fix: models from main repo; handle nan columns

draw_utils.py

@@ -39,6 +39,9 @@ def load_results():
     old_results_filtered = old_results[~old_results['normalized_name'].isin(duplicate_models)]
     res = pd.concat([old_results_filtered, new_results])
     res.drop('normalized_name', axis=1, inplace=True)
+    for col in LENGTHS:
+        if col not in res.columns:
+            res[col] = np.nan
 
     res['task'] = res['task'].str.replace('avg', 'avg(qa1-5)')
 

setup.sh

@@ -1,8 +1,9 @@
 #!/bin/bash
 # Clone repositories
-git clone https://github.com/booydar/babilong -b feat/babilong_evals_hf
+git clone https://github.com/booydar/babilong
 git clone https://huggingface.co/datasets/RMT-team/babilong_evals babilong/babilong_evals_new
 
 # Run the evaluation script
 cd babilong
-python -m babilong.collect_results --model_name all --save_path ./babilong_results --evals_path ./babilong_evals_new
+python -m babilong.collect_results --model_name all --save_path ./babilong_results --evals_path ./babilong_evals_new
+python -m babilong.collect_results --model_name all --save_path ./babilong_results --evals_path ./babilong_evals