Hugging Face's logo

Spaces:
inflaton-ai
/
logical-reasoning
Build error

App Files Community
1
dh-mc commited on
Commit
af72c0c
·
1 Parent(s): 685f9c5

reduce batch size to 16

Browse files
Files changed (2) hide show
  1. llama-factory/config/qwen2_72b_lora_sft_4bit-p1.yaml +2 -2
  2. llama-factory/config/qwen2_72b_lora_sft_4bit-p2.yaml +2 -2
llama-factory/config/qwen2_72b_lora_sft_4bit-p1.yaml CHANGED
@@ -28,10 +28,10 @@ overwrite_output_dir: true
28
  # resume_from_checkpoint: true
29
 
30
  ### train
31
- per_device_train_batch_size: 32
32
  gradient_accumulation_steps: 8
33
  learning_rate: 1.0e-4
34
- num_train_epochs: 4.0
35
  lr_scheduler_type: cosine
36
  warmup_ratio: 0.1
37
  bf16: true
 
28
  # resume_from_checkpoint: true
29
 
30
  ### train
31
+ per_device_train_batch_size: 16
32
  gradient_accumulation_steps: 8
33
  learning_rate: 1.0e-4
34
+ num_train_epochs: 3.0
35
  lr_scheduler_type: cosine
36
  warmup_ratio: 0.1
37
  bf16: true
llama-factory/config/qwen2_72b_lora_sft_4bit-p2.yaml CHANGED
@@ -28,10 +28,10 @@ overwrite_output_dir: true
28
  # resume_from_checkpoint: true
29
 
30
  ### train
31
- per_device_train_batch_size: 32
32
  gradient_accumulation_steps: 8
33
  learning_rate: 1.0e-4
34
- num_train_epochs: 4.0
35
  lr_scheduler_type: cosine
36
  warmup_ratio: 0.1
37
  bf16: true
 
28
  # resume_from_checkpoint: true
29
 
30
  ### train
31
+ per_device_train_batch_size: 16
32
  gradient_accumulation_steps: 8
33
  learning_rate: 1.0e-4
34
+ num_train_epochs: 3.0
35
  lr_scheduler_type: cosine
36
  warmup_ratio: 0.1
37
  bf16: true