Forwarding broadcasted scalar metadata in Scalify tracer. #97
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
balancap
force-pushed
the
forward-broadcasted-scalar-metadata-in-scalify-tracer
branch
4 times, most recently
from
48b4b98 to
f77aa00
Compare
`scalify` interpreter/tracer is now properly tracking which tensors are just broadcasted scalars, helping then to refine the conversion rule to ScaledArray for these. In practice: it means (finally!) proper full scale propagation in MNIST training, resulting in stable training with dynamic rescale. TODO: we still need to understand why `scaled_mul` requires ScaledArray promotion to get the MNIST training example running. This requirement has been lifted in `div/add/sub` thanks to this PR.
balancap
force-pushed
the
forward-broadcasted-scalar-metadata-in-scalify-tracer
branch
from
f77aa00 to
ccc4355
Compare
balancap
deleted the
forward-broadcasted-scalar-metadata-in-scalify-tracer
branch
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Outdated suggestions cannot be applied.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.
scalifyinterpreter/tracer is now properly tracking which tensors arejust broadcasted scalars, helping then to refine the conversion rule to
ScaledArray for these.
In practice: it means (finally!) proper full scale propagation in MNIST training,
resulting in stable training with dynamic rescale.
TODO: we still need to understand why
scaled_mulrequires ScaledArray promotion toget the MNIST training example running. This requirement has been lifted in
div/add/subthanks to this PR.