13 lines
285 B
YAML
13 lines
285 B
YAML
|
model_name: /home/toolkit/huggingface/tldr_sft_pythia7b
|
||
|
|
train_split: train
|
||
|
|
eval_split: test
|
||
|
|
load_in_8bit: False
|
||
|
|
bf16: True
|
||
|
|
per_device_train_batch_size: 32
|
||
|
|
gradient_accumulation_steps: 1
|
||
|
|
gradient_checkpointing: True
|
||
|
|
learning_rate: 1e-4
|
||
|
|
optimizer_type: paged_adamw_32bit
|
||
|
|
lora_alpha: 32
|
||
|
|
|