17 lines
389 B
YAML
17 lines
389 B
YAML
|
model_name: /home/toolkit/huggingface/tldr_sft_pythia1b_fp16_trainall_3epochs
|
||
|
|
load_in_8bit: False
|
||
|
|
bf16: False
|
||
|
|
fp16: True
|
||
|
|
per_device_train_batch_size: 8
|
||
|
|
per_device_eval_batch_size: 8
|
||
|
|
gradient_accumulation_steps: 4
|
||
|
|
gradient_checkpointing: False
|
||
|
|
learning_rate: 2e-5
|
||
|
|
optimizer_type: adamw_torch
|
||
|
|
num_train_epochs: 3
|
||
|
|
seq_length: 560
|
||
|
|
use_lora: True
|
||
|
|
lora_all_linear: True
|
||
|
|
lora_r: 8
|
||
|
|
lora_alpha: 32
|