Austin comments

Results 7 comments of


                                            Austin

`strict = False` does not work when the checkpoint is distributed

Maybe you can try as follows: 1.Adjust the strict parameter: Modify the code so that it sets allow_partial_load to not strict when calling torch.distributed.checkpoint.load, ensuring that partial loading is allowed...

`Trainer`'s `.init_module()` context does not initialize model on target device

By default, PyTorch creates the model on the CPU, unless you explicitly move it to a different device (e.g. via .to(device)). As a result, even with trainer.init_module(), the model will...

OneVsRestClassifier cannot be used with TunedThresholdClassifierCV

Can I work on this? @glemaitre

Deepspeed Startegy doesn't set num_checkpoints while using activation partitions

Here are some of my opinions: Causes: num_checkpoints Parameter not passed: num_checkpoints is the number of checkpoints used to activate, but is not properly passed to the relevant function during...

Unreadable font color theme of YAML files

Maybe you can change your IDE to light mode to make the background white.

local_action with mapping still documented despite producing a deprecation warning

reference : #32121

Import error on shutdown/KeyboardInterrupt if ran from Jupyter Lab notebook cell

Avoid exit(1): In a Jupyter environment, exit() can cause problems. exit is possible in standard Python scripts, but should not be called in Jupyter notebooks. You can use sys.exit() instead:...