Unverified Commit 7b152a83 authored by Xiaowei Ren's avatar Xiaowei Ren Committed by GitHub
Browse files

Fix CP unit test on A100 and L40s (#1211)



skip FP8 CP tests if hardware does not support FP8
Signed-off-by: default avatarXiaowei Ren <xren@nvidia.com>
parent 8a1b7ee2
...@@ -112,6 +112,8 @@ def test_cp_with_fused_attention(dtype, model, qkv_format, cp_comm_type): ...@@ -112,6 +112,8 @@ def test_cp_with_fused_attention(dtype, model, qkv_format, cp_comm_type):
pytest.skip("THD format is only supported on sm90+!") pytest.skip("THD format is only supported on sm90+!")
if cp_comm_type == "all_gather" and get_cudnn_version() < (9, 3, 0): if cp_comm_type == "all_gather" and get_cudnn_version() < (9, 3, 0):
pytest.skip("CP implementation with KV all-gather is only supported with cuDNN >= 9.3.0!") pytest.skip("CP implementation with KV all-gather is only supported with cuDNN >= 9.3.0!")
if dtype == "fp8" and get_device_compute_capability() < (9, 0):
pytest.skip("FP8 attention is only supported on sm90+!")
config = model_configs_fused_attn[model] config = model_configs_fused_attn[model]
if qkv_format == "thd" and config.num_heads != config.num_gqa_groups: if qkv_format == "thd" and config.num_heads != config.num_gqa_groups:
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment