File size: 8,774 Bytes
eb25f8a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os

import gradio as gr
from huggingface_hub import Repository
from text_generation import Client

# from dialogues import DialogueTemplate
from share_btn import (community_icon_html, loading_icon_html, share_btn_css,
                       share_js)

HF_TOKEN = os.environ.get("HF_TOKEN", None)
API_TOKEN = os.environ.get("API_TOKEN", None)
API_URL = os.environ.get("API_URL", None)
API_URL = "https://api-inference.huggingface.co/models/timdettmers/guanaco-33b-merged"

client = Client(
    API_URL,
    headers={"Authorization": f"Bearer {API_TOKEN}"},
)

repo = None


def get_total_inputs(inputs, chatbot, preprompt, user_name, assistant_name, sep):
    past = []
    for data in chatbot:
        user_data, model_data = data

        if not user_data.startswith(user_name):
            user_data = user_name + user_data
        if not model_data.startswith(sep + assistant_name):
            model_data = sep + assistant_name + model_data

        past.append(user_data + model_data.rstrip() + sep)

    if not inputs.startswith(user_name):
        inputs = user_name + inputs

    total_inputs = preprompt + "".join(past) + inputs + sep + assistant_name.rstrip()

    return total_inputs


def has_no_history(chatbot, history):
    return not chatbot and not history


header = "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. "
prompt_template = "### Human: {query} ### Assistant:{response}"

def generate(
    system_message,
    user_message,
    chatbot,
    history,
    temperature,
    top_k,
    top_p,
    max_new_tokens,
    repetition_penalty,
    do_save=True,
):
    # Don't return meaningless message when the input is empty
    if not user_message:
        print("Empty input")

    history.append(user_message)

    past_messages = []
    for data in chatbot:
        user_data, model_data = data

        past_messages.extend(
            [{"role": "user", "content": user_data}, {"role": "assistant", "content": model_data.rstrip()}]
        )
        
    if len(past_messages) < 1:
        prompt = header + prompt_template.format(query=user_message, response="")
    else:
        prompt = header
        for i in range(0, len(past_messages), 2):
            intermediate_prompt = prompt_template.format(query=past_messages[i]["content"], response=past_messages[i+1]["content"])
            print("intermediate: ", intermediate_prompt)
            prompt = prompt + intermediate_prompt

        prompt = prompt + prompt_template.format(query=user_message, response="")


    generate_kwargs = {
        "temperature": temperature,
        "top_k": top_k,
        "top_p": top_p,
        "max_new_tokens": max_new_tokens,
    }

    temperature = float(temperature)
    if temperature < 1e-2:
        temperature = 1e-2
    top_p = float(top_p)

    generate_kwargs = dict(
        temperature=temperature,
        max_new_tokens=max_new_tokens,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True,
        truncate=999,
        seed=42,
    )

    stream = client.generate_stream(
        prompt,
        **generate_kwargs,
    )

    output = ""
    for idx, response in enumerate(stream):
        print(f'step {idx} - {response.token.text}')
        if response.token.text == '':
            break

        if response.token.special:
            continue
        output += response.token.text
        if idx == 0:
            history.append(" " + output)
        else:
            history[-1] = output

        chat = [(history[i].strip(), history[i + 1].strip()) for i in range(0, len(history) - 1, 2)]

        yield chat, history, user_message, ""

    return chat, history, user_message, ""


examples = [
    "A Llama entered in my garden, what should I do?"
]


def clear_chat():
    return [], []


def process_example(args):
    for [x, y] in generate(args):
        pass
    return [x, y]


title = """<h1 align="center">Guanaco Playground 💬</h1>"""
custom_css = """
#banner-image {
    display: block;
    margin-left: auto;
    margin-right: auto;
}
#chat-message {
    font-size: 14px;
    min-height: 300px;
}
"""

with gr.Blocks(analytics_enabled=False, css=custom_css) as demo:
    gr.HTML(title)

    with gr.Row():
        with gr.Column():
            gr.Markdown(
                """
            💻 This demo showcases the Guanaco 33B model, released together with the paper [QLoRA](https://arxiv.org/abs/2305.14314)
    """
            )

    with gr.Row():
        do_save = gr.Checkbox(
            value=True,
            label="Store data",
            info="You agree to the storage of your prompt and generated text for research and development purposes:",
        )
    with gr.Accordion(label="System Prompt", open=False, elem_id="parameters-accordion"):
        system_message = gr.Textbox(
            elem_id="system-message",
            placeholder="Below is a conversation between a human user and a helpful AI coding assistant.",
            show_label=False,
        )
    with gr.Row():
        with gr.Box():
            output = gr.Markdown()
            chatbot = gr.Chatbot(elem_id="chat-message", label="Chat")

    with gr.Row():
        with gr.Column(scale=3):
            user_message = gr.Textbox(placeholder="Enter your message here", show_label=False, elem_id="q-input")
            with gr.Row():
                send_button = gr.Button("Send", elem_id="send-btn", visible=True)

                clear_chat_button = gr.Button("Clear chat", elem_id="clear-btn", visible=True)

            with gr.Accordion(label="Parameters", open=False, elem_id="parameters-accordion"):
                temperature = gr.Slider(
                    label="Temperature",
                    value=0.2,
                    minimum=0.0,
                    maximum=1.0,
                    step=0.1,
                    interactive=True,
                    info="Higher values produce more diverse outputs",
                )
                top_k = gr.Slider(
                    label="Top-k",
                    value=50,
                    minimum=0.0,
                    maximum=100,
                    step=1,
                    interactive=True,
                    info="Sample from a shortlist of top-k tokens",
                )
                top_p = gr.Slider(
                    label="Top-p (nucleus sampling)",
                    value=0.95,
                    minimum=0.0,
                    maximum=1,
                    step=0.05,
                    interactive=True,
                    info="Higher values sample more low-probability tokens",
                )
                max_new_tokens = gr.Slider(
                    label="Max new tokens",
                    value=512,
                    minimum=0,
                    maximum=1024,
                    step=4,
                    interactive=True,
                    info="The maximum numbers of new tokens",
                )
                repetition_penalty = gr.Slider(
                    label="Repetition Penalty",
                    value=1.2,
                    minimum=0.0,
                    maximum=10,
                    step=0.1,
                    interactive=True,
                    info="The parameter for repetition penalty. 1.0 means no penalty.",
                )
            with gr.Row():
                gr.Examples(
                    examples=examples,
                    inputs=[user_message],
                    cache_examples=False,
                    fn=process_example,
                    outputs=[output],
                )

    history = gr.State([])
    # To clear out "message" input textbox and use this to regenerate message
    last_user_message = gr.State("")

    user_message.submit(
        generate,
        inputs=[
            system_message,
            user_message,
            chatbot,
            history,
            temperature,
            top_k,
            top_p,
            max_new_tokens,
            repetition_penalty,
            do_save,
        ],
        outputs=[chatbot, history, last_user_message, user_message],
    )

    send_button.click(
        generate,
        inputs=[
            system_message,
            user_message,
            chatbot,
            history,
            temperature,
            top_k,
            top_p,
            max_new_tokens,
            repetition_penalty,
            do_save,
        ],
        outputs=[chatbot, history, last_user_message, user_message],
    )

    clear_chat_button.click(clear_chat, outputs=[chatbot, history])
    # share_button.click(None, [], [], _js=share_js)

demo.queue(concurrency_count=16).launch(debug=True)