expand tests

rewardbench/rewardbench.py

@@ -133,7 +133,7 @@ def push_results_to_hub(args, results):
     if not args.hf_name:
         args.hf_name = f"rewardbench_eval_{timestamp}"
 
-    full_repo_id = f"{args.hf_entity}/{args.hf_name}_{timestamp}"
+    full_repo_id = f"{args.hf_entity}/{args.hf_name}"
 
     # Create repository on Hugging Face Hub
     api.create_repo(full_repo_id, repo_type="dataset", exist_ok=True)
@@ -476,8 +476,8 @@ def rewardbench(args: Args):
 
     # Consolidate chosen and rejected scores along with prompts and texts
     if is_preference_ranking:
-        combined_data["scores_chosen"] = scores_chosen
-        combined_data["scores_rejected"] = scores_rejected
+        combined_data["scores_chosen"] = [item for sublist in scores_chosen for item in sublist]
+        combined_data["scores_rejected"] = [item for sublist in scores_rejected for item in sublist]
         combined_data["text_chosen"] = dataset["text_chosen"]
         combined_data["text_rejected"] = dataset["text_rejected"]
     # or take instruction

tests/test_package.py

@@ -18,7 +18,7 @@
 from fastchat.conversation import get_conv_template
 from transformers import AutoTokenizer
 
-from rewardbench import load_preference_dataset
+from rewardbench import load_and_process_dataset
 
 
 class LoadAnyDataTest(unittest.TestCase):
@@ -31,15 +31,21 @@ def setUp(self):
         self.conv = get_conv_template("tulu")
 
     def test_load_standard_tokenizer(self):
-        load_preference_dataset(
+        load_and_process_dataset(
             "allenai/ultrafeedback_binarized_cleaned", split="test_prefs", tokenizer=self.tokenizer
         )
 
     def test_load_standard_conv(self):
-        load_preference_dataset("allenai/ultrafeedback_binarized_cleaned", split="test_prefs", conv=self.conv)
+        load_and_process_dataset("allenai/ultrafeedback_binarized_cleaned", split="test_prefs", conv=self.conv)
 
     def test_load_alt_tokenizer(self):
-        load_preference_dataset("allenai/preference-test-sets", split="shp", tokenizer=self.tokenizer)
+        load_and_process_dataset("allenai/preference-test-sets", split="shp", tokenizer=self.tokenizer)
 
     def test_load_alt_conv(self):
-        load_preference_dataset("allenai/preference-test-sets", split="shp", conv=self.conv)
+        load_and_process_dataset("allenai/preference-test-sets", split="shp", conv=self.conv)
+
+    def test_load_sft_tokenizer(self):
+        load_and_process_dataset("HuggingFaceH4/no_robots", split="test", tokenizer=self.tokenizer)
+
+    def test_load_sft_conv(self):
+        load_and_process_dataset("HuggingFaceH4/no_robots", split="test", conv=self.conv)