transl8-srt 1.0.1 → 1.0.2

package/README.md

@@ -131,6 +131,23 @@ const client = new OpenAI({
 const result = await translate(content, {targetLang: "es"}, client);
 ```
 
+### Bring custom models - like Deepseek
+
+```typescript
+import OpenAI from "openai";
+import { translate } from "transl8-srt";
+
+const client = new OpenAI({
+  baseURL: "https://api.deepseek.com",
+  apiKey: process.env.DEEPSEEK_API_KEY,
+});
+
+const result = await translate(content, {
+  targetLang: "fr",
+  model: "deepseek-chat",   // ← explicitly set — no auto-selection for custom providers
+}, client);
+```
+
 ---
 
 ## Options reference

package/dist/index.cjs

@@ -7154,11 +7154,27 @@ Do not explain metaphors or idioms \u2014 translate them faithfully.`,
 Preserve the speaker's natural register and sentence rhythm.
 Do not add formality that isn't in the source.`
 };
-var NON_LATIN_SCRIPTS = /* @__PURE__ */ new Set(["ar", "am", "zh", "ja", "ko", "hi", "ru", "uk", "el"]);
-function pickModel(targetLang, override) {
-  if (override) return override;
-  const needsFullModel = NON_LATIN_SCRIPTS.has(targetLang) || ["yo", "ig", "ha", "sw", "zu"].includes(targetLang);
-  return needsFullModel ? "gpt-4o" : "gpt-4o-mini";
+var NEEDS_FULL_MODEL = /* @__PURE__ */ new Set([
+  "ar",
+  "am",
+  "zh",
+  "ja",
+  "ko",
+  "hi",
+  "ru",
+  "uk",
+  "el",
+  "yo",
+  "ig",
+  "ha",
+  "sw",
+  "zu"
+]);
+function pickModel(opts) {
+  if (opts.model) return opts.model;
+  const needsFull = NEEDS_FULL_MODEL.has(opts.targetLang);
+  if (needsFull) return "gpt-4o";
+  return opts.modelFallback ?? "gpt-4o-mini";
 }
 function buildGlossaryBlock(meta) {
   const lines = [];
@@ -7173,7 +7189,7 @@ function buildGlossaryBlock(meta) {
   }
   if (meta.glossary) {
     for (const [src, tgt] of Object.entries(meta.glossary)) {
-      lines.push(`"${src}" \u2192 "${tgt}"`);
+      lines.push(`"${src}" -> "${tgt}"`);
     }
   }
   return lines.length ? `GLOSSARY (must follow exactly):
@@ -7189,7 +7205,7 @@ function buildSystemPrompt(opts) {
     domainBlock,
     glossaryBlock,
     `RULES:
-- Preserve ALL-CAPS emphasis, ellipses (...), dashes (\u2014), and line-ending punctuation exactly.
+- Preserve ALL-CAPS emphasis, ellipses (...), dashes, and line-ending punctuation exactly.
 - Never merge or split sentences \u2014 one input item = one output item.
 - Never add explanatory text, parentheticals, or translator notes.
 - Output ONLY valid JSON. No markdown, no code fences, no preamble.`,
@@ -7304,7 +7320,7 @@ async function executeTranslation(client, sentences, opts) {
   const batchSize = opts.batchSize ?? 25;
   const overlap = opts.contextOverlap ?? 4;
   const concurrency = opts.concurrency ?? 8;
-  const model = pickModel(opts.targetLang, opts.model);
+  const model = pickModel(opts);
   const systemPrompt = buildSystemPrompt(opts);
   const payloads = buildBatches(sentences, batchSize, overlap);
   const tasks = payloads.map(
@@ -7417,12 +7433,13 @@ function emitVTT(cues) {
 // src/index.ts
 var import_promises = require("fs/promises");
 var import_path37 = require("path");
-var COST_PER_M_TOKENS = {
+var KNOWN_COSTS = {
   "gpt-4o": { input: 2.5, output: 10 },
   "gpt-4o-mini": { input: 0.15, output: 0.6 }
 };
 function estimateCost(model, inputTokens, outputTokens) {
-  const rates = COST_PER_M_TOKENS[model];
+  const rates = KNOWN_COSTS[model];
+  if (!rates) return 0;
   return inputTokens / 1e6 * rates.input + outputTokens / 1e6 * rates.output;
 }
 async function translate(fileContent, opts, client) {
@@ -7433,14 +7450,14 @@ async function translate(fileContent, opts, client) {
   const sentences = reconstructSentences(cues);
   const { translations, totalInputTokens, totalOutputTokens, totalBatches } = await executeTranslation(oi, sentences, opts);
   const translatedCues = reanchorCues(cues, sentences, translations);
-  const model = opts.model ?? (["yo", "ig", "ha", "sw", "zu", "ar", "am"].includes(opts.targetLang) ? "gpt-4o" : "gpt-4o-mini");
+  const resolvedModel = pickModel(opts);
   const stats = {
     totalCues: cues.length,
     totalSentences: sentences.length,
     totalBatches,
     inputTokens: totalInputTokens,
     outputTokens: totalOutputTokens,
-    estimatedCostUsd: estimateCost(model, totalInputTokens, totalOutputTokens),
+    estimatedCostUsd: estimateCost(resolvedModel, totalInputTokens, totalOutputTokens),
     durationMs: Date.now() - startTime
   };
   return {