Model
"bigscience/bloomz-560m"
Dataset
"Abirate/english_quotes" (English quotes)
Training procedure
PEFT using LoRA
lora configuration: - r=1, - lora_alpha=3, - target_modules=["query_key_value"], - lora_dropout=0.05, - bias="none", - task_type="CAUSAL_LM"
trainable params: 98,304 || all params: 559,312,896 || trainable%: 0.01757585078102687 None
training with Pytorch using the folowing parameters: - learning_rate= 1e-2 - num_train_epochs=10
Framework versions
- PEFT 0.4.0