File size: 11,790 Bytes
cb80e0b
 
 
 
 
 
 
 
d7f914d
 
 
 
 
 
 
cb80e0b
 
 
 
 
 
 
 
 
 
d7f914d
cb80e0b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d7f914d
 
cb80e0b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d7f914d
cb80e0b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import time

import gradio as gr
import random
from conversation import Conversation


def get_tab_arena_side_by_side(download_bot_config, get_bot_profile, model_mapping, client):
    gr.Markdown("""
    # ⚔️ Chatbot Arena (side-by-side) ⚔️
    ## Rules
    * Chat with two models side-by-side and vote for which one is better!
    * You pick the models you want to chat with.
    * You can continue chatting and voting or click “Clear” to start a new round.
    """)
    default_bot_id = "_bot_e21de304-6151-4a04-b025-4c553ae8cbca"
    bot_config = download_bot_config(default_bot_id)
    user_state = gr.State(
        bot_config
    )
    with gr.Row():
        bot_id = gr.Textbox(label="Chai bot ID", value=default_bot_id, interactive=True)
        reload_bot_button = gr.Button("Reload bot")
    bot_profile = gr.HTML(get_bot_profile(bot_config))
    with gr.Accordion("Bot config:", open=False):
        bot_config_text = gr.Markdown(f"# Memory\n{bot_config['memory']}\n# Prompt\n{bot_config['prompt']}\n")

    with gr.Row():
        values = list(model_mapping.keys())
        first_message = (None, bot_config["firstMessage"])
        height = 450
        model_a_value, model_b_value = random.sample(values, 2)
        with gr.Column():
            model_a = gr.Dropdown(values, value=model_a_value, label="Model A")
            chatbot_a = gr.Chatbot([first_message])
            chatbot_a.style(height=height)
        with gr.Column():
            model_b = gr.Dropdown(values, value=model_b_value, label="Model B")
            chatbot_b = gr.Chatbot([first_message])
            chatbot_b.style(height=height)

    with gr.Row():
        with gr.Column(scale=3):
            msg = gr.Textbox(show_label=False, value="Hi there!", interactive=True)
        with gr.Column(scale=3):
            send = gr.Button("Send")
    with gr.Row():
        vote_a = gr.Button("👈 A is better", interactive=False)
        vote_b = gr.Button("👉 B is better", interactive=False)
        vote_tie = gr.Button("🤝 Tie", interactive=False)
        vote_bad = gr.Button("💩 Both are bad", interactive=False)
    with gr.Row():
        regenerate = gr.Button("Regenerate", interactive=False)
        clear = gr.Button("Clear")

    with gr.Accordion("Generation parameters for model A", open=False):
        model = model_mapping[model_a.value]
        temperature_model_a = gr.Slider(minimum=0.0, maximum=1.0, value=model.generation_params["temperature"],
                                        interactive=True, label="Temperature")
        repetition_penalty_model_a = gr.Slider(minimum=0.0, maximum=2.0,
                                               value=model.generation_params["repetition_penalty"],
                                               interactive=True, label="Repetition penalty")
        max_new_tokens_model_a = gr.Slider(minimum=1, maximum=512, value=model.generation_params["max_new_tokens"],
                                           interactive=True, label="Max new tokens")
        top_k_model_a = gr.Slider(minimum=1, maximum=100, value=model.generation_params["top_k"],
                                  interactive=True, label="Top-K")
        top_p_model_a = gr.Slider(minimum=0.0, maximum=1.0, value=model.generation_params["top_p"],
                                  interactive=True, label="Top-P")

    with gr.Accordion("Generation parameters for model B", open=False):
        model = model_mapping[model_b.value]
        temperature_model_b = gr.Slider(minimum=0.0, maximum=1.0, value=model.generation_params["temperature"],
                                        interactive=True, label="Temperature")
        repetition_penalty_model_b = gr.Slider(minimum=0.0, maximum=2.0,
                                               value=model.generation_params["repetition_penalty"],
                                               interactive=True, label="Repetition penalty")
        max_new_tokens_model_b = gr.Slider(minimum=1, maximum=512, value=model.generation_params["max_new_tokens"],
                                           interactive=True, label="Max new tokens")
        top_k_model_b = gr.Slider(minimum=1, maximum=100, value=model.generation_params["top_k"],
                                  interactive=True, label="Top-K")
        top_p_model_b = gr.Slider(minimum=0.0, maximum=1.0, value=model.generation_params["top_p"],
                                  interactive=True, label="Top-P")

    def clear_chat(user_state):
        return "", [(None, user_state["firstMessage"])], [(None, user_state["firstMessage"])]

    def reload_bot(bot_id):
        bot_config = download_bot_config(bot_id)
        bot_profile = get_bot_profile(bot_config)
        return bot_profile, [(None, bot_config["firstMessage"])], [(None, bot_config[
            "firstMessage"])], bot_config, f"# Memory\n{bot_config['memory']}\n# Prompt\n{bot_config['prompt']}"

    def get_generation_args(model_tag):
        model = model_mapping[model_tag]
        return (
            model.generation_params["temperature"],
            model.generation_params["repetition_penalty"],
            model.generation_params["max_new_tokens"],
            model.generation_params["top_k"],
            model.generation_params["top_p"],
        )

    def respond(message, chat_history, user_state, model_tag,
                temperature, repetition_penalty, max_new_tokens, top_k, top_p):
        custom_generation_params = {
            'temperature': temperature,
            'repetition_penalty': repetition_penalty,
            'max_new_tokens': max_new_tokens,
            'top_k': top_k,
            'top_p': top_p,
        }
        conv = Conversation(user_state)
        conv.set_chat_history(chat_history)
        conv.add_user_message(message)
        model = model_mapping[model_tag]
        bot_message = model.generate_response(conv, custom_generation_params)
        chat_history.append(
            (message, bot_message)
        )
        return "", chat_history

    def record_vote(user_state, vote,
                    chat_history_a, model_tag_a,
                    chat_history_b, model_tag_b):
        if len(chat_history_a) < 2:
            return
        conv_a = Conversation(user_state)
        conv_a.set_chat_history(chat_history_a)
        conv_b = Conversation(user_state)
        conv_b.set_chat_history(chat_history_b)
        if "A is better" in vote:
            vote_str = "model_a"
        elif "B is better" in vote:
            vote_str = "model_b"
        elif "Tie" in vote:
            vote_str = "tie"
        else:
            vote_str = "tie (bothbad)"
        row = {
            "timestamp": time.time(),
            "bot_id": user_state["bot_id"],
            "vote": vote_str,
            "model_a": model_tag_a,
            "model_b": model_tag_b,
            "is_anonymous": int(False)
        }
        sheet = client.open("Chat Arena").sheet1
        num_rows = len(sheet.get_all_records())
        sheet.insert_row(list(row.values()), index=num_rows + 2)
        return

    def regenerate_response(chat_history, user_state, model_tag,
                            temperature, repetition_penalty, max_new_tokens, top_k, top_p):
        custom_generation_params = {
            'temperature': temperature,
            'repetition_penalty': repetition_penalty,
            'max_new_tokens': max_new_tokens,
            'top_k': top_k,
            'top_p': top_p,
        }
        last_row = chat_history.pop(-1)
        chat_history.append((last_row[0], None))
        model = model_mapping[model_tag]
        conv = Conversation(user_state)
        conv.set_chat_history(chat_history)
        bot_message = model.generate_response(conv, custom_generation_params)
        chat_history[-1] = (last_row[0], bot_message)
        return "", chat_history

    def disable_voting():
        return [gr.Button.update(interactive=False)] * 4

    def enable_voting():
        return [gr.Button.update(interactive=True)] * 4

    def enable_send():
        return [gr.Button.update(interactive=True), gr.Button.update(interactive=False)]

    def enable_regenerate():
        return gr.Button.update(interactive=True)

    for vote in [vote_a, vote_b, vote_tie, vote_bad]:
        vote.click(record_vote,
                   [user_state, vote, chatbot_a, model_a, chatbot_b, model_b],
                   None,
                   queue=False)
        vote.click(disable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)

    model_a.change(get_generation_args, [model_a],
                   [temperature_model_a, repetition_penalty_model_a, max_new_tokens_model_a, top_k_model_a,
                    top_p_model_a], queue=False)
    model_b.change(get_generation_args, [model_b],
                   [temperature_model_b, repetition_penalty_model_b, max_new_tokens_model_b, top_k_model_b,
                    top_p_model_b], queue=False)
    reload_bot_button.click(reload_bot, [bot_id], [bot_profile, chatbot_a, chatbot_b, user_state, bot_config_text],
                            queue=False)
    clear.click(clear_chat, [user_state], [msg, chatbot_a, chatbot_b], queue=False)
    model_a.change(clear_chat, [user_state], [msg, chatbot_a, chatbot_b], queue=False)
    model_b.change(clear_chat, [user_state], [msg, chatbot_a, chatbot_b], queue=False)
    clear.click(enable_send, None, [send, regenerate], queue=False)
    reload_bot_button.click(enable_send, None, [send, regenerate], queue=False)

    model_a.change(enable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)
    model_b.change(enable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)
    reload_bot_button.click(disable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)
    send.click(enable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)
    clear.click(disable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)
    regenerate.click(enable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)
    msg.submit(enable_voting, None, [vote_a, vote_b, vote_tie, vote_bad], queue=False)

    send.click(respond,
               [msg, chatbot_a, user_state, model_a, temperature_model_a, repetition_penalty_model_a,
                max_new_tokens_model_a, top_k_model_a, top_p_model_a], [msg, chatbot_a],
               queue=False)
    msg.submit(respond,
               [msg, chatbot_a, user_state, model_a, temperature_model_a, repetition_penalty_model_a,
                max_new_tokens_model_a, top_k_model_a, top_p_model_a], [msg, chatbot_a],
               queue=False)

    send.click(respond,
               [msg, chatbot_b, user_state, model_b, temperature_model_b, repetition_penalty_model_b,
                max_new_tokens_model_b, top_k_model_b, top_p_model_b], [msg, chatbot_b],
               queue=False)
    msg.submit(respond,
               [msg, chatbot_b, user_state, model_b, temperature_model_b, repetition_penalty_model_b,
                max_new_tokens_model_b, top_k_model_b, top_p_model_b], [msg, chatbot_b],
               queue=False)

    send.click(enable_regenerate, None, [regenerate], queue=False)
    msg.submit(enable_regenerate, None, [regenerate], queue=False)

    regenerate.click(regenerate_response,
                     [chatbot_a, user_state, model_a, temperature_model_a, repetition_penalty_model_a,
                      max_new_tokens_model_a, top_k_model_a,
                      top_p_model_a], [msg, chatbot_a], queue=False)
    regenerate.click(regenerate_response,
                     [chatbot_b, user_state, model_b, temperature_model_b, repetition_penalty_model_b,
                      max_new_tokens_model_b, top_k_model_b,
                      top_p_model_b], [msg, chatbot_b], queue=False)