Set ft dataloader name explicitly (#1187)

llmfoundry/data/finetuning/dataloader.py

@@ -53,7 +53,6 @@ def build_finetuning_dataloader(
     pin_memory: bool = True,
     prefetch_factor: int = 2,
     persistent_workers: bool = True,
-    name: Optional[str] = None,
     timeout: int = 0,
 ) -> DataSpec:
     """Builds a finetuning dataloader for training or evaluating.
@@ -161,7 +160,7 @@ def build_finetuning_dataloader(
 
     # this full config is necessary for properly profiling the packing ratio
     dataloader_cfg = {
-        'name': name,
+        'name': 'finetuning',
         'dataset': dataset_cfg,
         'drop_last': drop_last,
         'num_workers': num_workers,

llmfoundry/utils/config_utils.py

@@ -500,15 +500,13 @@ def log_config(cfg: Dict[str, Any]) -> None:
     config with different variables.
     """
     print(om.to_yaml(cfg))
-    if 'wandb' in cfg.get('loggers', {}):
-        try:
-            import wandb
-        except ImportError as e:
-            raise e
+    loggers = cfg.get('loggers', None) or {}
+    if 'wandb' in loggers:
+        import wandb
         if wandb.run:
             wandb.config.update(cfg)
 
-    if 'mlflow' in cfg.get('loggers', {}) and mlflow.active_run():
+    if 'mlflow' in loggers and mlflow.active_run():
         mlflow.log_params(params=om.to_container(cfg, resolve=True))
         _log_dataset_uri(cfg)
 

tests/a_scripts/inference/test_convert_composer_to_hf.py

@@ -849,6 +849,7 @@ def test_huggingface_conversion_callback(
         tokenizer_kwargs={'model_max_length': max_seq_len},
     )
 
+    dataloader_cfg.pop('name')
     train_dataloader = build_finetuning_dataloader(
         tokenizer=tokenizer,
         device_batch_size=device_batch_size,

tests/data/test_dataloader.py

@@ -318,7 +318,6 @@ def test_invalid_jsonl_data():
     packing_ratio = 'auto'
     allow_pad_trimming = False
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': 'iamroot/chat_malformatted_examples',
             'split': 'train',
@@ -374,7 +373,6 @@ def test_finetuning_dataloader(
     max_seq_len = 2048 if decoder_only_format else 1024
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name':
                 'iamroot/chat_formatted_examples' if use_chat_formatting else
@@ -444,7 +442,6 @@ def test_finetuning_dataloader_safe_load(
     # Clear the folder
     shutil.rmtree(DOWNLOADED_FT_DATASETS_DIRPATH, ignore_errors=True)
     cfg = DictConfig({
-        'name': 'finetuning',
         'dataset': {
             'hf_name': hf_name,
             'split': 'train',
@@ -507,7 +504,6 @@ def test_finetuning_dataloader_small_data(
         )
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': tiny_dataset_folder_path,
             'split': 'train',
@@ -564,7 +560,6 @@ def test_finetuning_dataloader_custom_split(tmp_path: pathlib.Path, split: str):
         make_tiny_ft_dataset(path=tiny_dataset_path, size=16)
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': tiny_dataset_folder_path,
             'split': split,
@@ -611,7 +606,6 @@ def test_finetuning_dataloader_custom_split_remote(split: str):
     max_seq_len = 2048
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': 's3://test-bucket/path/to/data',
             'split': split,
@@ -696,7 +690,6 @@ def test_finetuning_dataloader_streaming(
         }
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'max_seq_len': 2048,
             'decoder_only_format': True,
@@ -889,7 +882,6 @@ def test_malformed_data(
         )
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': str(tiny_dataset_folder_path),
             'split': 'train',
@@ -999,7 +991,6 @@ def test_malformed_conversation_data(
         )
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': str(tiny_dataset_folder_path),
             'split': 'train',
@@ -1238,7 +1229,6 @@ def test_token_counting_func_dataloader_setting(
 
     if dataloader_type == 'finetuning-hf':
         cfg = DictConfig({
-            'name': 'finetuning',
             'dataset': {
                 'hf_name': 'dummy-path',
                 'split': 'train',
@@ -1262,7 +1252,6 @@ def test_token_counting_func_dataloader_setting(
         )
     elif dataloader_type == 'finetuning-streaming':
         cfg = DictConfig({
-            'name': 'finetuning',
             'dataset': {
                 'remote': 'dummy-path',
                 'local': 'dummy-path',
@@ -1378,7 +1367,6 @@ def test_sharegpt_format(
         )
 
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': str(tiny_dataset_folder_path),
             'preprocessing_fn': 'teknium/OpenHermes-2.5',

tests/data/test_packing.py

@@ -177,7 +177,6 @@ def test_auto_packing_with_streaming_dataloader(tmp_path: Path):
     with MDSWriter(out=remote_dir, columns=columns, compression=None) as out:
         out.write({'prompt': 'HELLO', 'response': 'WORLD'})
     cfg = DictConfig({
-        'name': 'finetuning',
         'dataset': {
             'remote': remote_dir,
             'local': local_dir,
@@ -218,7 +217,6 @@ def test_packing_with_dataloader(packing_ratio: Any):
     reproducibility.seed_all(17)
     tokenizer = build_tokenizer('gpt2', {})
     cfg = {
-        'name': 'finetuning',
         'dataset': {
             'hf_name': 'tatsu-lab/alpaca',
             'split': 'train',

tests/fixtures/data.py

@@ -34,7 +34,6 @@ def tiny_ft_dataloader(
     device_batch_size: int = 1,
 ) -> DataLoader:
     dataloader_cfg = DictConfig({
-        'name': 'finetuning',
         'dataset': {
             'hf_name': str(tiny_ft_dataset_path),
             'split': 'train',