forked from tatz1101/Edge-AI-Platform-Tutorials
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathconfig2.prototxt.relative
More file actions
27 lines (18 loc) · 797 Bytes
/
config2.prototxt.relative
File metadata and controls
27 lines (18 loc) · 797 Bytes
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
# directory where output files are saved
workspace: "INSERT_ABSOLUTE_PATH_HERE/deephi/alexnetBNnoLRN/pruning"
gpu: "0"
# the number of iterations to run in test phase.
# Max num =validation set / batch_size = 4000 / 50
test_iter: 80
# the accuracy name of the model output you care about
#acc_name: "top-5"
acc_name: "top-1"
model: "INSERT_ABSOLUTE_PATH_HERE/deephi/alexnetBNnoLRN/pruning/train_val.prototxt"
#second run
weights: "INSERT_ABSOLUTE_PATH_HERE/deephi/alexnetBNnoLRN/pruning/regular_rate_0.1/snapshots/_iter_12000.caffemodel"
# solver used for fine tuning
solver: "INSERT_ABSOLUTE_PATH_HERE/deephi/alexnetBNnoLRN/pruning/solver.prototxt"
# expected compression ratio: 0.1 means 10%
rate: 0.2
#compression method used (only REGULAR is available)
pruner { method: REGULAR }