Support Sample packing for phi arch (#586)

* phi sequence packing

* sample packing fixes

* fix linting

* fix inference and phi e2e tests

* update phi example now that sample packing works

* wandb import keeps getting moved around

tests/e2e/test_lora_llama.py

@@ -7,39 +7,23 @@ import os
 import tempfile
 import unittest
 
+from axolotl.cli import load_datasets
 from axolotl.common.cli import TrainerCliArgs
-from axolotl.train import TrainDatasetMeta, train
+from axolotl.train import train
 from axolotl.utils.config import normalize_config
-from axolotl.utils.data import prepare_dataset
 from axolotl.utils.dict import DictDefault
-from axolotl.utils.models import load_tokenizer
 
 LOG = logging.getLogger("axolotl.tests.e2e")
 os.environ["WANDB_DISABLED"] = "true"
 
 
-def load_datasets(
-    *,
-    cfg: DictDefault,
-    cli_args: TrainerCliArgs,  # pylint:disable=unused-argument
-) -> TrainDatasetMeta:
-    tokenizer = load_tokenizer(cfg)
-
-    train_dataset, eval_dataset, total_num_steps = prepare_dataset(cfg, tokenizer)
-
-    return TrainDatasetMeta(
-        train_dataset=train_dataset,
-        eval_dataset=eval_dataset,
-        total_num_steps=total_num_steps,
-    )
-
-
 class TestLoraLlama(unittest.TestCase):
     """
     Test case for Llama models using LoRA
     """
 
     def test_lora(self):
+        # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
                 "base_model": "JackFram/llama-68m",
@@ -80,6 +64,7 @@ class TestLoraLlama(unittest.TestCase):
         train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
 
     def test_lora_packing(self):
+        # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
                 "base_model": "JackFram/llama-68m",