Audio sinusoidal position embedding with conv projection
Audio encoder's sinusoidal position embedding generation followed by 3-stage Conv2d downsampling (8x total reduction), GELU activations, and linear projection to model dimension. Processes mel-spectrogram through conv layers, adds cached sinusoidal position embeddings, and returns padded output.
Reported evidence · last observed 2026-07-24. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyone = 1 · d_model = 1024 · kernel_size = 3 · conv_out_dim = 3840 · num_mel_bins = 80 · freq_dim_after_conv = 10 · max_source_positions = 1500 · downsample_hidden_size = 38416 cases
Not measured on H100 for this workload. Challenges →
1337.4µs81% SOL1.00×Reported · license unknown · no source2026-07-24
Measured exactly what you asked. Reported by source; not independently reproduced.
2389.7µs78% SOL1.15×Reported · license unknown · no source2026-08-03
Measured exactly what you asked. Reported by source; not independently reproduced.
3399.8µs77% SOL1.18×Reported · license unknown · no source2026-07-17
Measured exactly what you asked. Reported by source; not independently reproduced.
4421.2µs76% SOL1.25×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
5503.9µs71% SOL1.49×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
6515.7µs71% SOL1.53×Reported · license unknown · no source2026-08-17
Measured exactly what you asked. Reported by source; not independently reproduced.
7523.4µs70% SOL1.55×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
8524.3µs70% SOL1.55×Reported · license unknown · no source2026-08-16
Measured exactly what you asked. Reported by source; not independently reproduced.
9536.2µs70% SOL1.59×Reported · license unknown · no source2026-07-14
Measured exactly what you asked. Reported by source; not independently reproduced.
10578.6µs68% SOL1.71×Reported · license unknown · no source2026-07-11
Measured exactly what you asked. Reported by source; not independently reproduced.
11625.1µs66% SOL1.85×Reported · license unknown · no source2026-08-14
Measured exactly what you asked. Reported by source; not independently reproduced.
12747.6µs61% SOL2.22×Reported · license unknown · no source2026-07-21
Measured exactly what you asked. Reported by source; not independently reproduced.
13754.9µs61% SOL2.24×Reported · license unknown · no source2026-07-21
Measured exactly what you asked. Reported by source; not independently reproduced.
14755.3µs61% SOL2.24×Reported · license unknown · no source2026-07-21
Measured exactly what you asked. Reported by source; not independently reproduced.
15787.0µs60% SOL2.33×Reported · license unknown · no source2026-07-21
Measured exactly what you asked. Reported by source; not independently reproduced.
16822.8µs58% SOL2.44×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
171.13ms50% SOL3.36×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
181.16ms49% SOL3.43×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.