Conversation
Benchmark Results
Benchmark PlotsA plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR. |
There was a problem hiding this comment.
Reactant.jl Benchmarks
Details
| Benchmark suite | Current: 601c867 | Previous: 9904590 | Ratio |
|---|---|---|---|
ViT base (256 x 256 x 3 x 32)/forward/CUDA/Reactant |
1322553433 ns |
1349734013 ns |
0.98 |
ViT base (256 x 256 x 3 x 32)/forward/CUDA/Lux |
228722429 ns |
206256598 ns |
1.11 |
ViT base (256 x 256 x 3 x 32)/forward/CPU/Reactant |
5172361198 ns |
5640616179 ns |
0.92 |
ViT base (256 x 256 x 3 x 32)/forward/CPU/Lux |
13577737011 ns |
21088566845 ns |
0.64 |
ViT small (256 x 256 x 3 x 4)/forward/CUDA/Reactant |
1282941637.5 ns |
1271785039.5 ns |
1.01 |
ViT small (256 x 256 x 3 x 4)/forward/CUDA/Lux |
8852342 ns |
8622303 ns |
1.03 |
ViT small (256 x 256 x 3 x 4)/forward/CPU/Reactant |
1620252176 ns |
1620712407 ns |
1.00 |
ViT small (256 x 256 x 3 x 4)/forward/CPU/Lux |
2243369927 ns |
2515728657 ns |
0.89 |
ViT tiny (256 x 256 x 3 x 32)/forward/CUDA/Reactant |
1326488266 ns |
1309552971.5 ns |
1.01 |
ViT tiny (256 x 256 x 3 x 32)/forward/CUDA/Lux |
91954474 ns |
88993975 ns |
1.03 |
ViT tiny (256 x 256 x 3 x 32)/forward/CPU/Reactant |
2172923706 ns |
2241028351 ns |
0.97 |
ViT tiny (256 x 256 x 3 x 32)/forward/CPU/Lux |
5386117875.5 ns |
5778081364 ns |
0.93 |
ViT tiny (256 x 256 x 3 x 4)/forward/CUDA/Reactant |
1343810206.5 ns |
1267184417.5 ns |
1.06 |
ViT tiny (256 x 256 x 3 x 4)/forward/CUDA/Lux |
7564674.5 ns |
7556327.5 ns |
1.00 |
ViT tiny (256 x 256 x 3 x 4)/forward/CPU/Reactant |
1467526138.5 ns |
1485636377.5 ns |
0.99 |
ViT tiny (256 x 256 x 3 x 4)/forward/CPU/Lux |
1579589341 ns |
1618719051 ns |
0.98 |
ViT tiny (256 x 256 x 3 x 16)/forward/CUDA/Reactant |
1305271383.5 ns |
1277921381 ns |
1.02 |
ViT tiny (256 x 256 x 3 x 16)/forward/CUDA/Lux |
11605366 ns |
11579756 ns |
1.00 |
ViT tiny (256 x 256 x 3 x 16)/forward/CPU/Reactant |
1751549749.5 ns |
1771132012 ns |
0.99 |
ViT tiny (256 x 256 x 3 x 16)/forward/CPU/Lux |
2633901708.5 ns |
2573621234.5 ns |
1.02 |
ViT small (256 x 256 x 3 x 16)/forward/CUDA/Reactant |
1304055953.5 ns |
1308609555 ns |
1.00 |
ViT small (256 x 256 x 3 x 16)/forward/CUDA/Lux |
89974859 ns |
86522099 ns |
1.04 |
ViT small (256 x 256 x 3 x 16)/forward/CPU/Reactant |
2204867166 ns |
2232948717 ns |
0.99 |
ViT small (256 x 256 x 3 x 16)/forward/CPU/Lux |
3598317281 ns |
4522528342 ns |
0.80 |
ViT small (256 x 256 x 3 x 32)/forward/CUDA/Reactant |
1322428054.5 ns |
1299910120.5 ns |
1.02 |
ViT small (256 x 256 x 3 x 32)/forward/CUDA/Lux |
122634765 ns |
107942560 ns |
1.14 |
ViT small (256 x 256 x 3 x 32)/forward/CPU/Reactant |
3087499412 ns |
3089121407 ns |
1.00 |
ViT small (256 x 256 x 3 x 32)/forward/CPU/Lux |
10528143013 ns |
12146634131 ns |
0.87 |
ViT base (256 x 256 x 3 x 16)/forward/CUDA/Reactant |
1287379469 ns |
1317150051 ns |
0.98 |
ViT base (256 x 256 x 3 x 16)/forward/CUDA/Lux |
128362123 ns |
121564537 ns |
1.06 |
ViT base (256 x 256 x 3 x 16)/forward/CPU/Reactant |
3212420542 ns |
3268410602 ns |
0.98 |
ViT base (256 x 256 x 3 x 16)/forward/CPU/Lux |
10296045155 ns |
10888829737 ns |
0.95 |
ViT base (256 x 256 x 3 x 4)/forward/CUDA/Reactant |
1353433771.5 ns |
1345840157 ns |
1.01 |
ViT base (256 x 256 x 3 x 4)/forward/CUDA/Lux |
87599937.5 ns |
78626182 ns |
1.11 |
ViT base (256 x 256 x 3 x 4)/forward/CPU/Reactant |
1885686384.5 ns |
2037797510.5 ns |
0.93 |
ViT base (256 x 256 x 3 x 4)/forward/CPU/Lux |
2450179426 ns |
2617867593 ns |
0.94 |
This comment was automatically generated by workflow using github-action-benchmark.
Val constraint on Base._cat signatureBase._cat to non-Val, typed Base._cat_t implementation
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
|
I was thinking to add some tests, but I'll wait to #161 to land. @jofrevalles this should fix your problem. can you test it please? |
I found that 0-dim arrays have a similar behavior to numbers when using |
|
@avik-pal I don't know why the shapes are wrong. It seems like it's always doing a concatenation of the first direction... but did some print debugging and It's failing on |
julia> x = fill(true)
0-dimensional Array{Bool, 0}:
1
julia> x_concrete = Reactant.to_rarray(x)
0-dimensional Array{Bool, 0}:
1I don't think it is hitting the dispatches at all |
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Fixed! But probably you want to change this Line 383 in 0520fb7 PrimitiveTypes in your PR.
|
|
Yes that PR has the patch |
|
|
|
@avik-pal it's ready now |
Base._cat to non-Val, typed Base._cat_t implementationBase._cat to non-Val, typed Base._cat_t and implement typed_hcat, typed_vcat, typed_hvcat, typed_hvncat
It's not yet finished, but this should fix #148