File size: 10,991 Bytes
07423df
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import asyncio
import gc
import logging
import os
import threading
from typing import Any, Callable, Dict, List, Optional

import numpy as np
import torch
from h2o_wave import Q, ui
from transformers import AutoTokenizer, TextStreamer

from llm_studio.app_utils.utils import parse_ui_elements
from llm_studio.src.models.text_causal_language_modeling_model import Model
from llm_studio.src.utils.modeling_utils import (
    EnvVariableStoppingCriteria,
    get_torch_dtype,
    set_generation_config,
)

logger = logging.getLogger(__name__)

__all__ = ["chat_update", "is_app_blocked_while_streaming"]

USER = True
BOT = False


@torch.inference_mode(mode=True)
async def chat_update(q: Q) -> None:
    """
    Update the chatbot with the new message.
    """
    q.client["experiment/display/chat/finished"] = False
    try:
        await update_chat_window(q)
    finally:
        q.client["experiment/display/chat/finished"] = True
        # Hide the "Stop generating" button
        q.page["experiment/display/chat"].generating = False


async def update_chat_window(q):
    cfg_prediction = parse_ui_elements(
        cfg=q.client["experiment/display/chat/cfg"].prediction,
        q=q,
        pre="chat/cfg_predictions/cfg/",
    )
    q.client["experiment/display/chat/cfg"].prediction = cfg_prediction
    # Update generation config
    q.client["experiment/display/chat/model"].backbone = set_generation_config(
        q.client["experiment/display/chat/model"].backbone, cfg_prediction
    )

    # could also invoke cfg.check() here, but leave it explicit as cfg.check()
    # may raise other issues not related to the chatbot
    if cfg_prediction.do_sample and cfg_prediction.temperature == 0.0:
        q.page["meta"].dialog = ui.dialog(
            title="Invalid Text Generation configuration.",
            name="chatbot_invalid_settings",
            items=[
                ui.text(
                    "Do Sample enabled and Temperature = 0 are mutually exclusive. "
                    "Please increase Temperature or disable sampling."
                ),
            ],
            closable=True,
        )
        await q.page.save()
        return

    # populate chat window with user message
    logger.info(f"Using chatbot config: {cfg_prediction}")
    if q.events["experiment/display/chat/chatbot"]:
        prompt = q.events["experiment/display/chat/chatbot"]["suggestion"]
    else:
        prompt = q.client["experiment/display/chat/chatbot"]
    message = [prompt, USER]
    q.client["experiment/display/chat/messages"].append(message)
    q.page["experiment/display/chat"].data += message
    q.page["experiment/display/chat"].data += ["", BOT]
    await q.page.save()

    predicted_text = await answer_chat(q)

    # populate chat window with bot message
    logger.info(f"Predicted Answer: {predicted_text}")
    message = [predicted_text, BOT]
    q.client["experiment/display/chat/messages"].append(message)
    q.page["experiment/display/chat"].data[-1] = message


async def answer_chat(q: Q) -> str:
    cfg = q.client["experiment/display/chat/cfg"]
    model: Model = q.client["experiment/display/chat/model"]
    tokenizer = q.client["experiment/display/chat/tokenizer"]

    full_prompt = ""
    if len(q.client["experiment/display/chat/messages"]):
        for prev_message in q.client["experiment/display/chat/messages"][
            -(cfg.prediction.num_history + 1) :
        ]:
            if prev_message[1] is USER:
                prev_message = cfg.dataset.dataset_class.parse_prompt(
                    cfg, prev_message[0]
                )
            else:
                prev_message = prev_message[0]
                if cfg.dataset.add_eos_token_to_answer:
                    prev_message += cfg._tokenizer_eos_token

            full_prompt += prev_message
    logger.info(f"Full prompt: {full_prompt}")

    inputs = cfg.dataset.dataset_class.encode(
        tokenizer, full_prompt, cfg.tokenizer.max_length_prompt, "left"
    )
    inputs["prompt_input_ids"] = (
        inputs.pop("input_ids").unsqueeze(0).to(cfg.environment._device)
    )
    inputs["prompt_attention_mask"] = (
        inputs.pop("attention_mask").unsqueeze(0).to(cfg.environment._device)
    )

    def text_cleaner(text: str) -> str:
        return cfg.dataset.dataset_class.clean_output(
            output={"predicted_text": np.array([text])}, cfg=cfg
        )["predicted_text"][0]

    if cfg.prediction.num_beams == 1:
        streamer = WaveChatStreamer(tokenizer=tokenizer, q=q, text_cleaner=text_cleaner)
        # Need to start generation in a separate thread, otherwise streaming is blocked
        thread = threading.Thread(
            target=generate,
            kwargs=dict(model=model, inputs=inputs, cfg=cfg, streamer=streamer),
        )
        q.client["currently_chat_streaming"] = True
        # Show the "Stop generating" button
        q.page["experiment/display/chat"].generating = True
        # Hide suggestions
        q.page["experiment/display/chat"].suggestions = None

        try:
            thread.start()
            max_wait_time_in_seconds = 60 * 3
            for current_wait_time in range(max_wait_time_in_seconds):
                thread_is_dead = not thread.is_alive()
                takes_too_much_time = current_wait_time == max_wait_time_in_seconds - 1
                streaming_finished = streamer.finished

                if streaming_finished or takes_too_much_time or thread_is_dead:
                    if takes_too_much_time:
                        # this is more of a safety measure
                        # to ensure the app gets responsive eventually
                        logger.warning(
                            "Chat generation took too much time. "
                            "Stopping chat generation."
                        )
                    if thread_is_dead:  # some error occurred during streaming
                        logger.warning(
                            "Chat generation thread is not alive anymore. "
                            "Please check logs!"
                        )
                    if streaming_finished:
                        logger.info("Chat Stream has been completed")

                    predicted_text = streamer.answer
                    break
                await q.sleep(1)  # 1 second, see max_wait_time_in_seconds
        finally:
            del q.client["currently_chat_streaming"]
            if thread.is_alive():
                thread.join()
    else:
        # ValueError: `streamer` cannot be used with beam search (yet!).
        # Make sure that `num_beams` is set to 1.
        logger.info("Not streaming output, as it cannot be used with beam search.")
        q.page["experiment/display/chat"].data[-1] = ["...", BOT]
        await q.page.save()
        predicted_answer_ids = generate(model, inputs, cfg)[0]
        predicted_text = tokenizer.decode(
            predicted_answer_ids, skip_special_tokens=True
        )
        predicted_text = text_cleaner(predicted_text)

    del inputs
    gc.collect()
    torch.cuda.empty_cache()
    return predicted_text


class WaveChatStreamer(TextStreamer):
    """
    Utility class that updates the chabot card in a streaming fashion
    """

    def __init__(
        self,
        tokenizer: AutoTokenizer,
        q: Q,
        text_cleaner: Optional[Callable] = None,
        **decode_kwargs,
    ):
        super().__init__(tokenizer, skip_prompt=True, **decode_kwargs)
        self.text_cleaner = text_cleaner
        self.words_predicted_answer: List[str] = []
        self.q = q
        self.loop = asyncio.get_event_loop()
        self.finished = False

    def on_finalized_text(self, text: str, stream_end: bool = False):
        self.words_predicted_answer += [text]
        self.loop.create_task(self.update_chat_page())

    async def update_chat_page(self):
        self.q.page["experiment/display/chat"].data[-1] = [self.answer, BOT]
        await self.q.page.save()

    @property
    def answer(self):
        """
        Create the answer by joining the generated words.
        By this, self.text_cleaner does not need to be idempotent.
        """
        answer = "".join(self.words_predicted_answer)
        if answer.endswith(self.tokenizer.eos_token):
            # text generation is stopped
            answer = answer.replace(self.tokenizer.eos_token, "")
        if self.text_cleaner:
            answer = self.text_cleaner(answer)
        return answer

    def end(self):
        super().end()
        self.finished = True


def generate(model: Model, inputs: Dict, cfg: Any, streamer: TextStreamer = None):
    with torch.cuda.amp.autocast(
        dtype=get_torch_dtype(cfg.environment.mixed_precision_dtype)
    ):
        output = model.generate(batch=inputs, cfg=cfg, streamer=streamer).detach().cpu()
    return output


async def show_chat_is_running_dialog(q):
    q.page["meta"].dialog = ui.dialog(
        title="Text Generation is streaming.",
        name="chatbot_running_dialog",
        items=[
            ui.text("Please wait till the text generation has stopped."),
        ],
        closable=True,
    )
    await q.page.save()


async def show_stream_is_aborted_dialog(q):
    q.page["meta"].dialog = ui.dialog(
        title="Text Generation will be stopped.",
        name="chatbot_stopping_dialog",
        items=[
            ui.text("Please wait"),
        ],
        closable=False,
    )
    await q.page.save()


async def is_app_blocked_while_streaming(q: Q):
    """
    Check whether the app is blocked with current answer generation.
    """
    if (
        q.events["experiment/display/chat/chatbot"] is not None
        and q.events["experiment/display/chat/chatbot"]["stop"]
        and q.client["currently_chat_streaming"]
    ):
        # Cancel the streaming task.
        try:
            # User clicks abort button while the chat is currently streaming
            logger.info("Stopping Chat Stream")
            os.environ[EnvVariableStoppingCriteria.stop_streaming_env] = "True"
            await show_stream_is_aborted_dialog(q)
            await q.page.save()

            for _ in range(20):  # don't wait longer than 10 seconds
                await q.sleep(0.5)
                if q.client["currently_chat_streaming"] is None:
                    q.page["meta"].dialog = None
                    await q.page.save()
                    return True
            else:
                logger.warning("Could not terminate stream")
                return True
        finally:
            if EnvVariableStoppingCriteria.stop_streaming_env in os.environ:
                del os.environ[EnvVariableStoppingCriteria.stop_streaming_env]
            # Hide the "Stop generating" button.
            q.page["experiment/display/chat"].generating = False

    elif q.client["experiment/display/chat/finished"] is False:
        await show_chat_is_running_dialog(q)
        return True
    return False