Safe serialization #116
Merged
Safe serialization #116
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
dacorvo
force-pushed
the
flattened_state_dict
branch
from
3076298 to
0267158
Compare
dacorvo
force-pushed
the
flattened_state_dict
branch
2 times, most recently
from
81c79a9 to
54b62ab
Compare
The docstring also explains why this method is required.
Now that activations are always quantized per-tensor, the shape of the input and output scales is always a scalar.
This will help serialization.
We make sure the parameters are serialized as string to be compatible with the stricter serialization formats.
dacorvo
force-pushed
the
flattened_state_dict
branch
from
54b62ab to
52a9455
Compare
This will allow to support other serializers than Pickle. This also removes the need to force assignment when loading a state_dict.
Flattened tensors meta is serialized using only strings, which are compatible with Abstract Syntax Trees evluation for converting back to python generic types. The only exception is qtype, for which we use a custom helper. This makes QModuleMixin serialization compatible with the weights_only parameter of torch.load. Note that this does NOT make QTensor compatible with safe serialization, as we don't override the pickle code used when directly saving QTensor and it uses its own representation of flattened tensors using non-primitive types.
dacorvo
force-pushed
the
flattened_state_dict
branch
from
52a9455 to
55e15b0
Compare
3 tasks
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Outdated suggestions cannot be applied.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.
This extends the
QModuleMixinclass to support two new serialization mode (in addition to the defaultPicklemode):weight_onlyserialization (fixes Incompatible with torch.load(...,weights_only=True) #93),safetensorsserialization (fixes Incompatible with safetensors serialization #100).