Add safetensors export feature #3345
Open
+812
−0
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Outdated suggestions cannot be applied.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.
This is still a work in progress, I started out following antimora's indications in: #3260 (comment)
I did not find a clean way to walk the Module with the Serializer, and perform the serialization only with the Recorder. The problem is that once the Serializer is finished I'd need to reconstruct each TensorData back from its
bytes
,dtype
andshape
fields as the serializer will walk up to the basic types.The alternative I found was to use the Serializer to get a mapping from ParamId to the TensorName, and use a ModuleVisitor to link to the TensorData. I think this is a bit cleaner, but the API is pretty different from the rest of the serializations, so I'm not sure, if it would be better to do it the other way, even if the logic ends up a bit more convoluted.
Any comments are appreciated :)