[Perf] Optimize clamp to clamp_ (#48143)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Wentao Ye
2026-07-16 18:41:07 -04:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 2cab53ddee
commit f17be06fbe
7 changed files with 48 additions and 32 deletions
@@ -1500,7 +1500,8 @@ def build_mla_chunked_context_metadata(
chunk_ends = torch.min(
context_lens_cpu.unsqueeze(0), chunk_starts + max_context_chunk
)
chunk_seq_lens = (chunk_ends - chunk_starts).clamp(min=0)
chunk_seq_lens = chunk_ends - chunk_starts
chunk_seq_lens.clamp_(min=0)
cu_seq_lens_cpu = torch.zeros(
num_chunks, num_prefills + 1, dtype=torch.int32, pin_memory=True
@@ -1550,9 +1551,8 @@ def build_mla_chunked_context_metadata(
padded_local_context_lens_cpu.unsqueeze(0),
local_chunk_starts + padded_local_max_context_chunk,
)
padded_local_chunk_seq_lens = (local_chunk_ends - local_chunk_starts).clamp(
min=0
)
padded_local_chunk_seq_lens = local_chunk_ends - local_chunk_starts
padded_local_chunk_seq_lens.clamp_(min=0)
padded_local_cu_seq_lens_cpu = torch.zeros(
num_chunks, num_prefills + 1, dtype=torch.int32, pin_memory=True