xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 05:04:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix benchmark_moe.py for blockwise fp8. (#23823)

Browse Source 
Signed-off-by: crischeng <420985011@qq.com>
Co-authored-by: cris <grace@guisenbindeMacBook-Pro.local>
This commit is contained in:
YUQI.CHENG 2025-08-28 21:44:09 +08:00 committed by GitHub
parent d3da2eea54
commit 66548f6603
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
benchmarks/kernels/benchmark_moe.py
View File

@ -419,8 +419,10 @@ class BenchmarkWorker:
)
# NOTE(woosuk): The current naming convention uses w2.shape[2], which
# is the intermediate size after silu_and_mul.
block_n = block_quant_shape[0] if block_quant_shape else None
block_k = block_quant_shape[1] if block_quant_shape else None
op_config = get_moe_configs(
num_experts, shard_intermediate_size // 2, dtype_str
num_experts, shard_intermediate_size // 2, dtype_str, block_n, block_k
)
if op_config is None:
config = get_default_config(
@ -430,6 +432,7 @@ class BenchmarkWorker:
hidden_size,
topk,
dtype_str,
block_quant_shape,
)
else:
config = op_config[min(op_config.keys(), key=lambda x: abs(x - num_tokens))]