Track large file with Git LFs

.gitattributes

@@ -0,0 +1 @@
+data/wandb/tzebw6rb.parquet filter=lfs diff=lfs merge=lfs -text

data/wandb/tzebw6rb.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97537f22180da6992b01c2ba7566fabc7c265d27c2c317eda2a191d52e08b843
+size 13624028

utils.py

@@ -136,11 +136,12 @@ def load_downloaded_runs(time, cols=KEYS):
         'date-based question answering': 'date_qa',
         'question-answering': 'qa',
     }
-    if 'task' in df_all.columns:
-        df_all.task = df_all.task.apply(lambda x: task_mapping.get(x, x))
+    
+    df_all['task'] = df_all.task.apply(lambda x: task_mapping.get(x, x))
+
 
     # Runs which do not have a turn field are imputed to be turn zero (single turn)
-    df_all.turn.fillna(0, inplace=True)
+    df_all['turn'] = df_all.turn.fillna(0)
 
     df_all.sort_values(by=['_timestamp'], inplace=True)
 
@@ -230,7 +231,7 @@ def download_runs(time, df_vali):
             continue
 
         try:
-            pbar.set_description(f'* Downloading run {row.run_id!r}', flush=True)
+            pbar.set_description(f'* Downloading run {row.run_id!r}')
             run = api.run(row.run_path)
 
             # By default we just download a subset of events (500 most recent)