Skip to content

[PyT] Reduce test sizes in fused attn fp8 vs fp16 to avoid OOM #3020

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

Open
vedaanta wants to merge 4 commits into
base: main
Choose a base branch
from
+12 −10
Open

[PyT] Reduce test sizes in fused attn fp8 vs fp16 to avoid OOM #3020

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2a1cd8a
tests/attention: shrink fp8_vs_f16 configs from B=2 to B=1
vedaanta May 21, 2026
a48e7c5
address changes recommended by Kshitij
vedaanta May 21, 2026
c3f1e50
Merge branch 'main' into vedaanta/te-fp8-vs-f16-shrink-b1
vedaanta May 21, 2026
cd29763
tests/attention: black format fp8_13 ModelConfig
vedaanta May 21, 2026
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
22 changes: 12 additions & 10 deletions tests/pytorch/attention/test_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1912,36 +1912,38 @@ def get_model(dtype, config):
# test: ModelConfig(b, sq, hq, dqk)
"fp8_9": ModelConfig(
2,
4096,
2048,
128,
192,
head_dim_v=128,
),
"fp8_10": ModelConfig(
1,
4096,
2,
2048,
128,
192,
head_dim_v=128,
attn_mask_type="causal",
),
"fp8_11": ModelConfig(
2,
4096,
2048,
128,
192,
head_dim_v=128,
attn_mask_type="causal_bottom_right",
),
"fp8_12": ModelConfig(2, 8192, 32, 128, num_gqa_groups=4, attn_mask_type="causal"),
"fp8_13": ModelConfig(2, 8192, 32, 128, attn_mask_type="causal", window_size=(128, 0)),
"fp8_14": ModelConfig(2, 8192, 64, 64, num_gqa_groups=8, attn_mask_type="causal"),
"fp8_15": ModelConfig(2, 8192, 64, 64, attn_mask_type="causal", window_size=(128, 0)),
"fp8_12": ModelConfig(1, 8192, 32, 128, num_gqa_groups=4, attn_mask_type="causal"),
"fp8_13": ModelConfig(
2, 8192, 32, 128, num_gqa_groups=4, attn_mask_type="causal", window_size=(128, 0)
Copy link
Copy Markdown
Collaborator

@cyanguwa cyanguwa May 26, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is the "num_gqa_groups=4" intended here? Should we focus on just reducing the size of the config via B or S, instead of adding GQA?

),
"fp8_14": ModelConfig(2, 4096, 64, 64, num_gqa_groups=8, attn_mask_type="causal"),
"fp8_15": ModelConfig(1, 8192, 64, 64, attn_mask_type="causal", window_size=(128, 0)),
"fp8_16": ModelConfig(
2, 8192, 64, 64, num_gqa_groups=8, attn_mask_type="causal", softmax_type="learnable"
1, 8192, 64, 64, num_gqa_groups=8, attn_mask_type="causal", softmax_type="learnable"
),
"fp8_17": ModelConfig(
2, 8192, 64, 64, attn_mask_type="causal", window_size=(128, 0), softmax_type="learnable"
2, 4096, 64, 64, attn_mask_type="causal", window_size=(128, 0), softmax_type="learnable"
),
"fp8_18": ModelConfig(1, 8192, 32, 128, num_gqa_groups=4, attn_mask_type="padding"),
"fp8_19": ModelConfig(2, 2048, 16, 128, attn_mask_type="padding_causal"),
Expand Down
Loading