fixed crashes due to content filter stop reason

providers/nvidia_nim.py

@@ -147,6 +147,7 @@ class NvidiaNimProvider(BaseProvider):
 
                 if choice.get("finish_reason"):
                     finish_reason = choice["finish_reason"]
+                    logger.debug(f"NIM finish_reason: {finish_reason}")
 
                 # Handle reasoning content from delta
                 reasoning = extract_reasoning_from_delta(delta)

providers/utils/sse_builder.py

@@ -17,7 +17,7 @@ STOP_REASON_MAP = {
     "stop": "end_turn",
     "length": "max_tokens",
     "tool_calls": "tool_use",
-    "content_filter": "stop_sequence",
+    "content_filter": "end_turn",
 }
 
 

server.py

@@ -26,11 +26,7 @@ from fastapi.responses import StreamingResponse, JSONResponse
 import tiktoken
 
 # Initialize tokenizer
-try:
-    ENCODER = tiktoken.get_encoding("cl100k_base")
-except Exception:
-    logger.warning("Could not load cl100k_base encoding, falling back to p50k_base")
-    ENCODER = tiktoken.get_encoding("p50k_base")
+ENCODER = tiktoken.get_encoding("cl100k_base")
 
 # Load environment variables
 load_dotenv()
@@ -39,7 +35,7 @@ load_dotenv()
 logging.basicConfig(
     level=logging.DEBUG,
     format="%(asctime)s - %(levelname)s - %(message)s",
-    handlers=[logging.FileHandler("server.log", encoding="utf-8", mode="w")],
+    handlers=[logging.FileHandler("server.log", encoding="utf-8", mode="a")],
 )
 logger = logging.getLogger(__name__)