[Model Runner v2] Enable BGE M3 pooling embed token_classify (#50661)
Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
This commit is contained in:
@@ -18,9 +18,13 @@ from vllm.v1.worker.gpu.input_batch import InputBatch
|
||||
from vllm.v1.worker.gpu.states import RequestState
|
||||
|
||||
_SUPPORTED_TASKS: frozenset[PoolingTask] = frozenset(
|
||||
{"embed", "classify", "token_embed", "token_classify"}
|
||||
{"embed", "classify", "token_embed", "token_classify", "embed&token_classify"}
|
||||
)
|
||||
# `embed&token_classify` concatenates a fixed-size embedding with per-token
|
||||
# weights, so its output length scales with the prompt like the others here.
|
||||
_TOKEN_TASKS: frozenset[PoolingTask] = frozenset(
|
||||
{"token_embed", "token_classify", "embed&token_classify"}
|
||||
)
|
||||
_TOKEN_TASKS: frozenset[PoolingTask] = frozenset({"token_embed", "token_classify"})
|
||||
|
||||
|
||||
class PoolingRunner:
|
||||
|
||||
Reference in New Issue
Block a user