Hugging Face's logo

Spaces:
inflaton-ai
/
logical-reasoning
Build error

App Files Community
1
dh-mc commited on
Commit
64655af
·
1 Parent(s): b87bc1a

fix training configs

Browse files
Files changed (2) hide show
  1. llama-factory/config/mgtv_template.yaml +2 -2
  2. llama-factory/config/mgtv_template_4bit.yaml +2 -2
llama-factory/config/mgtv_template.yaml CHANGED
@@ -26,14 +26,14 @@ plot_loss: true
26
  per_device_train_batch_size: 16
27
  gradient_accumulation_steps: 8
28
  learning_rate: 1.0e-4
29
- num_train_epochs: 6.0
30
  lr_scheduler_type: cosine
31
  warmup_ratio: 0.1
32
  bf16: true
33
  ddp_timeout: 180000000
34
 
35
  ### eval
36
- val_size: 0.01
37
  per_device_eval_batch_size: 1
38
  eval_strategy: steps
39
  eval_steps: 35
 
26
  per_device_train_batch_size: 16
27
  gradient_accumulation_steps: 8
28
  learning_rate: 1.0e-4
29
+ num_train_epochs: 2.0
30
  lr_scheduler_type: cosine
31
  warmup_ratio: 0.1
32
  bf16: true
33
  ddp_timeout: 180000000
34
 
35
  ### eval
36
+ val_size: 0.1
37
  per_device_eval_batch_size: 1
38
  eval_strategy: steps
39
  eval_steps: 35
llama-factory/config/mgtv_template_4bit.yaml CHANGED
@@ -27,14 +27,14 @@ plot_loss: true
27
  per_device_train_batch_size: 16
28
  gradient_accumulation_steps: 8
29
  learning_rate: 1.0e-4
30
- num_train_epochs: 6.0
31
  lr_scheduler_type: cosine
32
  warmup_ratio: 0.1
33
  bf16: true
34
  ddp_timeout: 180000000
35
 
36
  ### eval
37
- val_size: 0.01
38
  per_device_eval_batch_size: 1
39
  eval_strategy: steps
40
  eval_steps: 35
 
27
  per_device_train_batch_size: 16
28
  gradient_accumulation_steps: 8
29
  learning_rate: 1.0e-4
30
+ num_train_epochs: 2.0
31
  lr_scheduler_type: cosine
32
  warmup_ratio: 0.1
33
  bf16: true
34
  ddp_timeout: 180000000
35
 
36
  ### eval
37
+ val_size: 0.1
38
  per_device_eval_batch_size: 1
39
  eval_strategy: steps
40
  eval_steps: 35