xiezhihui.x commited on
Commit
9db537a
·
1 Parent(s): 4265423

cleanup viewer

Browse files
Files changed (2) hide show
  1. app.py +1 -1
  2. data_reviewer.py +12 -11
app.py CHANGED
@@ -90,7 +90,7 @@ with demo:
90
  leaderboard = init_leaderboard(df)
91
 
92
  with gr.TabItem("📊 Data Viewer"):
93
- dataset_name, dataset_split, sample_idx = create_data_viewer()
94
 
95
  with gr.TabItem("ℹ️ About"):
96
  gr.Markdown(ABOUT_TEXT)
 
90
  leaderboard = init_leaderboard(df)
91
 
92
  with gr.TabItem("📊 Data Viewer"):
93
+ dataset_split, sample_idx = create_data_viewer()
94
 
95
  with gr.TabItem("ℹ️ About"):
96
  gr.Markdown(ABOUT_TEXT)
data_reviewer.py CHANGED
@@ -8,6 +8,9 @@ import gradio as gr
8
  from datasets import load_dataset
9
  from PIL import Image
10
 
 
 
 
11
 
12
  @lru_cache(maxsize=1)
13
  def load_cached_dataset(dataset_name, split):
@@ -31,9 +34,9 @@ def get_responses(responses, rankings):
31
  return chosen, rejected
32
 
33
 
34
- def load_and_display_sample(dataset_name, split, idx):
35
  try:
36
- dataset = load_cached_dataset(dataset_name, split)
37
  max_idx = len(dataset) - 1
38
  idx = min(max(0, int(idx)), max_idx)
39
 
@@ -75,14 +78,12 @@ def load_and_display_sample(dataset_name, split, idx):
75
 
76
  def create_data_viewer():
77
  # Pre-fetch initial data
78
- initial_dataset_name = "MMInstruction/VRewardBench"
79
  initial_split = "test"
80
  initial_idx = 0
81
- initial_data = load_and_display_sample(initial_dataset_name, initial_split, initial_idx)
82
 
83
  with gr.Column():
84
  with gr.Row():
85
- dataset_name = gr.Textbox(label="Dataset Name", value=initial_dataset_name, interactive=True)
86
  dataset_split = gr.Radio(choices=["test"], value=initial_split, label="Dataset Split")
87
  sample_idx = gr.Number(label="Sample Index", value=initial_idx, minimum=0, step=1, interactive=True)
88
  total_samples = gr.Textbox(
@@ -92,6 +93,7 @@ def create_data_viewer():
92
  with gr.Row():
93
  with gr.Column():
94
  image = gr.Image(label="Sample Image", type="pil", value=initial_data[0]) # Set initial image
 
95
 
96
  with gr.Column():
97
  sample_id = gr.Textbox(
@@ -106,14 +108,13 @@ def create_data_viewer():
106
  interactive=False,
107
  )
108
 
109
- with gr.Row():
110
  judge = gr.Textbox(label="Judge", value=initial_data[4], interactive=False) # Set initial judge
111
  query_source = gr.Textbox(
112
  label="Query Source", value=initial_data[5], interactive=False # Set initial query source
113
  )
114
- query = gr.Textbox(label="Query", value=initial_data[6], interactive=False) # Set initial query
115
 
116
- with gr.Row():
117
  with gr.Column():
118
  models_json = gr.JSON(label="Models", value=json.loads(initial_data[7])) # Set initial models
119
  meta_json = gr.JSON(label="Meta", value=json.loads(initial_data[8])) # Set initial meta
@@ -130,10 +131,10 @@ def create_data_viewer():
130
  )
131
 
132
  # Auto-update when any input changes
133
- for input_component in [dataset_name, dataset_split, sample_idx]:
134
  input_component.change(
135
  fn=load_and_display_sample,
136
- inputs=[dataset_name, dataset_split, sample_idx],
137
  outputs=[
138
  image,
139
  sample_id,
@@ -151,4 +152,4 @@ def create_data_viewer():
151
  ],
152
  )
153
 
154
- return dataset_name, dataset_split, sample_idx
 
8
  from datasets import load_dataset
9
  from PIL import Image
10
 
11
+ IGNORE_DETAILS = True
12
+ DATASET_NAME = "MMInstruction/VRewardBench"
13
+
14
 
15
  @lru_cache(maxsize=1)
16
  def load_cached_dataset(dataset_name, split):
 
34
  return chosen, rejected
35
 
36
 
37
+ def load_and_display_sample(split, idx):
38
  try:
39
+ dataset = load_cached_dataset(DATASET_NAME, split)
40
  max_idx = len(dataset) - 1
41
  idx = min(max(0, int(idx)), max_idx)
42
 
 
78
 
79
  def create_data_viewer():
80
  # Pre-fetch initial data
 
81
  initial_split = "test"
82
  initial_idx = 0
83
+ initial_data = load_and_display_sample(initial_split, initial_idx)
84
 
85
  with gr.Column():
86
  with gr.Row():
 
87
  dataset_split = gr.Radio(choices=["test"], value=initial_split, label="Dataset Split")
88
  sample_idx = gr.Number(label="Sample Index", value=initial_idx, minimum=0, step=1, interactive=True)
89
  total_samples = gr.Textbox(
 
93
  with gr.Row():
94
  with gr.Column():
95
  image = gr.Image(label="Sample Image", type="pil", value=initial_data[0]) # Set initial image
96
+ query = gr.Textbox(label="Query", value=initial_data[6], interactive=False) # Set initial query
97
 
98
  with gr.Column():
99
  sample_id = gr.Textbox(
 
108
  interactive=False,
109
  )
110
 
111
+ with gr.Row(visible=not IGNORE_DETAILS):
112
  judge = gr.Textbox(label="Judge", value=initial_data[4], interactive=False) # Set initial judge
113
  query_source = gr.Textbox(
114
  label="Query Source", value=initial_data[5], interactive=False # Set initial query source
115
  )
 
116
 
117
+ with gr.Row(visible=not IGNORE_DETAILS):
118
  with gr.Column():
119
  models_json = gr.JSON(label="Models", value=json.loads(initial_data[7])) # Set initial models
120
  meta_json = gr.JSON(label="Meta", value=json.loads(initial_data[8])) # Set initial meta
 
131
  )
132
 
133
  # Auto-update when any input changes
134
+ for input_component in [dataset_split, sample_idx]:
135
  input_component.change(
136
  fn=load_and_display_sample,
137
+ inputs=[dataset_split, sample_idx],
138
  outputs=[
139
  image,
140
  sample_id,
 
152
  ],
153
  )
154
 
155
+ return dataset_split, sample_idx