fix gpt oss tool calling

src/exo/worker/engines/mlx/utils_mlx.py

@@ -316,6 +316,8 @@ def get_eos_token_ids_for_model(model_id: ModelId) -> list[int] | None:
         return [154820, 154827, 154829]
     elif "glm" in model_id_lower:
         return [151336, 151329, 151338]
+    elif "gpt-oss" in model_id_lower:
+        return [200002, 200012]
     return None
 
 

src/exo/worker/runner/runner.py

@@ -11,6 +11,7 @@ from mlx_lm.models.gpt_oss import Model as GptOssModel
 from mlx_lm.tokenizer_utils import TokenizerWrapper
 from openai_harmony import (  # pyright: ignore[reportMissingTypeStubs]
     HarmonyEncodingName,
+    HarmonyError,  # pyright: ignore[reportUnknownVariableType]
     Role,
     StreamableParser,
     load_harmony_encoding,
@@ -568,7 +569,11 @@ def parse_gpt_oss(
 
     for response in responses:
         assert isinstance(response, GenerationResponse)
-        stream.process(response.token)
+        try:
+            stream.process(response.token)
+        except HarmonyError:
+            logger.error("Encountered critical Harmony Error, returning early")
+            return
 
         delta = stream.last_content_delta
         ch = stream.current_channel