Spaces:
Build error
Build error
Delete qwen2_7b_lora_sft.yaml
Browse files
llama-factory/config/qwen2_7b_lora_sft.yaml
DELETED
|
@@ -1,46 +0,0 @@
|
|
| 1 |
-
### model
|
| 2 |
-
model_name_or_path: Qwen/Qwen2-7B-Instruct
|
| 3 |
-
|
| 4 |
-
### method
|
| 5 |
-
stage: sft
|
| 6 |
-
do_train: true
|
| 7 |
-
finetuning_type: lora
|
| 8 |
-
lora_target: all
|
| 9 |
-
quantization_bit: 4 # use 4-bit QLoRA
|
| 10 |
-
loraplus_lr_ratio: 16.0 # use LoRA+ with lambda=16.0
|
| 11 |
-
# use_unsloth: true # use UnslothAI's LoRA optimization for 2x faster training
|
| 12 |
-
|
| 13 |
-
### dataset
|
| 14 |
-
dataset: alpaca_mac
|
| 15 |
-
template: chatml
|
| 16 |
-
cutoff_len: 1024
|
| 17 |
-
max_samples: 50
|
| 18 |
-
overwrite_cache: true
|
| 19 |
-
preprocessing_num_workers: 16
|
| 20 |
-
|
| 21 |
-
### output
|
| 22 |
-
output_dir: /content/qwen2-7b/
|
| 23 |
-
logging_steps: 10
|
| 24 |
-
save_steps: 10
|
| 25 |
-
plot_loss: true
|
| 26 |
-
overwrite_output_dir: true
|
| 27 |
-
# resume_from_checkpoint: true
|
| 28 |
-
|
| 29 |
-
### train
|
| 30 |
-
per_device_train_batch_size: 1
|
| 31 |
-
gradient_accumulation_steps: 8
|
| 32 |
-
learning_rate: 1.0e-4
|
| 33 |
-
num_train_epochs: 6.0
|
| 34 |
-
lr_scheduler_type: cosine
|
| 35 |
-
warmup_ratio: 0.1
|
| 36 |
-
bf16: true
|
| 37 |
-
ddp_timeout: 180000000
|
| 38 |
-
|
| 39 |
-
### eval
|
| 40 |
-
val_size: 0.02
|
| 41 |
-
per_device_eval_batch_size: 1
|
| 42 |
-
eval_strategy: steps
|
| 43 |
-
eval_steps: 10
|
| 44 |
-
|
| 45 |
-
report_to: wandb
|
| 46 |
-
run_name: qwen2_7b_mac_colab # optional
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|