Views
No views yet
1<|im_start|>system
2You are a helpful assistant that gives long and detailed answers.<|im_end|>
3<|im_start|>user
4{{instruction}}<|im_end|>
5<|im_start|>assistant
6{{response}}<|im_end|>1### model
2model_name_or_path: danube2-base-chatml
3
4### method
5stage: sft
6do_train: true
7finetuning_type: full
8use_badam: true
9badam_switch_mode: ascending
10badam_switch_interval: 50
11badam_verbose: 1
12badam_start_block: 6
13seed: 720
14
15### dataset
16dataset: wizardlm_evol_v2_196k_unfiltered
17template: ninja_chatml
18cutoff_len: 8192
19overwrite_cache: false
20preprocessing_num_workers: 12
21
22### output
23output_dir: wizardlm-evol-v2-chatml-badam
24logging_steps: 5
25save_steps: 1
26save_strategy: epoch
27plot_loss: true
28overwrite_output_dir: false
29
30### train
31per_device_train_batch_size: 2
32gradient_accumulation_steps: 8
33learning_rate: 0.00001
34num_train_epochs: 1
35lr_scheduler_type: constant_with_warmup
36warmup_ratio: 0.01
37pure_bf16: true
38flash_attn: fa2
39
40### eval
41val_size: 0.01
42per_device_eval_batch_size: 1
43eval_strategy: steps
44eval_steps: 1000| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.6195 | 0.1050 | 1000 | 0.7363 |
| 0.6788 | 0.2100 | 2000 | 0.7252 |
| 0.689 | 0.3150 | 3000 | 0.7172 |
| 0.6707 | 0.4200 | 4000 | 0.7133 |
| 0.6674 | 0.5250 | 5000 | 0.7091 |
| 0.7365 | 0.6301 | 6000 | 0.7085 |
| 0.7037 | 0.7351 | 7000 | 0.7066 |
| 0.709 | 0.8401 | 8000 | 0.7041 |
| 0.6652 | 0.9451 | 9000 | 0.7042 |