runpod: let an OCR engine be placed per model, like everything else

OCR engines are picked by name (paddle|doctr|surya) rather than registered like
a model, so nothing could place one individually and the test for them failed
with "not in models.json" before it ever reached a pod. The only way to send
OCR to a pod was a capability remote — a production routing switch, and the
mechanism that once sent live embedding traffic to a rented GPU.

An ocr_models section maps to the ocr capability, so an engine gets the same
per-model placement every other model type already has, and the harness creates
a temporary entry for one when the catalogue has none.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0182DhR1eFNPQyGmDrpHbedY

codai/api/models_transfer.py

@@ -45,7 +45,7 @@ router = APIRouter()
 _SECTIONS = {
     "text_models", "image_models", "audio_models", "gguf_models", "tts_models",
     "vision_models", "video_models", "audio_gen_models", "embedding_models",
-    "spatial_models",
+    "spatial_models", "ocr_models",
 }
 
 

codai/api/runpod_worker.py

@@ -209,6 +209,12 @@ MODEL_TYPE_CAPABILITY = {
     "tts_models": "tts",
     "embedding_models": "embeddings",
     "spatial_models": "spatial",
+    # OCR engines are picked by name (paddle|doctr|surya) rather than registered
+    # like a model, so nothing could place one individually: a test had to move
+    # the whole capability, which is a production routing switch. An entry in
+    # this section gives an engine the same per-model placement everything else
+    # already has.
+    "ocr_models": "ocr",
     "audio_gen_models": "audio_gen",
 }
 

tools/runpod_model_test.py

@@ -102,11 +102,26 @@ def _find(data: dict, model: str):
     return None, None, None
 
 
+#: Picked by name rather than registered like a model, so there is nothing in the
+#: catalogue to pin. A temporary entry gives them the same per-model placement —
+#: the alternative was moving the whole OCR capability, which is a production
+#: routing switch.
+_OCR_ENGINES = ("paddle", "doctr", "surya")
+
+
 def _pin_to_runpod(model: str, engine: str, vram: float, usd: float,
                    image: str = "") -> bool:
     """Give this ONE model a runpod block. True when the catalogue changed."""
     data = _load_models()
     section, index, entry = _find(data, model)
+    if entry is None and model.strip().lower() in _OCR_ENGINES:
+        section = "ocr_models"
+        data.setdefault(section, [])
+        data[section].append({"path": model, "model_type": "ocr_models"})
+        index = len(data[section]) - 1
+        entry = data[section][index]
+        print(f"   {model}: OCR engine — added a temporary catalogue entry",
+              flush=True)
     if entry is None:
         print(f"   {model}: not in models.json — nothing to place", flush=True)
         return False