sedrickkeh commited on
Commit
d8d607a
·
verified ·
1 Parent(s): 138dfbe

Upload configs.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. configs.yaml +39 -0
configs.yaml ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ assistant_tag: assistant
2
+ bf16: 'True'
3
+ content_tag: value
4
+ cutoff_len: '16384'
5
+ dataloader_num_workers: '4'
6
+ dataloader_persistent_workers: 'True'
7
+ dataloader_pin_memory: 'True'
8
+ dataset: mlfoundations-dev/openthoughts_316
9
+ dataset_dir: ONLINE
10
+ ddp_timeout: '180000000'
11
+ deepspeed: /opt/ml/code/zero3.json
12
+ do_train: 'True'
13
+ enable_liger_kernel: 'True'
14
+ finetuning_type: full
15
+ formatting: sharegpt
16
+ global_batch_size: '32'
17
+ gradient_accumulation_steps: '4'
18
+ hub_model_id: mlfoundations-dev/316_globalbatchsize32_lr4e5_epochs15
19
+ learning_rate: 4e-05
20
+ logging_steps: '1'
21
+ lr_scheduler_type: cosine
22
+ messages: conversations
23
+ model_name_or_path: Qwen/Qwen2.5-7B-Instruct
24
+ num_train_epochs: '15.0'
25
+ output_dir: /opt/ml/model
26
+ overwrite_cache: 'True'
27
+ per_device_train_batch_size: '1'
28
+ plot_loss: 'True'
29
+ preprocessing_num_workers: '16'
30
+ push_to_db: 'True'
31
+ push_to_hub: 'True'
32
+ report_to: wandb
33
+ role_tag: from
34
+ run_name: 316_globalbatchsize32_lr4e5_epochs15
35
+ save_strategy: epoch
36
+ stage: sft
37
+ template: qwen25
38
+ user_tag: user
39
+ warmup_ratio: '0.1'