.keys(INDICTRANS_MODELS).join(', ')}`)
+  }
+  return modelInfo.filename
+}
+
+// ============================================================================
+// Exports
+// ============================================================================
+
+module.exports = {
+  ensureIndicTransModelFile,
+  getIndicTransFileName,
+  downloadIndicTransFromRegistry,
+  INDICTRANS_MODELS
+}

packages/qvac-lib-infer-nmtcpp/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/translation-nmtcpp",
-  "version": "2.0.2",
+  "version": "2.0.3",
   "description": "translation addon for qvac",
   "addon": true,
   "engines": {
@@ -66,6 +66,7 @@
   "homepage": "https://github.com/tetherto/qvac/tree/main/packages/qvac-lib-infer-nmtcpp#readme",
   "devDependencies": {
     "@qvac/dl-base": "^0.1.0",
+    "@qvac/registry-client": "^0.4.0",
     "bare-fetch": "^2.5.1",
     "bare-fs": "^4.5.1",
     "bare-process": "^4.2.2",
@@ -91,7 +92,8 @@
       "default": "./addonLogging.js"
     },
     "./addonLogging.js": "./addonLogging.js",
-    "./lib/bergamot-model-fetcher": "./lib/bergamot-model-fetcher.js"
+    "./lib/bergamot-model-fetcher": "./lib/bergamot-model-fetcher.js",
+    "./lib/indictrans-model-fetcher": "./lib/indictrans-model-fetcher.js"
   },
   "types": "index.d.ts",
   "standard": {