Does this support training models from scratch assuming you can provide a tokenizer and a model configuration?
Does this support training models from scratch assuming you can provide a tokenizer and a model configuration?