google-deepmind / google-deepmind/optax

optax.MultiSteps out of memory

Open
#472 17 comments 0 reactions 1 assignee Claimed by @hbq1 View on GitHub
type:bug
Dominant language
Python
Stars
2.3k
Forks
369
Avg merge
10h 15m
Merged PRs (30d)
7

Description

I always get an out of memory error using optax.MultiSteps, even when every_k_schedule=1.
Using optax.apply_every(k=1) in a chain works fine.

```
optimizer = optax.chain(
optax.clip_by_global_norm(0.5),
optax.adam(lr),
#optax.apply_every(k=1)
)
optimizer = optax.MultiSteps(optimizer, every_k_schedule=1)
```
Later I'm using
`opt_state = optimizer.init(params)`
and
```
updates, opt_state = optimizer.update(grads, opt_state, params)
params = optax.apply_updates(params, updates)
```
I have no idea what I could be doing wrong. I'm not changing anything else, like batch size.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.