[codex] Harden macOS setup and MLX OCR output

backend/ocr_server.py

@@ -463,8 +463,8 @@ def extract_text_from_mlx_output(generated):
 
 
 GROUNDING_TOKEN_PATTERN = re.compile(
-    r'<\|ref\|>([^<]*)<\|/ref\|><\|det\|>\[\[([^\]]+)\]\]<\|/det\|>',
-    re.MULTILINE
+    r'<\|ref\|>(.*?)<\|/ref\|><\|det\|>\s*(\[\[.*?\]\])\s*<\|/det\|>',
+    re.DOTALL
 )
 
 
@@ -1210,7 +1210,7 @@ def get_env_flag(name, default=False):
 PDF_EVAL_MODE = get_env_flag('DEEPSEEK_OCR_PDF_EVAL_MODE', True)
 PDF_MAX_NEW_TOKENS = max(
     128,
-    int(os.environ.get('DEEPSEEK_OCR_PDF_MAX_NEW_TOKENS', '384' if PREFERRED_DEVICE_HINT == 'mps' else '1024'))
+    int(os.environ.get('DEEPSEEK_OCR_PDF_MAX_NEW_TOKENS', '1024'))
 )
 
 DEFAULT_MAX_NEW_TOKENS_MPS = max(256, int(os.environ.get('DEEPSEEK_OCR_MAX_NEW_TOKENS_MPS', '768')))

main.js

@@ -442,6 +442,7 @@ function getUvSetupEnv(paths) {
   return {
     ...process.env,
     UV_CACHE_DIR: paths.uvCacheDir,
+    UV_HTTP_TIMEOUT: process.env.UV_HTTP_TIMEOUT || '180',
     PIP_CACHE_DIR: paths.pipCacheDir
   };
 }

package.json

@@ -1,6 +1,6 @@
 {
   "name": "deepseek-ocr-client",
-  "version": "2.0.0",
+  "version": "2.0.1",
   "description": "Electron GUI for DeepSeek-OCR-2",
   "main": "main.js",
   "scripts": {

scripts/test-backend-helpers.py

@@ -0,0 +1,39 @@
+#!/usr/bin/env python3
+"""Focused regression tests for backend helper behavior."""
+
+import unittest
+from pathlib import Path
+import sys
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(PROJECT_ROOT))
+
+from backend import ocr_server
+
+
+MULTI_BOX_TOKEN_TEXT = (
+    "Before\n"
+    "<|ref|>Important Safeguards<|/ref|>"
+    "<|det|>[[78, 321, 465, 349], [532, 134, 888, 172]]<|/det|>\n"
+    "After"
+)
+
+
+class MlxOutputNormalizationTests(unittest.TestCase):
+    def test_document_output_strips_grounding_tokens_with_multiple_boxes(self):
+        cleaned, raw_tokens = ocr_server.normalize_mlx_output("document", MULTI_BOX_TOKEN_TEXT)
+
+        self.assertEqual(raw_tokens, MULTI_BOX_TOKEN_TEXT)
+        self.assertEqual(cleaned, "Before\n\nAfter")
+        self.assertNotIn("<|ref|>", cleaned)
+        self.assertNotIn("<|det|>", cleaned)
+
+    def test_ocr_output_extracts_ref_text_from_multiple_box_grounding_token(self):
+        cleaned, raw_tokens = ocr_server.normalize_mlx_output("ocr", MULTI_BOX_TOKEN_TEXT)
+
+        self.assertEqual(raw_tokens, MULTI_BOX_TOKEN_TEXT)
+        self.assertEqual(cleaned, "Important Safeguards")
+
+
+if __name__ == "__main__":
+    unittest.main()