Log checkpoints as mlflow artifacts (#1976)

* Ensure hf_mlflow_log_artifact config var is set in env

* Add transformer MLflowCallback to callbacks list when mlflow enabled

* Test hf_mlflow_log_artifacts is set correctly

* Test mlflow not being used by default

src/axolotl/core/trainer_builder.py

@@ -1119,12 +1119,17 @@ class TrainerBuilderBase(abc.ABC):
                 SaveAxolotlConfigtoWandBCallback(self.cfg.axolotl_config_path)
             )
         if self.cfg.use_mlflow and is_mlflow_available():
+            from transformers.integrations.integration_utils import MLflowCallback
+
             from axolotl.utils.callbacks.mlflow_ import (
                 SaveAxolotlConfigtoMlflowCallback,
             )
 
-            callbacks.append(
-                SaveAxolotlConfigtoMlflowCallback(self.cfg.axolotl_config_path)
+            callbacks.extend(
+                [
+                    SaveAxolotlConfigtoMlflowCallback(self.cfg.axolotl_config_path),
+                    MLflowCallback,
+                ]
             )
         if self.cfg.use_comet and is_comet_available():
             from axolotl.utils.callbacks.comet_ import SaveAxolotlConfigtoCometCallback

src/axolotl/utils/mlflow_.py

@@ -16,3 +16,7 @@ def setup_mlflow_env_vars(cfg: DictDefault):
     # Enable mlflow if experiment name is present
     if cfg.mlflow_experiment_name and len(cfg.mlflow_experiment_name) > 0:
         cfg.use_mlflow = True
+
+    # Enable logging hf artifacts in mlflow if value is truthy
+    if cfg.hf_mlflow_log_artifacts is True:
+        os.environ["HF_MLFLOW_LOG_ARTIFACTS"] = "true"