Revert "Configurable max_tokens/max_completion_tokens key (#399)"

This reverts commit 121dcdc.

Author: sjmonson

src/guidellm/backend/openai.py

@@ -31,11 +31,10 @@
 TEXT_COMPLETIONS_PATH = "/v1/completions"
 CHAT_COMPLETIONS_PATH = "/v1/chat/completions"
 
-CompletionEndpointType = Literal["text_completions", "chat_completions"]
-EndpointType = Union[Literal["models"], CompletionEndpointType]
-CHAT_COMPLETIONS: CompletionEndpointType = "chat_completions"
+EndpointType = Literal["chat_completions", "models", "text_completions"]
+CHAT_COMPLETIONS: EndpointType = "chat_completions"
 MODELS: EndpointType = "models"
-TEXT_COMPLETIONS: CompletionEndpointType = "text_completions"
+TEXT_COMPLETIONS: EndpointType = "text_completions"
 
 
 @Backend.register("openai_http")
@@ -448,7 +447,7 @@ def _extra_body(self, endpoint_type: EndpointType) -> dict[str, Any]:
 
     def _completions_payload(
         self,
-        endpoint_type: CompletionEndpointType,
+        endpoint_type: EndpointType,
         orig_kwargs: Optional[dict],
         max_output_tokens: Optional[int],
         **kwargs,
@@ -468,10 +467,8 @@ def _completions_payload(
                 self.__class__.__name__,
                 max_output_tokens or self.max_output_tokens,
             )
-            max_output_key = settings.openai.max_output_key.get(
-                endpoint_type, "max_tokens"
-            )
-            payload[max_output_key] = max_output_tokens or self.max_output_tokens
+            payload["max_tokens"] = max_output_tokens or self.max_output_tokens
+            payload["max_completion_tokens"] = payload["max_tokens"]
 
             if max_output_tokens:
                 # only set stop and ignore_eos if max_output_tokens set at request level

src/guidellm/config.py

@@ -88,10 +88,6 @@ class OpenAISettings(BaseModel):
     base_url: str = "http://localhost:8000"
     max_output_tokens: int = 16384
     verify: bool = True
-    max_output_key: dict[Literal["text_completions", "chat_completions"], str] = {
-        "text_completions": "max_tokens",
-        "chat_completions": "max_completion_tokens",
-    }
 
 
 class ReportGenerationSettings(BaseModel):

tests/unit/conftest.py

@@ -132,6 +132,7 @@ async def _mock_completions_response(request) -> AsyncIterable[str]:
             assert payload["stream_options"] == {"include_usage": True}
             assert payload["prompt"] is not None
             assert len(payload["prompt"]) > 0
+            assert payload["max_completion_tokens"] > 0
             assert payload["max_tokens"] > 0
 
             return httpx.Response(  # type: ignore
@@ -140,7 +141,7 @@ async def _mock_completions_response(request) -> AsyncIterable[str]:
                     type_="text",
                     prompt=payload["prompt"],
                     output_token_count=(
-                        payload["max_tokens"]
+                        payload["max_completion_tokens"]
                         if payload.get("ignore_eos", False)
                         else None
                     ),
@@ -161,6 +162,7 @@ async def _mock_chat_completions_response(request):
             assert payload["messages"] is not None
             assert len(payload["messages"]) > 0
             assert payload["max_completion_tokens"] > 0
+            assert payload["max_tokens"] > 0
 
             return httpx.Response(  # type: ignore
                 200,