Skip to content

Commit 0ede58c

Browse files
[ci skip] more benchs
1 parent 36a0252 commit 0ede58c

File tree

1 file changed

+40
-0
lines changed

1 file changed

+40
-0
lines changed

src/layers/attention.jl

Lines changed: 40 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -238,6 +238,46 @@ perf(128, 8, 128, 32)
238238
# 3.166 ms (81 allocations: 9.13 MiB)
239239
# 16.082 ms (1049 allocations: 20.58 MiB)
240240

241+
## Threadripper, NNlib v0.8.12
242+
# tullio
243+
# 5.658 ms (77 allocations: 7.25 MiB)
244+
# 22.373 ms (1124 allocations: 16.71 MiB)
245+
# nalib
246+
# 6.187 ms (89 allocations: 7.75 MiB)
247+
# 23.723 ms (604 allocations: 14.70 MiB)
248+
# nnlib
249+
# 6.473 ms (87 allocations: 9.25 MiB)
250+
# 24.966 ms (1055 allocations: 20.71 MiB)
251+
# tullio - gpu
252+
# 145.332 μs (520 allocations: 24.52 KiB)
253+
# 902.020 μs (2221 allocations: 117.19 KiB)
254+
# nalib - gpu
255+
# 162.354 μs (410 allocations: 18.03 KiB)
256+
# 604.111 μs (1263 allocations: 71.78 KiB)
257+
# nnlib - gpu
258+
# 156.383 μs (440 allocations: 20.00 KiB)
259+
# 835.374 μs (1969 allocations: 100.58 KiB)
260+
261+
## Threadripper, NNlib v0.8.13 (fast_maximum)
262+
# tullio
263+
# 4.599 ms (71 allocations: 7.13 MiB)
264+
# 20.699 ms (1118 allocations: 16.59 MiB)
265+
# nalib
266+
# 5.049 ms (84 allocations: 7.63 MiB)
267+
# 22.252 ms (599 allocations: 14.57 MiB)
268+
# nnlib
269+
# 5.378 ms (81 allocations: 9.13 MiB)
270+
# 23.453 ms (1049 allocations: 20.58 MiB)
271+
# tullio - gpu
272+
# 145.824 μs (520 allocations: 24.52 KiB)
273+
# 915.305 μs (2221 allocations: 117.19 KiB)
274+
# nalib - gpu
275+
# 164.789 μs (410 allocations: 18.03 KiB)
276+
# 610.835 μs (1263 allocations: 71.78 KiB)
277+
# nnlib - gpu
278+
# 157.785 μs (440 allocations: 20.00 KiB)
279+
# 852.087 μs (1969 allocations: 100.58 KiB)
280+
241281

242282
# function prof()
243283
# dim, len, batch_size, nheads = 128, 8, 128, 32;

0 commit comments

Comments
 (0)