-
Notifications
You must be signed in to change notification settings - Fork 2
/
baselines.condor
63 lines (52 loc) · 1.38 KB
/
baselines.condor
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
# run pytorch hogwild instances
Executable = /home/josers2/apa/pytorch.examples/main.py
# Executable = /home/josers2/pytorch.examples/mnist_hogwild/main.py
# InitialDir = /scratch/hogwild/mnist_hogwild
runname = $(optim)_bs$(batch_size)_base
Output = /shared/jose/pytorch/$(runname).out
Error = /shared/jose/pytorch/$(runname).err
Log = /shared/jose/pytorch/$(runname).log
stream_output = True
stream_error = True
# bahamut's gpu is ridiculously slow??
requirements = (machine == "weapon.cs.illinois.edu")
priority = 20
request_GPUs = 1
numCores = $(numWorkers) * 6 + 3
numWorkers = 1
request_CPUs = $INT(numCores)
should_transfer_files = NO
Arguments = "--lr 0.1 \
--batch-size $(batch_size) \
--tmp-dir /scratch/jose/apa_runs \
--final-dir /shared/jose/apa_res \
--attack-checkpoint-path $(runname) \
--max-steps $(max_steps) \
--num-processes 1 \
--optimizer $(optim) \
$(runname) \
baseline"
batch_size = 128
Queue 1 optim max_steps from (
sgd 350
adam 250
rms 350
)
batch_size = 64
Queue 1 optim max_steps from (
sgd 350
adam 250
rms 350
)
batch_size = 32
Queue 1 optim max_steps from (
sgd 350
adam 250
rms 350
)
batch_size = 16
Queue 1 optim max_steps from (
sgd 350
adam 250
rms 350
)