Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Collection of minor changes #26

Open
wants to merge 20 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
129d8c5
Remove unused NANException
spinjo Nov 8, 2024
1e51fc6
Add missing bracket in logs
spinjo Nov 10, 2024
cef1b4c
Remove schedulefree optimizer (didnt help), take Lion optimizer from …
spinjo Nov 18, 2024
38a04cb
Add ADOPT optimizer (recent hype)
spinjo Nov 18, 2024
f684317
Make the code work with training.clip_grad_norm=None (turns off gradi…
spinjo Nov 19, 2024
f409c16
Updated config files to match main, removed all mentions of ScheduleF…
Nov 19, 2024
dc83381
Merge branch 'main' into cleanup3
spinjo Nov 19, 2024
cba6778
Expanded the range of allowed numpy versions in the requirements
Nov 19, 2024
0962998
Speed up embedding code: Avoid for loop over batch indices
spinjo Nov 20, 2024
1d4dd0e
Speed up embedding code part 2: Also kill the 2nd for loop (very simi…
spinjo Nov 20, 2024
6c69fcf
Speed up embedding code part 3: Change the if conditions in case we h…
spinjo Nov 20, 2024
42d8a06
Commentary on the numpy range expansion and opt_einsum latest version…
Nov 20, 2024
df3f753
Added a disclaimer about compatibility issues of xformers on Mac
Nov 22, 2024
e8ab29c
Remove mentions of equivariant join from mlp documentation
spinjo Nov 23, 2024
4dd7d25
Extend requirements to make data/collect_data.py work
spinjo Nov 25, 2024
16e2aed
Fix inconsistencies in tests: remove gated_gelu_divide from tests; ma…
spinjo Nov 25, 2024
03a3224
For amplitudes DSI, define inv_mean and inv_std with register_buffer …
spinjo Nov 27, 2024
9a7cacb
Fix the last fix - setting buffers to None does not work
spinjo Nov 27, 2024
a559a36
Fixed a bug on the amplitude plot script and made the model store the…
Nov 27, 2024
d5c0bc7
Shortened the clip_grad_norm calculation accounting for all options
Nov 28, 2024
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions config_paper/default_tagging.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -20,9 +20,13 @@ data:
train: true
training:
iterations: 2e5
optimizer: Lion
batchsize: 128
lr: 3e-4
validate_every_n_steps: 5000
weight_decay: 0 # top tagging from scratch: 0.2; JetClass pretraining/training: 0; top finetuning: 0.01
scheduler: CosineAnnealingLR


force_xformers: true

Expand Down
4 changes: 0 additions & 4 deletions experiments/base_experiment.py
Original file line number Diff line number Diff line change
Expand Up @@ -499,8 +499,6 @@ def cycle(iterable):
for step in range(self.cfg.training.iterations):
# training
self.model.train()
if self.cfg.training.optimizer == "ScheduleFree":
self.optimizer.train()
data = next(iterator)
self._step(data, step)

Expand Down Expand Up @@ -632,8 +630,6 @@ def _validate(self, step):
metrics = self._init_metrics()

self.model.eval()
if self.cfg.training.optimizer == "ScheduleFree":
self.optimizer.eval()
with torch.no_grad():
for data in self.val_loader:
# use EMA for validation if available
Expand Down
2 changes: 1 addition & 1 deletion requirements.txt
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ black==22.3.0
## Standard science stack
h5py
matplotlib
numpy<1.25
numpy==1.25
spinjo marked this conversation as resolved.
Show resolved Hide resolved
scipy

## Standard utils
Expand Down