Add nnx.training
#3782
Merged
Add nnx.training
#3782
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
chiamp
force-pushed
the
nnx_module_subclass
branch
from
17ec8e0
to
e3949d3
Compare
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #3782 +/- ##
==========================================
+ Coverage 60.31% 60.86% +0.55%
==========================================
Files 101 105 +4
Lines 12860 13045 +185
==========================================
+ Hits 7756 7940 +184
- Misses 5104 5105 +1 ☔ View full report in Codecov by Sentry. |
chiamp
force-pushed
the
nnx_module_subclass
branch
2 times, most recently
from
d06c963
to
297c388
Compare
cgarciae
reviewed
Mar 25, 2024
chiamp
force-pushed
the
nnx_module_subclass
branch
from
297c388
to
5b8f202
Compare
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
chiamp
force-pushed
the
nnx_module_subclass
branch
from
5b8f202
to
d538b16
Compare
cgarciae
reviewed
Mar 26, 2024
| @@ -170,18 +174,23 @@ This function takes the `state` and a data `batch` and does the following: | |||
|
|
|||
| ```{code-cell} ipython3 | |||
| @jax.jit | |||
| def train_step(state: TrainState, batch): | |||
| def train_step(state: nnx.State, static: nnx.GraphDef, batch): | |||
There was a problem hiding this comment.
maybe we could use nnx.jit here to make it simpler?
There was a problem hiding this comment.
Done.
Should we show some examples elsewhere on how to use NNX with both nnx.jit and jax.jit; in toy examples, and/or a guide with a side-by-side comparison, similar to the upgrade guides?
chiamp
force-pushed
the
nnx_module_subclass
branch
from
d538b16
to
560137b
Compare
cgarciae
force-pushed
the
nnx_module_subclass
branch
from
560137b
to
dd06693
Compare
chiamp
force-pushed
the
nnx_module_subclass
branch
from
dd06693
to
7f778be
Compare
cgarciae
force-pushed
the
nnx_module_subclass
branch
from
7f778be
to
a294f84
Compare
chiamp
force-pushed
the
nnx_module_subclass
branch
from
a294f84
to
eb8d13b
Compare
Merged
chiamp
force-pushed
the
nnx_module_subclass
branch
3 times, most recently
from
cdd1c5d
to
4726935
Compare
cgarciae
approved these changes
Apr 1, 2024
chiamp
force-pushed
the
nnx_module_subclass
branch
from
4726935
to
8852a09
Compare
chiamp
force-pushed
the
nnx_module_subclass
branch
from
8852a09
to
543686b
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add
nnx.training.API reference for
nnx.metricsandnnx.optimizer.Had to importchex.ArrayTreeto make typing happy. Is there an alternative where we don't have to use this dependency? e.g. annotating with# type: ignore.Renamed
optimizer.updatemethod tooptimizer.apply_gradientsmethod, becauseoptimizer.updatewas being called instead of the base classnnx.Module.updatemethod when usingnnx.jit.Added
OptStateandMetricStateVariablewrapper classes fornnx.optimizer.Optimizerandnnx.metrics.Metric, respectively.Updated
toy_examples/02_lifted_transforms.pyto usennx.optimizer.Optimizer. Training loop runs slightly slower.Updated
docs/experimental/nnx/mnist_tutorial.ipynbto usennx.metrics.MultiMetricandnnx.optimizer.Optimizer.For some reason, I'm getting an
nbstripouterror on CI when I try to include the outputs of the mnist tutorial, so I have removed them for now.