fix preprocessing errors

data_reviewer.py

@@ -47,12 +47,6 @@ def load_and_display_sample(split, idx):
 
         # Process JSON data
         models = json.loads(sample["models"]) if isinstance(sample["models"], str) else sample["models"]
-        meta = json.loads(sample["meta"]) if isinstance(sample["meta"], str) else sample["meta"]
-        error_analysis = (
-            json.loads(sample["human_error_analysis"])
-            if isinstance(sample["human_error_analysis"], str)
-            else sample["human_error_analysis"]
-        )
 
         return (
             sample["image"],  # image
@@ -63,9 +57,7 @@ def load_and_display_sample(split, idx):
             sample["query_source"],  # query_source
             sample["query"],  # query
             json.dumps(models, indent=2),  # models_json
-            json.dumps(meta, indent=2),  # meta_json
             sample["rationale"],  # rationale
-            json.dumps(error_analysis, indent=2),  # error_analysis_json
             sample["ground_truth"],  # ground_truth
             f"Total samples: {len(dataset)}",  # total_samples
         )
@@ -78,53 +70,64 @@ def create_data_viewer():
     initial_split = "test"
     initial_idx = 0
     initial_data = load_and_display_sample(initial_split, initial_idx)
+    (
+        init_image,
+        init_sample_id,
+        init_chosen_response,
+        init_rejected_response,
+        init_judge,
+        init_query_source,
+        init_query,
+        init_models_json,
+        init_rationale,
+        init_ground_truth,
+        init_total_samples,
+    ) = initial_data
 
     with gr.Column():
         with gr.Row():
             dataset_split = gr.Radio(choices=["test"], value=initial_split, label="Dataset Split")
             sample_idx = gr.Number(label="Sample Index", value=initial_idx, minimum=0, step=1, interactive=True)
             total_samples = gr.Textbox(
-                label="Total Samples", value=initial_data[12], interactive=False  # Set initial total samples
+                label="Total Samples", value=init_total_samples, interactive=False  # Set initial total samples
             )
 
         with gr.Row():
             with gr.Column():
-                image = gr.Image(label="Sample Image", type="pil", value=initial_data[0])  # Set initial image
-                query = gr.Textbox(label="Query", value=initial_data[6], interactive=False)  # Set initial query
+                image = gr.Image(label="Sample Image", type="pil", value=init_image)  # Set initial image
+                query = gr.Textbox(label="Query", value=init_query, interactive=False)  # Set initial query
 
             with gr.Column():
                 sample_id = gr.Textbox(
-                    label="Sample ID", value=initial_data[1], interactive=False  # Set initial sample ID
+                    label="Sample ID", value=init_sample_id, interactive=False  # Set initial sample ID
                 )
                 chosen_response = gr.TextArea(
-                    label="Chosen Response ✅", value=initial_data[2], interactive=False  # Set initial chosen response
+                    label="Chosen Response ✅",
+                    value=init_chosen_response,
+                    interactive=False,  # Set initial chosen response
                 )
                 rejected_response = gr.TextArea(
                     label="Rejected Response ❌",
-                    value=initial_data[3],  # Set initial rejected response
+                    value=init_rejected_response,  # Set initial rejected response
                     interactive=False,
                 )
 
         with gr.Row(visible=not IGNORE_DETAILS):
-            judge = gr.Textbox(label="Judge", value=initial_data[4], interactive=False)  # Set initial judge
+            judge = gr.Textbox(label="Judge", value=init_judge, interactive=False)  # Set initial judge
             query_source = gr.Textbox(
-                label="Query Source", value=initial_data[5], interactive=False  # Set initial query source
+                label="Query Source", value=init_query_source, interactive=False  # Set initial query source
             )
 
         with gr.Row(visible=not IGNORE_DETAILS):
             with gr.Column():
-                models_json = gr.JSON(label="Models", value=json.loads(initial_data[7]))  # Set initial models
-                meta_json = gr.JSON(label="Meta", value=json.loads(initial_data[8]))  # Set initial meta
+                models_json = gr.JSON(label="Models", value=json.loads(init_models_json))  # Set initial models
                 rationale = gr.TextArea(
-                    label="Rationale", value=initial_data[9], interactive=False  # Set initial rationale
+                    label="Rationale", value=init_rationale, interactive=False  # Set initial rationale
                 )
 
             with gr.Column():
-                error_analysis_json = gr.JSON(
-                    label="Human Error Analysis", value=json.loads(initial_data[10])  # Set initial error analysis
-                )
                 ground_truth = gr.TextArea(
-                    label="Ground Truth", value=initial_data[11], interactive=False  # Set initial ground truth
+                    label="Ground Truth", value=init_ground_truth, interactive=False  # Set initial ground truth
                 )
 
         # Auto-update when any input changes
@@ -141,9 +144,7 @@ def create_data_viewer():
                     query_source,
                     query,
                     models_json,
-                    meta_json,
                     rationale,
-                    error_analysis_json,
                     ground_truth,
                     total_samples,
                 ],