[V0 deprecation] Remove more V0 references (#29088)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
@@ -586,13 +586,11 @@ class IsHybrid(Protocol):
|
||||
def get_mamba_state_shape_from_config(
|
||||
cls,
|
||||
vllm_config: VllmConfig,
|
||||
use_v1: bool = True,
|
||||
) -> tuple[tuple[int, int], tuple[int, int, int]]:
|
||||
"""Calculate shapes for Mamba's convolutional and state caches.
|
||||
|
||||
Args:
|
||||
vllm_config: vLLM config
|
||||
use_v1: Get shapes for V1 (or V0)
|
||||
|
||||
Returns:
|
||||
Tuple containing:
|
||||
|
||||
@@ -290,7 +290,6 @@ class Plamo2MambaMixer(MambaBase, CustomOp):
|
||||
has_decode = num_decodes > 0
|
||||
num_actual_tokens = num_prefill_tokens + num_decodes
|
||||
|
||||
# NOTE: V0 put prefill before decode, v1 puts decode before prefill
|
||||
# Separate prefill and decode by splitting varlen input
|
||||
# Split along token dimension
|
||||
hidden_states_d, hidden_states_p = torch.split(
|
||||
|
||||
Reference in New Issue
Block a user