A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 12:15 UTC
[ROCm] Remove unused IS_FNUZ parameter from reshape_and_cache_shuffle_kernel (#39123)
Signed-off-by: Bortlesboat <bortstheboat@gmail.com>
Andrew Barnes committed
Apr 7, 2026 at 03:17 UTC
5c35517a3e66740ec728a124b81029b57998d937
1 file changed
-2
vllm/v1/attention/backends/rocm_aiter_fa.py
-2
@@ -228,7 +228,6 @@ if current_platform.is_rocm():
228
num_kv_heads,
229
BLOCK_SIZE: tl.constexpr,
230
QUANT: tl.constexpr,
231
- IS_FNUZ: tl.constexpr,
231
):
232
tid = tl.program_id(0)
233
head_id = tl.program_id(1)
@@ -314,7 +313,6 @@ if current_platform.is_rocm():
313
num_kv_heads,
314
BLOCK_SIZE=head_size,
315
QUANT=QUANT,
317
- IS_FNUZ=current_platform.fp8_dtype() == torch.float8_e4m3fnuz,
316
)
317
318