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 16, 02:00 UTC
[bugfix] moe tuning. rm is_navi() (#12273)
Signed-off-by: Divakar Verma <divakar.verma@amd.com>
Divakar Verma committed
Jan 21, 2025 at 16:47 UTC
2acba47d9bf97135d33355eff303d61a2c8d3d8a
1 file changed
+2
-2
benchmarks/kernels/benchmark_moe.py
+2
-2
@@ -12,10 +12,10 @@ from transformers import AutoConfig
12
13
from vllm.model_executor.layers.fused_moe.fused_moe import *
14
from vllm.platforms import current_platform
15
-from vllm.utils import FlexibleArgumentParser, is_navi
15
+from vllm.utils import FlexibleArgumentParser
16
17
FP8_DTYPE = torch.float8_e4m3fnuz if current_platform.is_rocm(
18
-) and not is_navi() else torch.float8_e4m3fn
18
+) else torch.float8_e4m3fn
19
20
21
class BenchmarkConfig(TypedDict):