Trainer - Single Accelerator¶
Category: trainer
Version: 1.0.0
UUID: d99e00d027d065e42ef2d0d0d611d9e0a7b9f2aa22ab7d23078aa17e4d2ab3c3
Execution context: leip_af
Choice Priority: 1000.0
Value Parameters¶
| Name | Synonyms | Type | Values | Help |
|---|---|---|---|---|
| The devices to use | trainer.devices |
choice | [0] (gpu0)[1] (gpu1)[2] (gpu2)[3] (gpu3) |
|
| Dry run | trainer.fast_dev_run |
scalar | bool | |
| Gradient clipping algorithm | trainer.gradient_clip_algorithm |
choice | normvalue |
Refer here for more details. |
| Gradient clipping value | trainer.gradient_clip_value |
scalar | float | Refer here for more details. |
| How much of training data to use | trainer.train_batches_percentage |
scalar | float | Refer (here)[https://lightning.ai/docs/pytorch/stable/common/trainer.html#limit-train-batches] |
| Log data every N steps | trainer.logging_frequency,trainer.log_every_n_steps |
scalar | int | Refer here for more details. |
| Maximum number of epochs | train.num_epochs |
scalar | int, min: 1 | |
| Maximum training time | train.max_time |
scalar | string | Time before training stops, formatted as a string "DD:HH:MM:SS". A checkpoint will be saved when training stops, even if an epoch didn't complete. Refer (here)[https://lightning.ai/docs/pytorch/stable/common/trainer.html#max-time] |
| Trainer numerical precision | trainer.precision |
choice | 321616-mixedbf16 |
Refer here for more details. |
| Frequency of validation | train.val_check_interval,train.val_check_frequency |
scalar | float | How often within one training epoch to check the validation set. Refer (here)[https://lightning.ai/docs/pytorch/stable/common/trainer.html#val-check-interval] |
This component fits into¶
| Name | UUID | Synonyms |
|---|---|---|
| Full Recipe | 4511dd... |
trainer |