@tiens.nguyen/gonext-local-worker 1.0.72 → 1.0.74

package/gonext-local-worker.mjs

@@ -1061,8 +1061,10 @@ async function correctOcrText(extractedText, modelOverride = "") {
   if (override.startsWith("http://") || override.startsWith("https://")) {
     const baseUrl = /\/v1\/?$/i.test(override) ? override : `${override}/v1`;
     console.log(`[gonext-worker] OCR correction via server ${baseUrl}`);
+    console.log(`[gonext-worker] OCR correction input: ${extractedText.slice(0, 300)}`);
     try {
       const corrected = await correctOcrTextViaServer(extractedText, baseUrl);
+      console.log(`[gonext-worker] OCR correction output: ${corrected.slice(0, 300)}`);
       console.log(`[gonext-worker] OCR correction done: ${extractedText.length} → ${corrected.length} chars`);
       return corrected;
     } catch (e) {
@@ -1081,23 +1083,40 @@ async function correctOcrText(extractedText, modelOverride = "") {
     "Return only the corrected text without any explanation.\n\n" +
     `Text:\n${extractedText}`;
   console.log(`[gonext-worker] OCR correction via CLI model=${modelPath}`);
-  // Write prompt to a temp file so long text / special chars don't break --prompt arg.
+  console.log(`[gonext-worker] OCR correction input: ${extractedText.slice(0, 300)}`);
+  // Write prompt to a temp Python script that calls mlx_lm programmatically,
+  // avoiding CLI arg length/escaping limits and the missing --prompt-file flag.
+  const scriptFile = join(tmpdir(), `gonext-ocr-correct-${Date.now()}.py`);
   const promptFile = join(tmpdir(), `gonext-ocr-correct-${Date.now()}.txt`);
+  const pyScript = `import sys, mlx_lm
+model, tokenizer = mlx_lm.load(sys.argv[1])
+with open(sys.argv[2], encoding="utf-8") as f:
+    prompt = f.read()
+result = mlx_lm.generate(model, tokenizer, prompt=prompt, max_tokens=2048, verbose=False)
+print(result)
+`;
   try {
+    await writeFile(scriptFile, pyScript, "utf8");
     await writeFile(promptFile, prompt, "utf8");
-    const { stdout } = await execFile(
+    const { stdout, stderr } = await execFile(
       "python3",
-      ["-m", "mlx_lm.generate", "--model", modelPath, "--prompt-file", promptFile, "--max-tokens", "2048", "--temp", "0.0"],
+      [scriptFile, modelPath, promptFile],
       { timeout: OCR_CORRECT_TIMEOUT_MS, maxBuffer: 10 * 1024 * 1024 }
     );
+    if (stderr?.trim()) {
+      console.log(`[gonext-worker] OCR correction CLI stderr: ${stderr.trim().slice(0, 300)}`);
+    }
     const corrected = normalizeCorrection(stdout, extractedText);
+    console.log(`[gonext-worker] OCR correction output: ${corrected.slice(0, 300)}`);
     console.log(`[gonext-worker] OCR correction done: ${extractedText.length} → ${corrected.length} chars`);
     return corrected;
   } catch (e) {
+    const stderr = e && typeof e === "object" && "stderr" in e ? String(e.stderr ?? "").trim().slice(0, 300) : "";
     const msg = e instanceof Error ? e.message : String(e);
-    console.warn(`[gonext-worker] OCR correction failed (using raw OCR text): ${msg.slice(0, 200)}`);
+    console.warn(`[gonext-worker] OCR correction CLI failed (using raw OCR text): ${msg.slice(0, 200)}${stderr ? ` | stderr: ${stderr}` : ""}`);
     return extractedText;
   } finally {
+    await rm(scriptFile, { force: true }).catch(() => {});
     await rm(promptFile, { force: true }).catch(() => {});
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.72",
+  "version": "1.0.74",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",