sedrickkeh commited on
Commit
0c89198
1 Parent(s): 1c3ec6c

Upload configs.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. configs.yaml +46 -0
configs.yaml ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ adam_beta1: '0.9'
2
+ adam_beta2: '0.999'
3
+ assistant_tag: gpt
4
+ bf16: 'True'
5
+ content_tag: value
6
+ cutoff_len: '2048'
7
+ dataset: mlfoundations-dev/oh-dcft-v1.2_no-curation_gpt-4o-mini
8
+ dataset_dir: ONLINE
9
+ ddp_timeout: '180000000'
10
+ deepspeed: /opt/ml/code/zero3.json
11
+ do_train: 'True'
12
+ enable_liger_kernel: 'False'
13
+ eval_strategy: epoch
14
+ finetuning_type: full
15
+ formatting: sharegpt
16
+ global_batch_size: '512'
17
+ gradient_accumulation_steps: '8'
18
+ gradient_checkpointing: 'True'
19
+ hub_model_id: mlfoundations-dev/hp_ablations_mistral_scheduler_cosine_warmup0.15_dcftv1.2
20
+ learning_rate: 5e-06
21
+ logging_steps: '10'
22
+ lr_scheduler_type: cosine
23
+ max_grad_norm: '1'
24
+ messages: conversations
25
+ model_name_or_path: mistralai/Mistral-7B-v0.1
26
+ neat_packing: 'True'
27
+ num_train_epochs: '3.0'
28
+ output_dir: /opt/ml/checkpoints
29
+ overwrite_cache: 'True'
30
+ overwrite_output_dir: 'True'
31
+ packing: 'True'
32
+ per_device_train_batch_size: '8'
33
+ plot_loss: 'True'
34
+ preprocessing_num_workers: '16'
35
+ push_to_db: 'True'
36
+ push_to_hub: 'True'
37
+ report_to: wandb
38
+ role_tag: from
39
+ run_name: hp_ablations_mistral_scheduler_cosine_warmup0.15_dcftv1.2
40
+ save_strategy: epoch
41
+ stage: sft
42
+ template: mistral
43
+ user_tag: human
44
+ val_size: '0.05'
45
+ warmup_ratio: '0.15'
46
+ weight_decay: '0.1'