sedrickkeh commited on
Commit
7062cda
·
verified ·
1 Parent(s): 5cb3c1e

Upload configs.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. configs.yaml +40 -0
configs.yaml ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ adam_beta1: '0.9'
2
+ adam_beta2: '0.95'
3
+ assistant_tag: gpt
4
+ bf16: 'True'
5
+ content_tag: value
6
+ cutoff_len: '32768'
7
+ dataset: mlfoundations-dev/s1K_reformat
8
+ dataset_dir: ONLINE
9
+ ddp_timeout: '180000000'
10
+ deepspeed: /opt/ml/code/zero3_offload.json
11
+ do_train: 'True'
12
+ enable_liger_kernel: 'False'
13
+ finetuning_type: full
14
+ formatting: sharegpt
15
+ global_batch_size: '16'
16
+ gradient_accumulation_steps: '1'
17
+ hub_model_id: mlfoundations-dev/s1K_llama3.1_8b_32kcontext
18
+ learning_rate: 1e-05
19
+ logging_steps: '1'
20
+ lr_scheduler_type: cosine
21
+ max_samples: '1000000'
22
+ messages: conversations
23
+ model_name_or_path: meta-llama/Meta-Llama-3.1-8B
24
+ num_train_epochs: '5.0'
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: s1K_llama3.1_8b_32kcontext
35
+ save_strategy: epoch
36
+ stage: sft
37
+ template: llama3
38
+ user_tag: human
39
+ warmup_ratio: '0.05'
40
+ weight_decay: '0.0001'