p12475398
/
LLaMA-Factory-Mirror
forked from p04798526/LLaMA-Factory-Mirror
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

train: change yaml

This commit is contained in:
wql 2024-08-13 13:19:42 +08:00
parent 6f7bca808a
commit beb97a099c
1 changed files with 14 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
examples/train_lora/llama2_lora_sft.yaml
View File

@ -5,17 +5,18 @@ model_name_or_path: modelscope/Llama-2-7b-ms
stage: sft stage: sft
do_train: true do_train: true
finetuning_type: lora finetuning_type: lora
lora_target: q_proj,v_proj lora_target: all
### dataset ### dataset
dataset: alpaca_en dataset: alpaca_en
template: llama2 template: llama2
cutoff_len: 1024 cutoff_len: 1024
max_samples: 100000 max_samples: 10000
overwrite_cache: true overwrite_cache: true
preprocessing_num_workers: 16
### output ### output
output_dir: ./saves/LLaMA2-7B/lora/train_24_8_13_10_02 output_dir: ./saves/LLaMA2-7B/lora/train_24_8_13_13_16
logging_steps: 3 logging_steps: 3
save_steps: 100 save_steps: 100
plot_loss: true plot_loss: true
@ -24,15 +25,15 @@ overwrite_output_dir: true
### train ### train
per_device_train_batch_size: 2 per_device_train_batch_size: 2
gradient_accumulation_steps: 8 gradient_accumulation_steps: 8
learning_rate: 5.0e-05 learning_rate: 1.0e-4
num_train_epochs: 10 num_train_epochs: 10.0
lr_scheduler_type: cosine lr_scheduler_type: cosine
fp16: true warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000
max_grad_norm: 1.0 ### eval
warmup_steps: 0 val_size: 0.1
optim: adamw_torch per_device_eval_batch_size: 2
report_to: none eval_strategy: steps
lora_rank: 8 eval_steps: 500
lora_alpha: 16
lora_dropout: 0.1