neuro-ml thunder issues

neuro-ml / thunder

🌩️ The Deep Learning framework based on Lightning

https://neuro-ml.github.io/thunder/

MIT License

9 stars 1 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Dev

#89 arseniybelkov opened 4 months ago
0
Support for pydantic<3.0

#88 arseniybelkov opened 5 months ago
0
`thunder backend list` raises exception when executed with no added backends.

#87 arseniybelkov opened 5 months ago
0
Checkpoint loading outside of experiment is inconvinent now. We need `load_checkpoint` to fetch all the dependencies from the config.

#86 arseniybelkov opened 5 months ago
1
Callbacks for configs

#85 arseniybelkov closed 5 months ago
1
Fix `last_checkpoint`

#84 kirpoly closed 6 months ago
2
`group_metrics` consume too much memory

#83 arseniybelkov closed 5 months ago
0
Dev

#82 arseniybelkov closed 5 months ago
1
fixed last checkpoint loading

#81 shevtsovalexey closed 7 months ago
0
`last_checkpoint` doesn't work with symlinks

#80 arseniybelkov closed 5 months ago
0
accumalation of gradient (at least with 4, 8) leads to Nones in time stamps in TimeProfiler

#79 arseniybelkov opened 7 months ago
0
AMP during inference

#78 arseniybelkov opened 8 months ago
0
Customizable logging in `MetricMonitor`

#77 arseniybelkov opened 9 months ago
0
`ThunderModule` does not convert targets to the same type as outputs of the model.

#76 arseniybelkov opened 9 months ago
0
Make convenient `fix_seed` function for experiments.

#75 arseniybelkov closed 5 months ago
1
breaking changes incoming

#74 arseniybelkov closed 7 months ago
1
`MetricMonitor` api changes and fixes

#73 arseniybelkov closed 9 months ago
1
Learning Rate schedulers are not restored from checkpoint

#72 arseniybelkov closed 6 months ago
0
Default backends `cli` and `slurm` are not recognized by `thunder backend set`

#71 arseniybelkov opened 10 months ago
0
Named dataloaders for MetricLogger

#70 arseniybelkov opened 10 months ago
0
CLI speedup

#69 arseniybelkov opened 10 months ago
0
preprocessing for group metrics works only on `all_predictions` which is inconvenient

#68 arseniybelkov closed 6 months ago
0
add prediction as part of `thunder run`

#67 arseniybelkov closed 11 months ago
0
`thunder run NAME` does not fail when NAME does not exist

#66 arseniybelkov closed 5 months ago
0
`TimeProfiler` and `MetricLogger` don't save and load state after failed training.

#65 arseniybelkov closed 6 months ago
0
Dev

#64 arseniybelkov closed 11 months ago
1
add `FixedLayout` layout

#63 arseniybelkov closed 11 months ago
0
SingleSplit splits 255 cases with in relation 0.7 and 0.3 into 254 total cases.

#62 arseniybelkov closed 11 months ago
0
WandbLogger can't find project while deleting duplicate runs.

#61 arseniybelkov opened 1 year ago
1
test js

#60 arseniybelkov closed 1 year ago
1
WandbLogger creates new run after experiment was stopped.

#59 arseniybelkov closed 7 months ago
1
add support for other configs except lazycon

#58 arseniybelkov opened 1 year ago
0
add tests for wandb

#57 arseniybelkov opened 1 year ago
0
dev

#56 arseniybelkov closed 1 year ago
1
Dev

#55 arseniybelkov closed 1 year ago
1
fix (#53)

#54 arseniybelkov closed 1 year ago
1
`WandbLogger` spams experiments

#53 arseniybelkov closed 1 year ago
0
Asynchronous validation

#52 arseniybelkov opened 1 year ago
0
Add framework overview to docs.

#51 arseniybelkov opened 1 year ago
0
cli-features

#50 arseniybelkov closed 1 year ago
1
`MetricLogger` can't cast tensors to other devices.

#49 arseniybelkov closed 1 year ago
1
`MetricLogger` fails if outputs are None.

#48 arseniybelkov opened 1 year ago
0
Add flag for overwriting experiment directory for `thunder build`.

#47 arseniybelkov closed 1 year ago
0
TimeProfiler fixes

#46 arseniybelkov closed 1 year ago
1
`avg val downtime` can sometimes be negative during first epoch.

#45 arseniybelkov opened 1 year ago
0
Remove batch_size dependency from `TimeProfiler`.

#44 arseniybelkov closed 1 year ago
0
Subtract `optimizer step` and `backward` from `avg train downtime` in TimeProfiler.

#43 arseniybelkov closed 1 year ago
1
Make preprocessing outputs shareable.

#42 arseniybelkov opened 1 year ago
0
pydantic>2.0 causes `<Result AttributeError("'FieldInfo' object has no attribute 'outer_type_'")>.exit_code`

#41 arseniybelkov closed 5 months ago
1
Add flag for setting default backend

#40 arseniybelkov closed 1 year ago
0