[V0 Deprecation] Remove `best_of` (#29090)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

docs/usage/v1_guide.md

@@ -169,8 +169,8 @@ As part of the major architectural rework in vLLM V1, several legacy features ha
 - **best_of**: This feature has been deprecated due to limited usage. See details at [RFC #13361](https://github.com/vllm-project/vllm/issues/13361).
 - **Per-Request Logits Processors**: In V0, users could pass custom
   processing functions to adjust logits on a per-request basis. In vLLM V1, this
-  feature has been deprecated. Instead, the design is moving toward supporting **global logits
-  processors**, a feature the team is actively working on for future releases. See details at [RFC #13360](https://github.com/vllm-project/vllm/pull/13360).
+  feature has been deprecated. Instead, we now support **global logits processors**
+  which are set at startup time, see [RFC #17799](https://github.com/vllm-project/vllm/issues/17799).
 
 ##### KV Cache features
 

tests/v1/sample/test_sampling_params_e2e.py

@@ -22,14 +22,6 @@ def test_n_gt_1(llm):
     assert len(outputs[0].outputs) == 3
 
 
-def test_best_of(llm):
-    """Raise a ValueError since best_of is deprecated."""
-
-    params = SamplingParams(n=2, best_of=3)
-    with pytest.raises(ValueError):
-        _ = llm.generate(PROMPT, params)
-
-
 def test_penalties(llm):
     """Check that we do not get errors if applied."""
 

vllm/entrypoints/openai/protocol.py

@@ -565,7 +565,6 @@ class ChatCompletionRequest(OpenAIBaseModel):
     user: str | None = None
 
     # --8<-- [start:chat-completion-sampling-params]
-    best_of: int | None = None
     use_beam_search: bool = False
     top_k: int | None = None
     min_p: float | None = None
@@ -889,7 +888,6 @@ class ChatCompletionRequest(OpenAIBaseModel):
             extra_args["kv_transfer_params"] = self.kv_transfer_params
         return SamplingParams.from_optional(
             n=self.n,
-            best_of=self.best_of,
             presence_penalty=self.presence_penalty,
             frequency_penalty=self.frequency_penalty,
             repetition_penalty=repetition_penalty,
@@ -1088,7 +1086,6 @@ class CompletionRequest(OpenAIBaseModel):
     # https://platform.openai.com/docs/api-reference/completions/create
     model: str | None = None
     prompt: list[int] | list[list[int]] | str | list[str] | None = None
-    best_of: int | None = None
     echo: bool | None = False
     frequency_penalty: float | None = 0.0
     logit_bias: dict[str, float] | None = None
@@ -1375,7 +1372,6 @@ class CompletionRequest(OpenAIBaseModel):
             extra_args["kv_transfer_params"] = self.kv_transfer_params
         return SamplingParams.from_optional(
             n=self.n,
-            best_of=self.best_of,
             presence_penalty=self.presence_penalty,
             frequency_penalty=self.frequency_penalty,
             repetition_penalty=repetition_penalty,

vllm/entrypoints/openai/serving_completion.py

@@ -250,14 +250,8 @@ class OpenAIServingCompletion(OpenAIServing):
         model_name = self.models.model_name(lora_request)
         num_prompts = len(engine_prompts)
 
-        # Similar to the OpenAI API, when n != best_of, we do not stream the
-        # results. Noting that best_of is only supported in V0. In addition,
-        # we do not stream the results when use beam search.
-        stream = (
-            request.stream
-            and (request.best_of is None or request.n == request.best_of)
-            and not request.use_beam_search
-        )
+        # We do not stream the results when using beam search.
+        stream = request.stream and not request.use_beam_search
 
         # Streaming response
         if stream:

vllm/sampling_params.py

@@ -144,12 +144,6 @@ class SamplingParams(
         are generated and streamed cumulatively per request. To see all `n`
         outputs upon completion, use `output_kind=RequestOutputKind.FINAL_ONLY`
         in `SamplingParams`."""
-    best_of: int | None = None
-    """Number of output sequences that are generated from the prompt. From
-    these `best_of` sequences, the top `n` sequences are returned. `best_of`
-    must be greater than or equal to `n`. By default, `best_of` is set to `n`.
-    Warning, this is only supported in V0."""
-    _real_n: int | None = None
     presence_penalty: float = 0.0
     """Penalizes new tokens based on whether they appear in the generated text
     so far. Values > 0 encourage the model to use new tokens, while values < 0
@@ -265,7 +259,6 @@ class SamplingParams(
     @staticmethod
     def from_optional(
         n: int | None = 1,
-        best_of: int | None = None,
         presence_penalty: float | None = 0.0,
         frequency_penalty: float | None = 0.0,
         repetition_penalty: float | None = 1.0,
@@ -315,7 +308,6 @@ class SamplingParams(
 
         return SamplingParams(
             n=1 if n is None else n,
-            best_of=best_of,
             presence_penalty=0.0 if presence_penalty is None else presence_penalty,
             frequency_penalty=0.0 if frequency_penalty is None else frequency_penalty,
             repetition_penalty=1.0
@@ -348,22 +340,6 @@ class SamplingParams(
         )
 
     def __post_init__(self) -> None:
-        # how we deal with `best_of`:
-        # if `best_of` is not set, we default to `n`;
-        # if `best_of` is set, we set `n` to `best_of`,
-        # and set `_real_n` to the original `n`.
-        # when we return the result, we will check
-        # if we need to return `n` or `_real_n` results
-        if self.best_of:
-            if self.best_of < self.n:
-                raise ValueError(
-                    f"best_of must be greater than or equal to n, "
-                    f"got n={self.n} and best_of={self.best_of}."
-                )
-            if not self._real_n:
-                self._real_n = self.n
-                self.n = self.best_of
-
         if 0 < self.temperature < _MAX_TEMP:
             logger.warning(
                 "temperature %s is less than %s, which may cause numerical "
@@ -433,18 +409,6 @@ class SamplingParams(
             raise ValueError(f"n must be an int, but is of type {type(self.n)}")
         if self.n < 1:
             raise ValueError(f"n must be at least 1, got {self.n}.")
-        if self.best_of is not None:
-            if not isinstance(self.best_of, int):
-                raise ValueError(
-                    f"best_of must be an integer, got {type(self.best_of)}"
-                )
-            if self.best_of < 1:
-                raise ValueError(f"best_of must be at least 1, got {self.best_of}")
-            if self.best_of < self.n:
-                raise ValueError(
-                    f"best_of must be greater than or equal to n, "
-                    f"got n={self.n} and best_of={self.best_of}."
-                )
         if not -2.0 <= self.presence_penalty <= 2.0:
             raise ValueError(
                 f"presence_penalty must be in [-2, 2], got {self.presence_penalty}."
@@ -519,10 +483,6 @@ class SamplingParams(
                 "stop strings are only supported when detokenize is True. "
                 "Set detokenize=True to use stop."
             )
-        if self.best_of != self._real_n and self.output_kind == (
-            RequestOutputKind.DELTA
-        ):
-            raise ValueError("best_of must equal n to use output_kind=DELTA")
 
     def _verify_greedy_sampling(self) -> None:
         if self.n > 1:

vllm/v1/engine/processor.py

@@ -142,9 +142,6 @@ class Processor:
         self,
         params: SamplingParams,
     ) -> None:
-        # Best of not yet supported.
-        if params.best_of is not None and params.best_of > 1:
-            raise ValueError("vLLM V1 does not yet support best_of.")
         # Logits processors not supported.
         if params.logits_processors:
             raise ValueError(