ZIBEE/bi_150-vllm
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Upgrade to vllm 0.17.0 corex v4.1 overlay

Browse Source
This commit is contained in:
LZiBee
2026-04-29 19:38:22 +08:00
parent 8fac6062e4
commit 938d0854a5
430 changed files with 35969 additions and 14511 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/layers/fla/ops/index.py
View File

@@ -15,14 +15,12 @@ from .utils import tensor_cache
@tensor_cache
def prepare_lens(cu_seqlens: torch.LongTensor) -> torch.LongTensor:
def prepare_lens(cu_seqlens: torch.Tensor) -> torch.Tensor:
return cu_seqlens[1:] - cu_seqlens[:-1]
@tensor_cache
def prepare_chunk_indices(
cu_seqlens: torch.LongTensor, chunk_size: int
) -> torch.LongTensor:
def prepare_chunk_indices(cu_seqlens: torch.Tensor, chunk_size: int) -> torch.Tensor:
indices = torch.cat(
[
torch.arange(n)
@@ -33,9 +31,7 @@ def prepare_chunk_indices(
@tensor_cache
def prepare_chunk_offsets(
cu_seqlens: torch.LongTensor, chunk_size: int
) -> torch.LongTensor:
def prepare_chunk_offsets(cu_seqlens: torch.Tensor, chunk_size: int) -> torch.Tensor:
return torch.cat(
[cu_seqlens.new_tensor([0]), triton.cdiv(prepare_lens(cu_seqlens), chunk_size)]
).cumsum(-1)