[TRITON][TEST]Add nightly test for triton split_qkv_rmsnorm_rope (#5267)

### What this PR does / why we need it?
Add nightly test for triton split_rmsnorm_rope

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: Angazenn <supperccell@163.com>
This commit is contained in:
Angazenn
2026-01-05 21:35:37 +08:00
committed by GitHub
parent a2daacbd71
commit 11e75494b1
2 changed files with 216 additions and 2 deletions

View File

@@ -209,8 +209,8 @@ def split_qkv_rmsnorm_rope_impl(
kv_hidden_size: int,
head_dim: int,
eps: float,
q_bias: Optional[torch.Tensor],
k_bias: Optional[torch.Tensor],
q_bias: Optional[torch.Tensor] = None,
k_bias: Optional[torch.Tensor] = None,
) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
KV_BLOCK_SIZE = triton.next_power_of_2(head_dim)
assert KV_BLOCK_SIZE == head_dim