[misc] fix packing and eval plot (#7623)

examples/train_lora/llama3_lora_dpo.yaml

@@ -27,6 +27,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_kto.yaml

@@ -17,6 +17,7 @@ cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/kto
@@ -24,6 +25,7 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_ppo.yaml

@@ -17,6 +17,7 @@ cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
+dataloader_num_workers: 4
 
 ### output
 output_dir: saves/llama3-8b/lora/ppo
@@ -24,6 +25,7 @@ logging_steps: 10
 save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_pretrain.yaml

@@ -24,6 +24,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_reward.yaml

@@ -25,6 +25,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_sft.yaml

@@ -25,6 +25,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_sft_ds3.yaml

@@ -26,6 +26,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/llama3_lora_sft_ray.yaml

@@ -26,6 +26,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### ray
 ray_run_name: llama3_8b_sft_lora

examples/train_lora/llama4_lora_sft_ds3.yaml

@@ -28,10 +28,11 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1
-gradient_accumulation_steps: 8
+gradient_accumulation_steps: 2
 learning_rate: 1.0e-4
 num_train_epochs: 3.0
 lr_scheduler_type: cosine

examples/train_lora/llava1_5_lora_sft.yaml

@@ -25,6 +25,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/qwen2vl_lora_dpo.yaml

@@ -29,6 +29,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1

examples/train_lora/qwen2vl_lora_sft.yaml

@@ -27,6 +27,7 @@ save_steps: 500
 plot_loss: true
 overwrite_output_dir: true
 save_only_model: false
+report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]
 
 ### train
 per_device_train_batch_size: 1