hfl is a single-file static configuration pipeline tool for HF.
hfl only works with common NLP tasks & is configured with PyTorch.
This is currently experimental (WIP).
This section is incomplete.
- Basic pipelines.
- Inherited checkpoints (
model&tokenizer). - Multi-sequence tokenization.
- Training on train datasets (local & AWS SageMaker).
- Deploying fine-tuned models (
push_to_hub). - Localized metrics from test sets.
- Specifying custom preprocessor scripts.
- Integration with custom training loops.
Target a global pip installation to this repository.
You can either make create_env_win_cmd or if you require to change the execution policy then make create_env_win_pow followed by the relevant load target.
This section is incomplete.