File size: 10,308 Bytes
099e99c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14f85b1
099e99c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c2fbbc3
 
 
 
 
099e99c
14f85b1
099e99c
 
 
 
 
 
 
 
14f85b1
099e99c
 
 
 
14f85b1
099e99c
 
 
 
ab58a29
099e99c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ab58a29
099e99c
 
 
 
 
 
 
 
14f85b1
099e99c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
import json

import gradio as gr
import pandas as pd
from datasets import load_dataset
from gradio_huggingfacehub_search import HuggingfaceHubSearch

from src.distilabel_dataset_generator.utils import get_org_dropdown


def get_iframe(hub_repo_id) -> str:
    if not hub_repo_id:
        raise gr.Error("Hub repo id is required")
    url = f"https://huggingface.co/datasets/{hub_repo_id}/embed/viewer"
    iframe = f"""
    <iframe
  src="{url}"
  frameborder="0"
  width="100%"
  height="600px"
></iframe>
"""
    return iframe


def get_valid_columns(df: pd.DataFrame):
    valid_columns = []
    for col in df.columns:
        sample_val = df[col].iloc[0]
        if isinstance(sample_val, str) or (
            isinstance(sample_val, list)
            and all(isinstance(item, dict) for item in sample_val)
        ):
            valid_columns.append(col)
    return valid_columns


def load_dataset_from_hub(hub_repo_id: str, n_rows: int = 10):
    gr.Info(message="Loading dataset ...")
    if not hub_repo_id:
        raise gr.Error("Hub repo id is required")
    ds_dict = load_dataset(hub_repo_id)
    splits = list(ds_dict.keys())
    ds = ds_dict[splits[0]]
    if n_rows:
        ds = ds.select(range(n_rows))
    df = ds.to_pandas()
    # Get columns that contain either strings or lists of dictionaries
    valid_columns = get_valid_columns(df)
    return (
        df,
        gr.Dropdown(choices=valid_columns, label="Instruction Column"),
        gr.Dropdown(choices=valid_columns, label="Instruction Column"),
        gr.Dropdown(choices=valid_columns, label="Response Column"),
    )


def define_evaluation_aspects(task_type: str):
    if task_type == "instruction":
        return gr.Dropdown(
            value=["overall-rating"],
            choices=["complexity", "quality"],
            label="Evaluation Aspects",
            multiselect=True,
            interactive=True,
        )
    elif task_type == "instruction-response":
        return gr.Dropdown(
            value=["overall-rating"],
            choices=["helpfulness", "truthfulness", "overall-rating", "honesty"],
            label="Evaluation Aspects",
            multiselect=True,
            interactive=True,
        )
    else:
        return gr.Dropdown(interactive=False, visible=False)


def evaluate_instruction(df: pd.DataFrame, aspects: list[str], instruction_column: str):
    pass


def evaluate_instruction_response(
    df: pd.DataFrame, aspects: list[str], instruction_column: str, response_column: str
):
    pass


def evaluate_custom(
    df: pd.DataFrame, aspects: list[str], prompt_template: str, structured_output: dict
):
    pass


def _apply_to_dataset(
    df: pd.DataFrame,
    eval_type: str,
    aspects_instruction: list[str],
    instruction_column: str,
    aspects_instruction_response: list[str],
    instruction_column_response: str,
    response_column_response: str,
    aspects_custom: list[str],
    prompt_template: str,
    structured_output: dict,
):
    if eval_type == "instruction":
        df = evaluate_instruction(df, aspects_instruction, instruction_column)
    elif eval_type == "instruction-response":
        df = evaluate_instruction_response(
            df,
            aspects_instruction_response,
            instruction_column_response,
            response_column_response,
        )
    elif eval_type == "custom":
        df = evaluate_custom(df, aspects_custom, prompt_template, structured_output)
    return df


def apply_to_sample_dataset(
    repo_id: str,
    eval_type: str,
    aspects_instruction: list[str],
    aspects_instruction_response: list[str],
    aspects_custom: list[str],
    instruction_instruction: str,
    instruction_instruction_response: str,
    response_instruction_response: str,
    prompt_template: str,
    structured_output: dict,
):
    df, _, _, _ = load_dataset_from_hub(repo_id, n_rows=10)
    df = _apply_to_dataset(
        df,
        eval_type,
        aspects_instruction,
        instruction_instruction,
        aspects_instruction_response,
        instruction_instruction_response,
        response_instruction_response,
        aspects_custom,
        prompt_template,
        structured_output,
    )
    return df


def push_to_hub(
    org_name: str,
    repo_name: str,
    private: bool,
    n_rows: int,
    original_repo_id: str,
    eval_type: str,
    aspects_instruction: list[str],
    aspects_instruction_response: list[str],
    aspects_custom: list[str],
    instruction_instruction: str,
    instruction_instruction_response: str,
    response_instruction_response: str,
    prompt_template: str,
    structured_output: dict,
):
    df, _, _, _ = load_dataset_from_hub(original_repo_id, n_rows=n_rows)
    df = _apply_to_dataset(
        df,
        eval_type,
        aspects_instruction,
        instruction_instruction,
        aspects_instruction_response,
        instruction_instruction_response,
        response_instruction_response,
        aspects_custom,
        prompt_template,
        structured_output,
    )
    new_repo_id = f"{org_name}/{repo_name}"


######################
# Gradio UI
######################


with gr.Blocks() as app:
    gr.Markdown("## 1. Select your input dataset")
    with gr.Row():
        with gr.Column(scale=1):
            search_in = HuggingfaceHubSearch(
                label="Search",
                placeholder="Search for a Dataset",
                search_type="dataset",
                sumbit_on_select=True,
            )
            load_btn = gr.Button("Load dataset")
        with gr.Column(scale=3):
            search_out = gr.HTML(label="Dataset Preview")

    gr.HTML("<hr>")
    gr.Markdown("## 2. Configure your task")
    with gr.Row():
        with gr.Column(scale=1):
            eval_type = gr.Dropdown(
                label="Evaluation Type",
                choices=["instruction", "instruction-response", "custom-template"],
                visible=False,
            )
            with gr.Tab("instruction") as tab_instruction:
                aspects_instruction = define_evaluation_aspects("instruction")
                instruction_instruction = gr.Dropdown(
                    label="Instruction Column", interactive=True
                )
                tab_instruction.select(
                    lambda: "instruction",
                    inputs=[],
                    outputs=[eval_type],
                )
            with gr.Tab("instruction-response") as tab_instruction_response:
                aspects_instruction_response = define_evaluation_aspects(
                    "instruction-response"
                )
                instruction_instruction_response = gr.Dropdown(
                    label="Instruction Column", interactive=True
                )
                response_instruction_response = gr.Dropdown(
                    label="Response Column", interactive=True
                )
                tab_instruction_response.select(
                    lambda: "instruction-response",
                    inputs=[],
                    outputs=[eval_type],
                )
            with gr.Tab("custom") as tab_custom:
                aspects_custom = define_evaluation_aspects("custom")
                prompt_template = gr.Code(
                    label="Prompt Template",
                    value="{{column_1}} based on {{column_2}}",
                    language="markdown",
                    interactive=True,
                )
                structured_output = gr.Code(
                    label="Structured Output",
                    value=json.dumps({"eval_aspect": "str"}),
                    language="json",
                    interactive=True,
                )
                tab_custom.select(
                    lambda: "custom-template",
                    inputs=[],
                    outputs=[eval_type],
                )
            btn_apply_to_sample_dataset = gr.Button("Refresh dataset")
        with gr.Column(scale=3):
            dataframe = gr.Dataframe()

    gr.HTML("<hr>")
    gr.Markdown("## 3. Generate your dataset")
    with gr.Row():
        with gr.Column(scale=1):
            org_name = get_org_dropdown()
            repo_name = gr.Textbox(
                label="Repo name",
                placeholder="dataset_name",
                value="my-distiset",
                interactive=True,
            )
            n_rows = gr.Number(
                label="Number of rows",
                value=10,
                interactive=True,
                scale=1,
            )
            private = gr.Checkbox(
                label="Private dataset",
                value=False,
                interactive=True,
                scale=1,
            )
            btn_push_to_hub = gr.Button("Push to Hub", variant="primary", scale=2)
        with gr.Column(scale=3):
            success_message = gr.Markdown(visible=False)

    search_in.submit(get_iframe, inputs=search_in, outputs=search_out)
    load_btn.click(
        load_dataset_from_hub,
        inputs=[search_in],
        outputs=[
            dataframe,
            instruction_instruction,
            instruction_instruction_response,
            response_instruction_response,
        ],
    )
    btn_apply_to_sample_dataset.click(
        apply_to_sample_dataset,
        inputs=[
            search_in,
            eval_type,
            aspects_instruction,
            aspects_instruction_response,
            aspects_custom,
            instruction_instruction,
            instruction_instruction_response,
            response_instruction_response,
            prompt_template,
            structured_output,
        ],
        outputs=dataframe,
    )
    btn_push_to_hub.click(
        push_to_hub,
        inputs=[
            org_name,
            repo_name,
            private,
            n_rows,
            search_in,
            eval_type,
            aspects_instruction,
            aspects_instruction_response,
            aspects_custom,
            instruction_instruction,
            instruction_instruction_response,
            response_instruction_response,
            prompt_template,
            structured_output,
        ],
        outputs=success_message,
    )
    app.load(fn=get_org_dropdown, outputs=[org_name])