[Deprecation] Remove fallbacks for Embeddings API (#18795)

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

vllm/config.py

@@ -797,17 +797,12 @@ class ModelConfig:
         else:
             # Aliases
             if task_option == "embedding":
-                preferred_task = self._get_preferred_task(
-                    architectures, supported_tasks)
-                if preferred_task != "embed":
-                    msg = ("The 'embedding' task will be restricted to "
-                           "embedding models in a future release. Please "
-                           "pass `--task classify`, `--task score`, or "
-                           "`--task reward` explicitly for other pooling "
-                           "models.")
-                    warnings.warn(msg, DeprecationWarning, stacklevel=2)
+                msg = ("The 'embedding' task has been renamed to "
+                       "'embed', please use the new name. The old name "
+                       "will be removed in v1.0.")
+                warnings.warn(msg, DeprecationWarning, stacklevel=2)
 
-                task_option = preferred_task or "embed"
+                task_option = "embed"
 
             if task_option not in supported_tasks:
                 msg = (

vllm/entrypoints/openai/api_server.py

@@ -17,7 +17,7 @@ from contextlib import asynccontextmanager
 from functools import partial
 from http import HTTPStatus
 from json import JSONDecodeError
-from typing import Annotated, Optional, Union
+from typing import Annotated, Optional
 
 import prometheus_client
 import regex as re
@@ -59,9 +59,7 @@ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
                                               EmbeddingChatRequest,
                                               EmbeddingCompletionRequest,
                                               EmbeddingRequest,
-                                              EmbeddingResponse,
-                                              EmbeddingResponseData,
-                                              ErrorResponse,
+                                              EmbeddingResponse, ErrorResponse,
                                               LoadLoRAAdapterRequest,
                                               PoolingChatRequest,
                                               PoolingCompletionRequest,
@@ -627,37 +625,10 @@ async def create_completion(request: CompletionRequest, raw_request: Request):
 async def create_embedding(request: EmbeddingRequest, raw_request: Request):
     handler = embedding(raw_request)
     if handler is None:
-        fallback_handler = pooling(raw_request)
-        if fallback_handler is None:
-            return base(raw_request).create_error_response(
-                message="The model does not support Embeddings API")
+        return base(raw_request).create_error_response(
+            message="The model does not support Embeddings API")
 
-        logger.warning(
-            "Embeddings API will become exclusive to embedding models "
-            "in a future release. To return the hidden states directly, "
-            "use the Pooling API (`/pooling`) instead.")
-
-        res = await fallback_handler.create_pooling(request, raw_request)
-
-        generator: Union[ErrorResponse, EmbeddingResponse]
-        if isinstance(res, PoolingResponse):
-            generator = EmbeddingResponse(
-                id=res.id,
-                object=res.object,
-                created=res.created,
-                model=res.model,
-                data=[
-                    EmbeddingResponseData(
-                        index=d.index,
-                        embedding=d.data,  # type: ignore
-                    ) for d in res.data
-                ],
-                usage=res.usage,
-            )
-        else:
-            generator = res
-    else:
-        generator = await handler.create_embedding(request, raw_request)
+    generator = await handler.create_embedding(request, raw_request)
 
     if isinstance(generator, ErrorResponse):
         return JSONResponse(content=generator.model_dump(),

vllm/outputs.py

@@ -7,7 +7,7 @@ from dataclasses import dataclass
 from typing import Any, Generic, Optional, Union
 
 import torch
-from typing_extensions import TypeVar, deprecated
+from typing_extensions import TypeVar
 
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
@@ -76,14 +76,6 @@ class PoolingOutput:
         return (isinstance(other, self.__class__) and bool(
             (self.data == other.data).all()))
 
-    @property
-    @deprecated("`LLM.encode()` now stores raw outputs in the `data` "
-                "attribute. To return embeddings, use `LLM.embed()`. "
-                "To return class probabilities, use `LLM.classify()` "
-                "and access the `probs` attribute. ")
-    def embedding(self) -> list[float]:
-        return self.data.tolist()
-
 
 class RequestOutput:
     """The output data of a completion request to the LLM.
@@ -506,12 +498,6 @@ class ScoringOutput:
     def __repr__(self) -> str:
         return f"ScoringOutput(score={self.score})"
 
-    @property
-    @deprecated("`LLM.score()` now returns scalar scores. "
-                "Please access it via the `score` attribute. ")
-    def embedding(self) -> list[float]:
-        return [self.score]
-
 
 class ScoringRequestOutput(PoolingRequestOutput[ScoringOutput]):