DeepSeek/DeepGEMM
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepGEMM synced 2025-06-26 23:15:49 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Several cleanups

Browse Source
This commit is contained in:
Chenggang Zhao
2025-05-14 14:18:43 +08:00
parent 6233709c67
commit c4a7116e0a
7 changed files with 38 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/test_core.py
View File

@@ -291,7 +291,7 @@ def test_k_grouped_wgrad_gemm():
def test_func():
deep_gemm.k_grouped_wgrad_gemm_fp8_fp8_fp32_nt(x_fp8, y_fp8, out, k_sizes)
t = bench_kineto(test_func, 'fp8_wgrad_gemm', suppress_kineto_output=True, is_multiple=True) * num_groups
t = bench_kineto(test_func, 'fp8_wgrad_gemm', suppress_kineto_output=True, with_multiple_kernels=True) * num_groups
print(f' > Performance ({num_groups=}, m={m:5}, n={n:5}, avg_k={total_k//num_groups:5}): {t * 1e6:4.0f} us | '
f'throughput: {2 * num_groups * m * n * (total_k/num_groups) / t / 1e12:4.0f} TFLOPS, '
f'{(m * total_k + n * total_k + num_groups * m * n * 2) / 1e9 / t:4.0f} GB/s')