Skip to content

Support Context Parallel for Multi Latent Attention (MLA) #1729

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 11 commits into from
Jun 10, 2025
Merged

Support Context Parallel for Multi Latent Attention (MLA) #1729

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
c9e8bcc
Support MLA (qk_dim != v_dim) for AttnFuncWithCPAndKVP2P
yuzhongw-nvidia Apr 21, 2025
0ae3c05
add UT for MLA CP
yuzhongw-nvidia May 8, 2025
80f69e9
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 8, 2025
1ae58ef
Merge branch 'main' into mla_cp
yaox12 May 9, 2025
7ceed35
Merge branch 'main' into mla_cp
xrennvidia May 21, 2025
7a82fc8
refine the code
yuzhongw-nvidia May 22, 2025
6226f98
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] May 22, 2025
05962c3
Merge branch 'main' into mla_cp
xrennvidia Jun 5, 2025
08ba1fc
refine the code
yuzhongw-nvidia Jun 6, 2025
cf61f3d
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Jun 6, 2025
81a0c32
Merge branch 'main' into mla_cp
xrennvidia Jun 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions tests/pytorch/fused_attn/test_fused_attn_with_cp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,6 +107,18 @@ def test_cp_with_flash_attention(dtype, model, qkv_format, cp_comm_type):
"cp_2_4": ModelConfig(
2, 12, 2, 128, 4096, 4096, 0.0, "causal", "no_bias", window_size=(512, 0)
), # GQA
"cp_3_0": ModelConfig(
2, 12, 12, 128, 4096, 4096, 0.0, "causal", "no_bias", head_dim_v=64
), # MLA
"cp_3_1": ModelConfig(
2, 12, 12, 128, 4096, 4096, 0.0, "no_mask", "no_bias", head_dim_v=64
), # MLA
"cp_3_2": ModelConfig(
2, 12, 12, 128, 4096, 4096, 0.0, "causal", "post_scale_bias", head_dim_v=64
), # MLA
"cp_3_3": ModelConfig(
2, 12, 12, 128, 4096, 4096, 0.0, "no_mask", "post_scale_bias", head_dim_v=64
), # MLA
}


Expand Down Expand Up @@ -159,6 +171,8 @@ def test_cp_with_fused_attention(dtype, model, qkv_format, cp_comm_type, fp8_mha
)
if dtype != "fp8" and fp8_mha:
pytest.skip("Only fp8 works with fp8_mha=True!")
if "p2p" not in cp_comm_type and config.head_dim_qk != config.head_dim_v:
pytest.skip("MLA CP currently only support KV P2P!")

subprocess.run(
get_bash_arguments(
Expand Down
Loading