Default to generation_config from model (vllm-project#12622)

Signed-off-by: Harry Mellor <[email protected]>
Signed-off-by: Louis Ulmer <[email protected]>

Author: hmellor

tests/entrypoints/openai/correctness/test_lmeval.py

@@ -20,7 +20,7 @@
 TASK = "gsm8k"
 FILTER = "exact_match,strict-match"
 RTOL = 0.03
-EXPECTED_VALUE = 0.58
+EXPECTED_VALUE = 0.54
 DEFAULT_ARGS = ["--max-model-len", "4096", "--disable-log-requests"]
 MORE_ARGS_LIST = [
     [],  # Default

tests/entrypoints/openai/test_serving_chat.py

@@ -38,6 +38,7 @@ class MockModelConfig:
     diff_sampling_param: Optional[dict] = None
     allowed_local_media_path: str = ""
     encoder_config = None
+    generation_config: str = "auto"
 
     def get_diff_sampling_param(self):
         return self.diff_sampling_param or {}

tests/test_config.py

@@ -289,7 +289,7 @@ def test_uses_mrope(model_id, uses_mrope):
 def test_generation_config_loading():
     model_id = "Qwen/Qwen2.5-1.5B-Instruct"
 
-    # When set generation_config to None, the default generation config
+    # When set generation_config to "vllm", the default generation config
     # will not be loaded.
     model_config = ModelConfig(model_id,
                                task="auto",
@@ -298,7 +298,7 @@ def test_generation_config_loading():
                                trust_remote_code=False,
                                seed=0,
                                dtype="float16",
-                               generation_config=None)
+                               generation_config="vllm")
     assert model_config.get_diff_sampling_param() == {}
 
     # When set generation_config to "auto", the default generation config
@@ -340,7 +340,7 @@ def test_generation_config_loading():
 
     assert model_config.get_diff_sampling_param() == override_result
 
-    # When generation_config is set to None and override_generation_config
+    # When generation_config is set to "vllm" and override_generation_config
     # is set, the override_generation_config should be used directly.
     model_config = ModelConfig(
         model_id,
@@ -350,7 +350,7 @@ def test_generation_config_loading():
         trust_remote_code=False,
         seed=0,
         dtype="float16",
-        generation_config=None,
+        generation_config="vllm",
         override_generation_config=override_generation_config)
 
     assert model_config.get_diff_sampling_param() == override_generation_config

vllm/config.py

@@ -255,7 +255,7 @@ def __init__(
         override_neuron_config: Optional[dict[str, Any]] = None,
         override_pooler_config: Optional["PoolerConfig"] = None,
         logits_processor_pattern: Optional[str] = None,
-        generation_config: Optional[str] = None,
+        generation_config: str = "auto",
         enable_sleep_mode: bool = False,
         override_generation_config: Optional[dict[str, Any]] = None,
         model_impl: Union[str, ModelImpl] = ModelImpl.AUTO,
@@ -951,7 +951,7 @@ def get_multimodal_config(self) -> "MultiModalConfig":
         return self.multimodal_config
 
     def try_get_generation_config(self) -> dict[str, Any]:
-        if self.generation_config is None or self.generation_config == "auto":
+        if self.generation_config in ("auto", "vllm"):
             config = try_get_generation_config(
                 self.hf_config_path or self.model,
                 trust_remote_code=self.trust_remote_code,
@@ -971,17 +971,14 @@ def try_get_generation_config(self) -> dict[str, Any]:
     def get_diff_sampling_param(self) -> dict[str, Any]:
         """
         This method returns a dictionary containing the parameters
-        that differ from the default sampling parameters, but only
-        if `generation_config` is set. If `generation_config` is not
-        set, an empty dictionary is returned.
+        that differ from the default sampling parameters. If
+        `generation_config` is `"vllm"`, an empty dictionary is returned.
 
         Returns:
             dict[str, Any]: A dictionary with the differing sampling
-            parameters if `generation_config` is set, otherwise an
-            empty dictionary.
+            parameters, if `generation_config` is `"vllm"` an empty dictionary.
         """
-        if self.generation_config is None:
-            # When generation_config is not set
+        if self.generation_config == "vllm":
             config = {}
         else:
             config = self.try_get_generation_config()

vllm/engine/arg_utils.py

@@ -207,7 +207,7 @@ class EngineArgs:
 
     kv_transfer_config: Optional[KVTransferConfig] = None
 
-    generation_config: Optional[str] = None
+    generation_config: Optional[str] = "auto"
     override_generation_config: Optional[Dict[str, Any]] = None
     enable_sleep_mode: bool = False
     model_impl: str = "auto"
@@ -1034,13 +1034,13 @@ def add_cli_args(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
         parser.add_argument(
             "--generation-config",
             type=nullable_str,
-            default=None,
+            default="auto",
             help="The folder path to the generation config. "
-            "Defaults to None, no generation config is loaded, vLLM defaults "
-            "will be used. If set to 'auto', the generation config will be "
-            "loaded from model path. If set to a folder path, the generation "
-            "config will be loaded from the specified folder path. If "
-            "`max_new_tokens` is specified in generation config, then "
+            "Defaults to 'auto', the generation config will be loaded from "
+            "model path. If set to 'vllm', no generation config is loaded, "
+            "vLLM defaults will be used. If set to a folder path, the "
+            "generation config will be loaded from the specified folder path. "
+            "If `max_new_tokens` is specified in generation config, then "
             "it sets a server-wide limit on the number of output tokens "
             "for all requests.")
 

vllm/entrypoints/openai/serving_chat.py

@@ -109,8 +109,10 @@ def __init__(
         self.default_sampling_params = (
             self.model_config.get_diff_sampling_param())
         if self.default_sampling_params:
-            logger.info("Overwriting default chat sampling param with: %s",
-                        self.default_sampling_params)
+            source = self.model_config.generation_config
+            source = "model" if source == "auto" else source
+            logger.info("Using default chat sampling params from %s: %s",
+                        source, self.default_sampling_params)
 
     async def create_chat_completion(
         self,

vllm/entrypoints/openai/serving_completion.py

@@ -55,9 +55,10 @@ def __init__(
         self.default_sampling_params = (
             self.model_config.get_diff_sampling_param())
         if self.default_sampling_params:
-            logger.info(
-                "Overwriting default completion sampling param with: %s",
-                self.default_sampling_params)
+            source = self.model_config.generation_config
+            source = "model" if source == "auto" else source
+            logger.info("Using default completion sampling params from %s: %s",
+                        source, self.default_sampling_params)
 
     async def create_completion(
         self,