Added extra logging

apps/knowledge/tasks.py

@@ -36,7 +36,6 @@ def _extract_text_from_training_file(file_obj: TrainingFile) -> str:
     return _decode_text_bytes(raw_bytes).strip()
 
 def _get_text_chunks(text: str, size: int = 10000):
-    """Slices text into rough blocks to prevent HTTP timeouts."""
     for i in range(0, len(text), size):
         yield text[i:i + size]
 
@@ -50,8 +49,6 @@ def ingest_training_file_task(self, file_uuid):
     file_obj.status = 'ingesting'
     file_obj.save()
 
-    target_dimensions = RoleRagDocument._meta.get_field('embedding').dimensions
-
     try:
         raw_text = _extract_text_from_training_file(file_obj)
         if not raw_text:
@@ -70,7 +67,7 @@ def ingest_training_file_task(self, file_uuid):
                     json={
                         "text": text_segment,
                         "threshold": 95,
-                        "target_dimensions": target_dimensions,
+                        "target_dimensions": settings.EMBEDDING_DIMENSIONS,
                     },
                 )
                 response.raise_for_status()

gpu_server.py

@@ -77,14 +77,16 @@ async def health():
 def _resolve_target_dimensions(payload: Dict[str, Any]) -> int:
     raw_target = payload.get("target_dimensions")
     if raw_target in (None, ""):
-        raise HTTPException(status_code=400, detail="'target_dimensions' is required")
+        raise HTTPException(status_code=400, detail="'target_dimensions' is required and must be a positive integer")
 
     try:
         target = int(raw_target)
     except (TypeError, ValueError) as exc:
+        logger.warning("Invalid target_dimensions value: %s", raw_target)
         raise HTTPException(status_code=400, detail="'target_dimensions' must be an integer") from exc
 
     if target <= 0:
+        logger.warning("Non-positive target_dimensions value: %s", target)
         raise HTTPException(status_code=400, detail="'target_dimensions' must be > 0")
 
     return target
@@ -103,13 +105,18 @@ def pad_and_normalize(embeddings: Tensor, target_dimensions: int) -> Tensor:
 async def embeddings(request: Request):
     data = await request.json()
     input_data = data.get("input", "")
+    input_kind = type(input_data).__name__
+    input_count = len(input_data) if isinstance(input_data, list) else (1 if isinstance(input_data, str) else 0)
+    logger.info("/v1/embeddings request received: input_kind=%s input_count=%s", input_kind, input_count)
     target_dimensions = _resolve_target_dimensions(data)
+    logger.info("/v1/embeddings resolved target_dimensions=%s", target_dimensions)
 
     if isinstance(input_data, str):
         inputs = [input_data]
     elif isinstance(input_data, list):
         inputs = [str(item) for item in input_data if str(item).strip()]
     else:
+        logger.warning("/v1/embeddings bad input type: %s", input_kind)
         raise HTTPException(status_code=400, detail="'input' must be a string or list of strings")
 
     if not inputs:
@@ -157,16 +164,22 @@ async def semantic_chunk(request: Request):
     data = await request.json()
     raw_text = data.get("text", "")
     threshold_percentile = data.get("threshold", 95)
+    raw_text_len = len(raw_text) if isinstance(raw_text, str) else -1
+    logger.info("/v1/semantic-chunk request received: text_len=%s threshold=%s", raw_text_len, threshold_percentile,)
     target_dimensions = _resolve_target_dimensions(data)
+    logger.info("/v1/semantic-chunk resolved target_dimensions=%s", target_dimensions)
 
     if not raw_text:
+        logger.info("/v1/semantic-chunk empty text payload")
         return {"chunks": [], "embeddings": []}
 
     if len(raw_text) > 50000:
+        logger.warning("/v1/semantic-chunk payload too large: text_len=%s", len(raw_text))
         raise HTTPException(status_code=413, detail="Text block too large. Please batch on the client.")
 
     model = state.get("embed_model")
     if model is None:
+        logger.error("/v1/semantic-chunk embedding model not initialized")
         raise HTTPException(status_code=503, detail="Embedding model not initialized")
 
     sentences = [s.strip() for s in raw_text.replace('\n', ' ').split('. ') if s.strip()]