Upload 2 files
Browse files- dataset_info.json +22 -0
- llama3_guardrail_lora_sft.yaml +37 -0
dataset_info.json
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"train_sample": {
|
3 |
+
"file_name": "train_sample.json",
|
4 |
+
"columns": {
|
5 |
+
"prompt": "instruction",
|
6 |
+
"query": "input",
|
7 |
+
"response": "output",
|
8 |
+
"system": "system",
|
9 |
+
"history": "history"
|
10 |
+
}
|
11 |
+
},
|
12 |
+
"test_sample": {
|
13 |
+
"file_name": "train_sample.json",
|
14 |
+
"columns": {
|
15 |
+
"prompt": "instruction",
|
16 |
+
"query": "input",
|
17 |
+
"response": "output",
|
18 |
+
"system": "system",
|
19 |
+
"history": "history"
|
20 |
+
}
|
21 |
+
}
|
22 |
+
}
|
llama3_guardrail_lora_sft.yaml
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
### examples/train_lora/llama3_lora_sft.yaml
|
2 |
+
model_name_or_path: "meta-llama/Llama-Guard-3-1B"
|
3 |
+
|
4 |
+
stage: sft
|
5 |
+
do_train: true
|
6 |
+
do_eval: true
|
7 |
+
finetuning_type: lora
|
8 |
+
lora_target: all
|
9 |
+
|
10 |
+
dataset: train_sample
|
11 |
+
eval_dataset: test_sample
|
12 |
+
dataset_dir: ./data
|
13 |
+
template: llama3
|
14 |
+
cutoff_len: 4096
|
15 |
+
max_samples: 1000
|
16 |
+
overwrite_cache: true
|
17 |
+
preprocessing_num_workers: 64
|
18 |
+
|
19 |
+
output_dir: ./saves/llama3-1b/lora/sft
|
20 |
+
logging_steps: 1
|
21 |
+
save_steps: 10
|
22 |
+
plot_loss: true
|
23 |
+
overwrite_output_dir: true
|
24 |
+
|
25 |
+
per_device_train_batch_size: 4
|
26 |
+
gradient_accumulation_steps: 8
|
27 |
+
learning_rate: 1.0e-4
|
28 |
+
num_train_epochs: 3.0
|
29 |
+
lr_scheduler_type: cosine
|
30 |
+
warmup_ratio: 0.1
|
31 |
+
bf16: true
|
32 |
+
ddp_timeout: 180000000
|
33 |
+
|
34 |
+
# val_size: 0.1
|
35 |
+
per_device_eval_batch_size: 16
|
36 |
+
eval_strategy: steps
|
37 |
+
eval_steps: 1
|