Update big.yml

big.yml

@@ -27,7 +27,7 @@ datasets:
 
 shuffle_merged_datasets: true
 dataset_prepared_path: prepared_data
-val_set_size: 0.01
+val_set_size: 0.0
 output_dir: ./output
 
 plugins:
@@ -69,7 +69,7 @@ micro_batch_size: 2
 num_epochs: 4
 optimizer: paged_adamw_8bit
 lr_scheduler: cosine
-learning_rate: 5e-5
+learning_rate: 5e-7
 max_grad_norm: 0.2
 
 train_on_inputs: false
@@ -88,9 +88,9 @@ flash_attention: true
 s2_attention:
 
 warmup_steps: 40
-evals_per_epoch: 2
+evals_per_epoch: 
 eval_table_size:
-eval_max_new_tokens: 128
+eval_max_new_tokens: 
 saves_per_epoch: 2
 debug:
 deepspeed: ./deepspeed_configs/zero3_bf16.json
@@ -98,4 +98,3 @@ weight_decay: 0.01
 fsdp:
 fsdp_config:
 special_tokens:
-   pad_token: <|finetune_right_pad_id|>