mmcdermott nested_ragged_tensors issues

mmcdermott / nested_ragged_tensors

Utilities for efficiently working with, saving, and loading, collections of connected nested ragged tensors in PyTorch

MIT License

7 stars 1 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Schema may be setting keys with dimensional signfiers...

#44 mmcdermott opened 10 hours ago
0
github sigstore workflow is failing

#43 mmcdermott opened 1 week ago
0
Errors likely exist when sentinel key names are used as input raw tensor names

#42 mmcdermott opened 1 week ago
0
Hotfix 0.0.8 release for flatten at dim=1

#41 mmcdermott closed 1 week ago
2
Fixes flatten on dim=1

#40 mmcdermott closed 1 week ago
1
`flatten` does not work with `target_dim=1`

#39 mmcdermott closed 1 week ago
1
Sampling error in NRT

#38 payalchandak closed 10 hours ago
2
Documentation needs to be updated

#37 mmcdermott opened 1 week ago
0
_:hammer_and_wrench: Refactor suggestion_

#36 mmcdermott opened 1 week ago
0
Release Candidate

#35 mmcdermott closed 1 week ago
2
You should be able to easily extract from a tensor a contiguous slice of nested values, even when that slice does not correspond to a meaningful dense slice.

#34 Oufattole opened 2 weeks ago
4
Updated benchmark to have a better way to subselect sequences.

#33 mmcdermott closed 1 month ago
2
Making tensors from file paths or dictionaries share file paths for ease of development.

#32 mmcdermott closed 1 week ago
3
Benchmarking does not appear to be posting comments on commits in the correct manner

#31 mmcdermott closed 1 week ago
0
Made splitting only happen at densification.

#30 mmcdermott closed 1 month ago
4
Major revisions to the utility and generality of the benchmarking code.

#29 mmcdermott closed 1 month ago
2
Once separate methods are made into independent profiling benchmarks, set up GH actions to run them only when their code or the installed packages change versions.

#28 mmcdermott opened 1 month ago
0
Ensuring benchmark has full coverage of capabilities and usage modes of NRT.

#27 mmcdermott closed 1 week ago
0
Establishing separate, run-once code for other methods to compare.

#26 mmcdermott opened 1 month ago
0
Running the benchmark via GH Actions on new PRs to main.

#25 mmcdermott closed 1 month ago
0
Incorporating memory tracking into the benchmark code.

#24 mmcdermott closed 1 month ago
0
Converting the `performance_tests` code to a benchmark that can be run just on the NRT codebase with a static sample dataset.

#23 mmcdermott closed 1 month ago
0
Profiling should be expanded and tracked regularly.

#22 mmcdermott opened 1 month ago
0
Tensors should not be split into lists of lists until densification.

#21 mmcdermott closed 1 month ago
1
The interface should be streamlined so that tensors can either be directly realized in memory or stored on disk and the object can still do the same things.

#20 mmcdermott closed 1 week ago
0
The speed profiling code should be run and tracked in some way on pull requests via GH workflows.

#19 mmcdermott closed 1 month ago
1
Added a more graceful failure mode for the empty tensors case.

#18 mmcdermott closed 1 month ago
2
Adding test cases and fixes for dim-0 or size-0 tensors.

#17 mmcdermott closed 1 month ago
2
Updated pyproject toml for dynamic versioning and added python build workflows.

#16 mmcdermott closed 1 month ago
2
Build setup should be finalized with more recent improvements.

#15 mmcdermott closed 1 month ago
0
Add check when out_tensors[bounds_key] is empty

#14 juancq closed 1 month ago
4
IndexError when out_tensors[bounds_key] is empty

#13 juancq closed 1 month ago
0
If a set of tensors is empty, the pipeline should either do something more intelligent (such as drop that tensor) or error gracefully.

#12 mmcdermott closed 1 month ago
0
This, I think, correctly flattens the last dim during load, but needs more testing and error control

#11 mmcdermott closed 4 weeks ago
3
debugging flattening the last dimensions

#10 Oufattole closed 2 months ago
1
`load_slice` should be able to sub-splice more than just through the zeroth dimension.

#9 mmcdermott closed 1 week ago
0
`load` and `load_slice` should have options to flatten the data in the last level without padding during loading.

#8 mmcdermott closed 1 week ago
4
Updated tests

#7 mmcdermott closed 4 months ago
2
Update profiling to include `load_slice` usage.

#6 mmcdermott closed 1 week ago
0
Add test coverage checking

#5 mmcdermott closed 5 months ago
1
Add readthedocs

#4 mmcdermott opened 5 months ago
0
Unstack dim1 too

#3 mmcdermott closed 9 months ago
0
Simplify the code to only what is absolutely required (joint collections of nested ragged tensors) and further optimize

#2 mmcdermott closed 10 months ago
0
Numpy support

#1 mmcdermott closed 10 months ago
0