Update Engine Logic

.github/workflows/main.yml

@@ -33,3 +33,5 @@ jobs:
 
     - name: Test with pytest
       run: pytest tests -v
+      env:
+        PYTHONPATH: ${{ github.workspace }}

src/model_engine.py

@@ -7,8 +7,13 @@
 
 import fitz
 from PIL import Image
-from mlx_lm import load
-from mlx_lm.generate import stream_generate
+try:
+    from mlx_lm import load
+    from mlx_lm.generate import stream_generate
+except (ImportError, ModuleNotFoundError):
+    # CI / non-Apple-Silicon: module still importable, tests mock these
+    load = None
+    stream_generate = None
 
 from src.scanner import UniversalResource
 
@@ -39,7 +44,7 @@ def __init__(self, model_path="google/gemma-3-4b-it"):
             self.master_template = "# S T A R R Y N O T E \n\n[Recovery Mode Active]"
 
         # Pre-clean template: strip HTML comments to reduce prompt tokens
-        self._prompt_template = self._compact_template(self.master_template)
+        self._prompt_template = self._clean_template(self.master_template)
         log.info("S T A R R Y N O T E Engine is fully operational (template: %d → %d chars).",
                  len(self.master_template), len(self._prompt_template))
 
@@ -99,7 +104,7 @@ def _stream(self, prompt, on_token=None, images=None):
         for i, response in enumerate(stream_generate(
             self.model, self.tokenizer, prompt=prompt, **kwargs
         )):
-            text = response.text
+            text += response.text  # v0.30+ yields per-segment, must accumulate
             if on_token:
                 on_token(i + 1)
 
@@ -190,7 +195,7 @@ def _analyze_pdf(self, file_path: str, on_token=None) -> str:
             for page in doc:
                 text_buffer += page.get_text() + "\n"
 
-            content = text_buffer.strip()[:8000]
+            content = text_buffer.strip()[:12000]
 
             if len(content) < 100:
                 log.info("Image-based PDF detected — initializing Vision OCR…")