Merge branch 'main' into uc-not-enabled-error

llmfoundry/callbacks/hf_checkpointer.py

@@ -124,6 +124,7 @@ def _register_model_with_run_id_multiprocess(
         logging.basicConfig(
             format=
             f'%(asctime)s: rank{dist.get_global_rank()}[%(process)d][%(threadName)s]: %(levelname)s: %(name)s: %(message)s',
+            force=True,
         )
         logging.getLogger('composer').setLevel(composer_logging_level)
 

llmfoundry/command_utils/data_prep/convert_text_to_mds.py

@@ -515,6 +515,7 @@ def _configure_logging(logging_level: str):
     logging.basicConfig(
         format=
         f'%(asctime)s: [%(process)d][%(threadName)s]: %(levelname)s: %(name)s: %(message)s',
+        force=True,
     )
     logging_level = logging_level.upper()
     logging.getLogger('llmfoundry').setLevel(logging_level)

llmfoundry/command_utils/eval.py

@@ -272,6 +272,7 @@ def evaluate(cfg: DictConfig) -> tuple[list[Trainer], pd.DataFrame]:
             # 2022-06-29 11:22:26,152: rank0[822018][MainThread]: INFO: Message here
             format=
             f'%(asctime)s: rank{dist.get_global_rank()}[%(process)d][%(threadName)s]: %(levelname)s: %(name)s: %(message)s',
+            force=True,
         )
         logging.getLogger('llmfoundry').setLevel(
             eval_config.python_log_level.upper(),

llmfoundry/command_utils/train.py

@@ -227,6 +227,7 @@ def train(cfg: DictConfig) -> Trainer:
             # 2022-06-29 11:22:26,152: rank0[822018][MainThread]: INFO: Message here
             format=
             f'%(asctime)s: rank{dist.get_global_rank()}[%(process)d][%(threadName)s]: %(levelname)s: %(name)s: %(message)s',
+            force=True,
         )
         logging.getLogger('llmfoundry').setLevel(
             train_cfg.python_log_level.upper(),
@@ -311,20 +312,20 @@ def train(cfg: DictConfig) -> Trainer:
     eval_gauntlet_config = train_cfg.eval_gauntlet or train_cfg.eval_gauntlet_str
 
     # Optional parameters will be set to default values if not specified.
-    env_run_name: Optional[str] = os.environ.get('RUN_NAME', None)
-    run_name: str = (
-        train_cfg.run_name if train_cfg.run_name else env_run_name
-    ) or 'llm'
+    run_name: Optional[
+        str] = train_cfg.run_name if train_cfg.run_name else os.environ.get(
+            'RUN_NAME',
+            None,
+        )
     is_state_dict_sharded: bool = (
         fsdp_config.get('state_dict_type', 'full') == 'sharded'
     ) if fsdp_config else False
     save_latest_filename: str = train_cfg.save_latest_filename if train_cfg.save_latest_filename else 'latest-sharded-rank{rank}' if is_state_dict_sharded else 'latest-rank{rank}.pt'
     save_filename: str = train_cfg.save_filename if train_cfg.save_filename else 'ep{epoch}-ba{batch}-rank{rank}.pt'
 
     # Enable autoresume from model checkpoints if possible
-    is_user_set_run_name: bool = train_cfg.run_name is not None or env_run_name is not None
     autoresume_default: bool = False
-    if is_user_set_run_name and \
+    if run_name is not None and \
         train_cfg.save_folder is not None \
         and not train_cfg.save_overwrite \
         and not train_cfg.save_weights_only:

scripts/inference/endpoint_generate.py

@@ -25,7 +25,11 @@
 
 from llmfoundry.utils import prompt_files as utils
 
-logging.basicConfig(level=logging.INFO, format='%(asctime)s %(message)s')
+logging.basicConfig(
+    format=
+    f'%(asctime)s: [%(process)d][%(threadName)s]: %(levelname)s: %(name)s: %(message)s',
+    force=True,
+)
 log = logging.getLogger(__name__)
 
 ENDPOINT_API_KEY_ENV: str = 'ENDPOINT_API_KEY'

scripts/misc/download_model.py

@@ -31,8 +31,9 @@
 HF_TOKEN_ENV_VAR = 'HF_TOKEN'
 
 logging.basicConfig(
-    format=f'%(asctime)s: %(levelname)s: %(name)s: %(message)s',
-    level=logging.INFO,
+    format=
+    f'%(asctime)s: [%(process)d][%(threadName)s]: %(levelname)s: %(name)s: %(message)s',
+    force=True,
 )
 log = logging.getLogger(__name__)