Figure 4a reports forward-plus-backward attention throughput on A100 80GB SXM4; PyTorch is marked OOM at sequence length 16k. The panel covers non-causal attention with head dimension 64.

Figure 4a reports forward-plus-backward attention throughput on A100 80GB SXM4; PyTorch is marked OOM at sequence length 16k. The panel covers non-causal attention with head dimension 64. · CiteArk

Measurement results

0/29 assessed

0 supported0 disputed or conflicting29 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Metric and conditions	Reported	Latest observation	Evidence judgment
method: PyTorch · sequenceLength: 512	36 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 512	91 TFLOPs/s	—	Not assessed
method: xFormers · sequenceLength: 512	68 TFLOPs/s	—	Not assessed
method: FlashAttention Triton · sequenceLength: 512	90 TFLOPs/s	—	Not assessed
method: FlashAttention-2 · sequenceLength: 512	132 TFLOPs/s	—	Not assessed
method: PyTorch · sequenceLength: 1k	40 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 1k	92 TFLOPs/s	—	Not assessed
method: xFormers · sequenceLength: 1k	73 TFLOPs/s	—	Not assessed
method: FlashAttention Triton · sequenceLength: 1k	102 TFLOPs/s	—	Not assessed
method: FlashAttention-2 · sequenceLength: 1k	153 TFLOPs/s	—	Not assessed
method: PyTorch · sequenceLength: 2k	43 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 2k	104 TFLOPs/s	—	Not assessed

Figure 4a reports forward-plus-backward attention throughput on A100 80GB SXM4; PyTorch is marked OOM at sequence length 16k. The panel covers non-causal attention with head dimension 64.

Reported and observed measurements

Assessments (0)

Runs (0)

Measurement results