train model

scripts/TRAIN.md

@@ -36,5 +36,5 @@ pip install -U -r requirements-lit.in
 ```
 
 ```bash
-litgpt pretrain --data LitData --data.data_path "../data/" --config ./model.yaml
+litgpt pretrain --config ./model.yaml
 ```

scripts/model.yaml

@@ -42,7 +42,12 @@ initial_checkpoint_dir:
 resume: "auto"
 
 # Data-related arguments. If not provided, the default is ``litgpt.data.TinyLlama``.
-# data: LitData
+data:
+  class_path: LitData
+  init_args:
+    data_path: "../data/"
+    # num_workers: 16
+    num_workers: 3
 
 # Training-related arguments. See ``litgpt.args.TrainArgs`` for details
 train:
@@ -56,8 +61,8 @@ train:
   global_batch_size: 512
 
   # Number of samples per data-parallel rank (type: int, default: 4)
-  micro_batch_size: 16
-  # micro_batch_size: 15
+  # micro_batch_size: 16
+  micro_batch_size: 15
 
   # Number of iterations with learning rate warmup active (type: int, default: 2000)
   lr_warmup_steps: 2000