don't use zero first context for loading datasets

src/axolotl/utils/data/sft.py

@@ -53,7 +53,7 @@ from axolotl.utils.data.utils import (
     retry_on_request_exceptions,
 )
 from axolotl.utils.dict import DictDefault
-from axolotl.utils.distributed import is_local_main_process, zero_first
+from axolotl.utils.distributed import is_local_main_process
 from axolotl.utils.trainer import (
     calculate_total_num_steps,
     process_datasets_for_packing,
@@ -66,7 +66,6 @@ LOG = logging.getLogger(__name__)
 def prepare_dataset(cfg, tokenizer, processor=None, preprocess_iterable=None):
     prompters = []
     if not cfg.pretraining_dataset:
-        with zero_first(is_local_main_process()):
         if cfg.test_datasets:
             train_dataset, _, prompters = load_prepare_datasets(
                 tokenizer,
@@ -272,7 +271,7 @@ def load_tokenized_prepared_datasets(
         LOG.info("Loading raw datasets...")
         if not cfg.is_preprocess:
             LOG.warning(
-                "Processing datasets during training can lead to VRAM instability. Please pre-process your dataset."
+                "Processing datasets during training can lead to VRAM instability. Please use `axolotl preprocess` to prepare your dataset."
             )
 
         if cfg.seed: