Revert "Fix spec decoding acc length for dpsk-r1-fp4 tp8" (#12900)
This commit is contained in:
@@ -96,7 +96,7 @@ _workspace_manager = FlashInferWorkspaceManager()
|
|||||||
|
|
||||||
|
|
||||||
def ensure_workspace_initialized(
|
def ensure_workspace_initialized(
|
||||||
max_token_num: int = 2048, hidden_dim: int = 4096, use_fp32_lamport: bool = False
|
max_token_num: int = 16384, hidden_dim: int = 4096, use_fp32_lamport: bool = False
|
||||||
):
|
):
|
||||||
"""Ensure workspace is initialized"""
|
"""Ensure workspace is initialized"""
|
||||||
if not is_flashinfer_available() or _flashinfer_comm is None:
|
if not is_flashinfer_available() or _flashinfer_comm is None:
|
||||||
@@ -128,7 +128,7 @@ def flashinfer_allreduce_residual_rmsnorm(
|
|||||||
residual: torch.Tensor,
|
residual: torch.Tensor,
|
||||||
weight: torch.Tensor,
|
weight: torch.Tensor,
|
||||||
eps: float = 1e-6,
|
eps: float = 1e-6,
|
||||||
max_token_num: int = 2048,
|
max_token_num: int = 16384,
|
||||||
use_oneshot: Optional[bool] = None,
|
use_oneshot: Optional[bool] = None,
|
||||||
trigger_completion_at_end: bool = False,
|
trigger_completion_at_end: bool = False,
|
||||||
fp32_acc: bool = False,
|
fp32_acc: bool = False,
|
||||||
|
|||||||
Reference in New Issue
Block a user