shajmaan commited on
Commit
62c67ad
1 Parent(s): 535f57b

Upload 2 files

Browse files
Files changed (2) hide show
  1. 2024-07-18-08-15-04.yaml +65 -0
  2. training_loss.png +0 -0
2024-07-18-08-15-04.yaml ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ top.booster: auto
2
+ top.checkpoint_path: []
3
+ top.finetuning_type: lora
4
+ top.model_name: LLaMA2-7B
5
+ top.quantization_bit: '8'
6
+ top.quantization_method: bitsandbytes
7
+ top.rope_scaling: none
8
+ top.template: default
9
+ top.visual_inputs: false
10
+ train.additional_target: ''
11
+ train.badam_mode: layer
12
+ train.badam_switch_interval: 50
13
+ train.badam_switch_mode: ascending
14
+ train.badam_update_ratio: 0.05
15
+ train.batch_size: 2
16
+ train.compute_type: bf16
17
+ train.create_new_adapter: false
18
+ train.cutoff_len: 1024
19
+ train.dataset:
20
+ - an_data
21
+ train.dataset_dir: data
22
+ train.ds_offload: false
23
+ train.ds_stage: none
24
+ train.freeze_extra_modules: ''
25
+ train.freeze_trainable_layers: 2
26
+ train.freeze_trainable_modules: all
27
+ train.galore_rank: 16
28
+ train.galore_scale: 0.25
29
+ train.galore_target: all
30
+ train.galore_update_interval: 200
31
+ train.gradient_accumulation_steps: 8
32
+ train.learning_rate: 5e-5
33
+ train.logging_steps: 1
34
+ train.lora_alpha: 16
35
+ train.lora_dropout: 0
36
+ train.lora_rank: 8
37
+ train.lora_target: ''
38
+ train.loraplus_lr_ratio: 0
39
+ train.lr_scheduler_type: cosine
40
+ train.max_grad_norm: '1.0'
41
+ train.max_samples: '100000'
42
+ train.neat_packing: false
43
+ train.neftune_alpha: 0
44
+ train.num_train_epochs: '3.0'
45
+ train.optim: adamw_torch
46
+ train.packing: true
47
+ train.ppo_score_norm: false
48
+ train.ppo_whiten_rewards: false
49
+ train.pref_beta: 0.1
50
+ train.pref_ftx: 0
51
+ train.pref_loss: sigmoid
52
+ train.report_to: false
53
+ train.resize_vocab: false
54
+ train.reward_model: null
55
+ train.save_steps: 100
56
+ train.shift_attn: false
57
+ train.training_stage: Pre-Training
58
+ train.use_badam: false
59
+ train.use_dora: false
60
+ train.use_galore: false
61
+ train.use_llama_pro: false
62
+ train.use_pissa: false
63
+ train.use_rslora: false
64
+ train.val_size: 0
65
+ train.warmup_steps: 3
training_loss.png ADDED