Spaces:
Sleeping
Sleeping
File size: 6,947 Bytes
97a7635 c4165a7 30e2c08 c4165a7 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 30e2c08 97a7635 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 |
import gradio as gr
import os
from interface_utils import *
maxim = 'benevolence'
submaxims = ["The response does not exhibit insensitivity, rudeness, or harm.",
"The response does not reflect an engagement or endorsement with requests that are harmful or unethical."]
checkbox_choices = [
["Yes", "No", "NA"],
["Yes", "No", "NA"]
]
conversation_data_sliced = load_from_jsonl('./data/conversations_unlabeled_sliced.jsonl')
max_conversation_length = max([len(conversation['transcript']) for conversation in conversation_data_sliced])
conversation = get_conversation(conversation_data_sliced)
def save_labels(conv_id, slice_idx, skipped, submaxim_0=None, submaxim_1=None):
data = {
'conv_id': conv_id,
'slice_idx': int(slice_idx),
'maxim': maxim,
'skipped': skipped,
'submaxim_0': submaxim_0,
'submaxim_1': submaxim_1,
}
os.makedirs("./labels", exist_ok=True)
with open(f"./labels/{maxim}_human_labels_{conv_id}_{slice_idx}.json", 'w') as f:
json.dump(data, f, indent=4)
def update_interface(new_conversation):
new_conv_id = new_conversation['conv_id']
new_slice_idx = new_conversation['slice_idx']
new_transcript = new_conversation['transcript']
is_contextual_turn = True if new_slice_idx > 0 else False
markdown_blocks = [None] * max_conversation_length
for i in range(max_conversation_length):
if i < len(new_transcript) and new_transcript[i]['speaker'] != '':
markdown_blocks[i] = gr.Markdown(
f""" **{new_transcript[i]['speaker']}**: {new_transcript[i]['response']}""",
visible=True)
else:
markdown_blocks[i] = gr.Markdown("", visible=False)
new_last_response = gr.Text(value=get_last_response(new_transcript),
label="",
lines=1,
container=False,
interactive=False,
autoscroll=True,
visible=True)
new_radio_0_base = gr.Radio(label=submaxims[0],
choices=checkbox_choices[0],
value=None,
visible=True)
new_radio_1_base = gr.Radio(label=submaxims[1],
choices=checkbox_choices[1],
value=None,
visible=is_contextual_turn)
conv_len = gr.Number(value=len(new_transcript), visible=False)
return [new_conv_id] + [new_slice_idx] + list(markdown_blocks) + [new_last_response] + [new_radio_0_base] + [new_radio_1_base] + [conv_len]
def submit(*args):
conv_id = args[0]
slice_idx = args[1]
submaxim_0 = args[-3]
submaxim_1 = args[-2]
save_labels(conv_id, slice_idx, skipped=False, submaxim_0=submaxim_0, submaxim_1=submaxim_1)
new_conversation = get_conversation(conversation_data_sliced)
return update_interface(new_conversation)
def skip(*args):
conv_id = args[0]
slice_idx = args[1]
save_labels(conv_id, slice_idx, skipped=True)
new_conversation = get_conversation(conversation_data_sliced)
return update_interface(new_conversation, slice_idx)
with gr.Blocks(theme=gr.themes.Default()) as interface:
conv_id = conversation['conv_id']
slice_idx = conversation['slice_idx']
transcript = conversation['transcript']
conv_len = gr.Number(value=len(transcript), visible=False)
is_contextual_turn = True if slice_idx > 0 else False
markdown_blocks = [None] * max_conversation_length
with gr.Column(scale=1, min_width=600):
with gr.Group():
gr.Markdown("""<span style='font-size: 16px;'> **Conversational context** </span>""",
visible=True)
for i in range(max_conversation_length):
if i < len(transcript):
markdown_blocks[i] = gr.Markdown(
f""" **{transcript[i]['speaker']}**: {transcript[i]['response']}""")
else:
markdown_blocks[i] = gr.Markdown("")
if i >= conv_len.value:
markdown_blocks[i].visible = False
with gr.Row():
with gr.Group(elem_classes="bottom-aligned-group"):
speaker_adapted = gr.Markdown(
f"""<span style='font-size: 16px;'> **Response to label** </span>""",
visible=True)
last_response = gr.Textbox(value=get_last_response(transcript),
label="",
lines=1,
container=False,
interactive=False,
autoscroll=True,
visible=True)
radio_submaxim_0_base = gr.Radio(label=submaxims[0],
choices=checkbox_choices[0],
value=None,
visible=True)
radio_submaxim_1_base = gr.Radio(label=submaxims[1],
choices=checkbox_choices[1],
value=None,
visible=is_contextual_turn)
submit_button = gr.Button("Submit")
skip_button = gr.Button("Skip")
conv_id_element = gr.Text(value=conv_id, visible=False)
slice_idx_element = gr.Text(value=slice_idx, visible=False)
input_list = [conv_id_element] + \
[slice_idx_element] + \
markdown_blocks + \
[last_response] + \
[radio_submaxim_0_base] + \
[radio_submaxim_1_base] + \
[conv_len]
submit_button.click(
fn=submit,
inputs=input_list,
outputs=[conv_id_element,
slice_idx_element,
*markdown_blocks,
last_response,
radio_submaxim_0_base,
radio_submaxim_1_base,
conv_len]
)
skip_button.click(
fn=skip,
inputs=input_list,
outputs=[conv_id_element,
slice_idx_element,
*markdown_blocks,
last_response,
radio_submaxim_0_base,
radio_submaxim_1_base,
conv_len]
)
css = """
#textbox_id textarea {
background-color: white;
}
.bottom-aligned-group {
display: flex;
flex-direction: column;
justify-content: flex-end;
height: 100%;
}
"""
interface.css = css
interface.launch()
|