wip

examples/multimodal/pretrain-llava-mistral.yml

@@ -18,7 +18,7 @@ strict: false
 datasets:
   - path: liuhaotian/LLaVA-CC3M-Pretrain-595K
 dataset_prepared_path:
-val_set_size: 0.01
+val_set_size: 0.0
 output_dir: ./out
 
 sequence_len: 2048
@@ -33,8 +33,8 @@ wandb_log_model:
 
 gradient_accumulation_steps: 4
 micro_batch_size: 2
-num_epochs: 4
-optimizer: adamw_bnb_8bit
+num_epochs: 1
+optimizer: adamw_torch
 lr_scheduler: cosine
 learning_rate: 0.002