blob: e41a41963c9f2d0656105d827379c812b8825303 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
|
### Qwen3-0.6B Full SFT Config (H200x4 Optimized)
model_name_or_path: Qwen/Qwen3-0.6B
stage: sft
do_train: true
finetuning_type: full
freeze_trainable_layers: 0
dataset: preference_extractor_train
template: qwen
cutoff_len: 1024
overwrite_cache: true
preprocessing_num_workers: 16
output_dir: saves/qwen3-0.6b-full-sft-h200
logging_steps: 5
save_strategy: steps
save_steps: 200
plot_loss: true
overwrite_output_dir: true
# H200x4 Configuration
# Total Batch Size = 32 * 4 * 1 = 128
per_device_train_batch_size: 32
gradient_accumulation_steps: 1
learning_rate: 2.0e-5
num_train_epochs: 1.0
lr_scheduler_type: cosine
warmup_ratio: 0.05
bf16: true
flash_attn: fa2
val_size: 0.01
per_device_eval_batch_size: 32
eval_strategy: steps
eval_steps: 200
|
