[bugfix] bge-m3-sparse-plugin mismatch requests (#48112)

Signed-off-by: augusto.yjh <augusto.yjh@antgroup.com>
This commit is contained in:
Augusto Yao
2026-07-10 10:03:00 +08:00
committed by GitHub
parent a0f6d767e4
commit feb384ada2
@@ -3,9 +3,8 @@
from collections.abc import Sequence from collections.abc import Sequence
from vllm.config import ModelConfig, PoolerConfig, VllmConfig from vllm.config import PoolerConfig, VllmConfig
from vllm.entrypoints.openai.engine.protocol import UsageInfo from vllm.entrypoints.openai.engine.protocol import UsageInfo
from vllm.entrypoints.pooling.base.protocol import EmbedRequestMixin
from vllm.inputs import PromptType from vllm.inputs import PromptType
from vllm.outputs import PoolingRequestOutput from vllm.outputs import PoolingRequestOutput
from vllm.plugins.io_processors.interface import IOProcessor from vllm.plugins.io_processors.interface import IOProcessor
@@ -14,7 +13,6 @@ from vllm.renderers import BaseRenderer
from vllm.tokenizers.detokenizer_utils import convert_ids_list_to_tokens from vllm.tokenizers.detokenizer_utils import convert_ids_list_to_tokens
from .types import ( from .types import (
EMBED_TASKS,
SparseEmbeddingCompletionRequestMixin, SparseEmbeddingCompletionRequestMixin,
SparseEmbeddingResponse, SparseEmbeddingResponse,
SparseEmbeddingResponseData, SparseEmbeddingResponseData,
@@ -38,7 +36,6 @@ class BgeM3SparseEmbeddingsProcessor(
continue continue
self.default_pooling_params[param] = getattr(pooler_config, param) self.default_pooling_params[param] = getattr(pooler_config, param)
self.embed_dimensions = vllm_config.model_config.embedding_size self.embed_dimensions = vllm_config.model_config.embedding_size
self.embed_request_queue: list[EmbedRequestMixin] = []
def __repr__(self) -> str: def __repr__(self) -> str:
return ( return (
@@ -56,44 +53,9 @@ class BgeM3SparseEmbeddingsProcessor(
# refer to PoolingCompletionRequest.to_pooling_params # refer to PoolingCompletionRequest.to_pooling_params
# set and verify pooling params # set and verify pooling params
params.skip_reading_prefix_cache = True params.skip_reading_prefix_cache = True
raw_embed_request = self.embed_request_queue.pop(0)
if raw_embed_request.embed_task not in EMBED_TASKS:
raise ValueError(
f"Unsupported task {raw_embed_request}, "
f"Supported tasks are {EMBED_TASKS}"
)
params.task = "embed&token_classify" params.task = "embed&token_classify"
params.use_activation = raw_embed_request.use_activation params.use_activation = True
if params.use_activation is None: params.dimensions = self.embed_dimensions
params.use_activation = True
params.dimensions = raw_embed_request.dimensions
model_config: ModelConfig = self.vllm_config.model_config
for param in self.default_pooling_params:
if getattr(params, param, None) is None:
setattr(params, param, self.default_pooling_params[param])
if params.dimensions is not None:
if not model_config.is_matryoshka:
raise ValueError(
f'Model "{model_config.served_model_name}" does not '
f"support matryoshka representation, "
f"changing output dimensions will lead to poor results."
)
mds = model_config.matryoshka_dimensions
if mds is not None:
if params.dimensions not in mds:
raise ValueError(
f"Model {model_config.served_model_name!r} "
f"only supports {str(mds)} matryoshka dimensions, "
f"use other output dimensions will "
f"lead to poor results."
)
elif params.dimensions < 1:
raise ValueError("Dimensions must be greater than 0")
return params return params
def parse_request( def parse_request(
@@ -113,10 +75,8 @@ class BgeM3SparseEmbeddingsProcessor(
if request_id is not None: if request_id is not None:
assert request_id not in self.online_requests, "request_id duplicated" assert request_id not in self.online_requests, "request_id duplicated"
self.online_requests[request_id] = prompt self.online_requests[request_id] = prompt
self.embed_request_queue.extend(prompt.to_embed_requests_online())
else: else:
self.offline_requests.append(prompt) self.offline_requests.append(prompt)
self.embed_request_queue.extend(prompt.to_embed_requests_offline())
return prompt.input return prompt.input
def _get_sparse_embedding_request(self, request_id: str | None = None): def _get_sparse_embedding_request(self, request_id: str | None = None):
@@ -157,11 +117,7 @@ class BgeM3SparseEmbeddingsProcessor(
raw_request = self._get_sparse_embedding_request(request_id) raw_request = self._get_sparse_embedding_request(request_id)
has_dense_embed = raw_request.embed_task in ["dense", "dense&sparse"] has_dense_embed = raw_request.embed_task in ["dense", "dense&sparse"]
has_sparse_embed = raw_request.embed_task in ["sparse", "dense&sparse"] has_sparse_embed = raw_request.embed_task in ["sparse", "dense&sparse"]
embed_dimensions = ( embed_dimensions = self.embed_dimensions
self.embed_dimensions
if raw_request.dimensions is None
else raw_request.dimensions
)
for idx in range(len(model_output)): for idx in range(len(model_output)):
mo = model_output[idx] mo = model_output[idx]
sparse_embedding_dict: dict[int, float] = {} sparse_embedding_dict: dict[int, float] = {}