Upload root_scripts/fix_parquet.py with huggingface_hub

root_scripts/fix_parquet.py

@@ -0,0 +1,44 @@
+
+import pandas as pd
+
+path = "/workspace/rl4phyx/RL4Phyx/ZeroSearch/One-Shot-RLVR/data/train/physics_vlm/mechanics/mechanics_1_rl_numerical.parquet"
+df = pd.read_parquet(path)
+
+# Build a clean open-ended prompt from scratch
+open_ended_prompt = """Look at the image and answer the physics question.
+
+A patient with a dislocated shoulder is put into a traction apparatus as shown in figure. The pulls A and B have equal magnitudes and must combine to produce an outward traction force of 12.8 N on the patient's arm.
+
+Question: How large should these pulls be?
+
+Please reason step by step and put your final numerical answer (with units) in \\boxed{}."""
+
+# Rebuild all rows with clean prompt
+new_rows = []
+for i, row in df.iterrows():
+    r = {
+        "data_source": row["data_source"],
+        "prompt": [{"content": open_ended_prompt, "role": "user"}],
+        "ability": row["ability"],
+        "reward_model": row["reward_model"],  # keeps {'ground_truth': '7.55N', 'style': 'rule'}
+        "extra_info": row["extra_info"],
+    }
+    new_rows.append(r)
+
+new_df = pd.DataFrame(new_rows)
+new_df.to_parquet(path, index=False)
+
+# Verify
+df2 = pd.read_parquet(path)
+print("Shape:", df2.shape)
+print("Columns:", list(df2.columns))
+print()
+print("Prompt:")
+print(df2.iloc[0]["prompt"][0]["content"])
+print()
+print("reward_model:", df2.iloc[0]["reward_model"])
+print("extra_info:", df2.iloc[0]["extra_info"])
+print()
+has_options = "Options" in df2.iloc[0]["prompt"][0]["content"]
+print("Has Options:", has_options)
+print("DONE!")