File size: 35,355 Bytes
e679d69
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
import asyncio
import json
import os
import time
import traceback
import warnings
from concurrent.futures import ThreadPoolExecutor
from logging import getLogger
from threading import Lock
from typing import AsyncGenerator, Dict, List, Optional, Union

import aiohttp
import requests

from ..schema import ModelStatusCode
from ..utils import filter_suffix
from .base_api import AsyncBaseAPILLM, BaseAPILLM

warnings.simplefilter('default')

OPENAI_API_BASE = 'https://api.openai.com/v1/chat/completions'


class GPTAPI(BaseAPILLM):
    """Model wrapper around OpenAI's models.

    Args:
        model_type (str): The name of OpenAI's model.
        retry (int): Number of retires if the API call fails. Defaults to 2.
        key (str or List[str]): OpenAI key(s). In particular, when it
            is set to "ENV", the key will be fetched from the environment
            variable $OPENAI_API_KEY, as how openai defaults to be. If it's a
            list, the keys will be used in round-robin manner. Defaults to
            'ENV'.
        org (str or List[str], optional): OpenAI organization(s). If not
            specified, OpenAI uses the default organization bound to each API
            key. If specified, the orgs will be posted with each request in
            round-robin manner. Defaults to None.
        meta_template (Dict, optional): The model's meta prompt
            template if needed, in case the requirement of injecting or
            wrapping of any meta instructions.
        api_base (str): The base url of OpenAI's API. Defaults to
            'https://api.openai.com/v1/chat/completions'.
        gen_params: Default generation configuration which could be overridden
            on the fly of generation.
    """

    is_api: bool = True

    def __init__(self,
                 model_type: str = 'gpt-3.5-turbo',
                 retry: int = 2,
                 json_mode: bool = False,
                 key: Union[str, List[str]] = 'ENV',
                 org: Optional[Union[str, List[str]]] = None,
                 meta_template: Optional[Dict] = [
                     dict(role='system', api_role='system'),
                     dict(role='user', api_role='user'),
                     dict(role='assistant', api_role='assistant'),
                     dict(role='environment', api_role='system')
                 ],
                 api_base: str = OPENAI_API_BASE,
                 proxies: Optional[Dict] = None,
                 **gen_params):
        if 'top_k' in gen_params:
            warnings.warn('`top_k` parameter is deprecated in OpenAI APIs.',
                          DeprecationWarning)
            gen_params.pop('top_k')
        super().__init__(
            model_type=model_type,
            meta_template=meta_template,
            retry=retry,
            **gen_params)
        self.gen_params.pop('top_k')
        self.logger = getLogger(__name__)

        if isinstance(key, str):
            self.keys = [os.getenv('OPENAI_API_KEY') if key == 'ENV' else key]
        else:
            self.keys = key

        # record invalid keys and skip them when requesting API
        # - keys have insufficient_quota
        self.invalid_keys = set()

        self.key_ctr = 0
        if isinstance(org, str):
            self.orgs = [org]
        else:
            self.orgs = org
        self.org_ctr = 0
        self.url = api_base
        self.model_type = model_type
        self.proxies = proxies
        self.json_mode = json_mode

    def chat(
        self,
        inputs: Union[List[dict], List[List[dict]]],
        **gen_params,
    ) -> Union[str, List[str]]:
        """Generate responses given the contexts.

        Args:
            inputs (Union[List[dict], List[List[dict]]]): a list of messages
                or list of lists of messages
            gen_params: additional generation configuration

        Returns:
            Union[str, List[str]]: generated string(s)
        """
        assert isinstance(inputs, list)
        if 'max_tokens' in gen_params:
            raise NotImplementedError('unsupported parameter: max_tokens')
        gen_params = {**self.gen_params, **gen_params}
        with ThreadPoolExecutor(max_workers=20) as executor:
            tasks = [
                executor.submit(self._chat,
                                self.template_parser._prompt2api(messages),
                                **gen_params)
                for messages in (
                    [inputs] if isinstance(inputs[0], dict) else inputs)
            ]
        ret = [task.result() for task in tasks]
        return ret[0] if isinstance(inputs[0], dict) else ret

    def stream_chat(
        self,
        inputs: List[dict],
        **gen_params,
    ):
        """Generate responses given the contexts.

        Args:
            inputs (List[dict]): a list of messages
            gen_params: additional generation configuration

        Returns:
            str: generated string
        """
        assert isinstance(inputs, list)
        if 'max_tokens' in gen_params:
            raise NotImplementedError('unsupported parameter: max_tokens')
        gen_params = self.update_gen_params(**gen_params)
        gen_params['stream'] = True

        resp = ''
        finished = False
        stop_words = gen_params.get('stop_words')
        if stop_words is None:
            stop_words = []
        # mapping to role that openai supports
        messages = self.template_parser._prompt2api(inputs)
        for text in self._stream_chat(messages, **gen_params):
            if self.model_type.lower().startswith('qwen'):
                resp = text
            else:
                resp += text
            if not resp:
                continue
            # remove stop_words
            for sw in stop_words:
                if sw in resp:
                    resp = filter_suffix(resp, stop_words)
                    finished = True
                    break
            yield ModelStatusCode.STREAM_ING, resp, None
            if finished:
                break
        yield ModelStatusCode.END, resp, None

    def _chat(self, messages: List[dict], **gen_params) -> str:
        """Generate completion from a list of templates.

        Args:
            messages (List[dict]): a list of prompt dictionaries
            gen_params: additional generation configuration

        Returns:
            str: The generated string.
        """
        assert isinstance(messages, list)

        header, data = self.generate_request_data(
            model_type=self.model_type,
            messages=messages,
            gen_params=gen_params,
            json_mode=self.json_mode)

        max_num_retries, errmsg = 0, ''
        while max_num_retries < self.retry:
            with Lock():
                if len(self.invalid_keys) == len(self.keys):
                    raise RuntimeError('All keys have insufficient quota.')

                # find the next valid key
                while True:
                    self.key_ctr += 1
                    if self.key_ctr == len(self.keys):
                        self.key_ctr = 0

                    if self.keys[self.key_ctr] not in self.invalid_keys:
                        break

                key = self.keys[self.key_ctr]
                header['Authorization'] = f'Bearer {key}'

            if self.orgs:
                with Lock():
                    self.org_ctr += 1
                    if self.org_ctr == len(self.orgs):
                        self.org_ctr = 0
                header['OpenAI-Organization'] = self.orgs[self.org_ctr]

            response = dict()
            try:
                raw_response = requests.post(
                    self.url,
                    headers=header,
                    data=json.dumps(data),
                    proxies=self.proxies)
                response = raw_response.json()
                return response['choices'][0]['message']['content'].strip()
            except requests.ConnectionError:
                errmsg = 'Got connection error ' + str(traceback.format_exc())
                self.logger.error(errmsg)
                continue
            except requests.JSONDecodeError:
                errmsg = 'JsonDecode error, got ' + str(raw_response.content)
                self.logger.error(errmsg)
                continue
            except KeyError:
                if 'error' in response:
                    if response['error']['code'] == 'rate_limit_exceeded':
                        time.sleep(1)
                        continue
                    elif response['error']['code'] == 'insufficient_quota':
                        self.invalid_keys.add(key)
                        self.logger.warn(f'insufficient_quota key: {key}')
                        continue

                    errmsg = 'Find error message in response: ' + str(
                        response['error'])
                    self.logger.error(errmsg)
            except Exception as error:
                errmsg = str(error) + '\n' + str(traceback.format_exc())
                self.logger.error(errmsg)
            max_num_retries += 1

        raise RuntimeError('Calling OpenAI failed after retrying for '
                           f'{max_num_retries} times. Check the logs for '
                           f'details. errmsg: {errmsg}')

    def _stream_chat(self, messages: List[dict], **gen_params) -> str:
        """Generate completion from a list of templates.

        Args:
            messages (List[dict]): a list of prompt dictionaries
            gen_params: additional generation configuration

        Returns:
            str: The generated string.
        """

        def streaming(raw_response):
            for chunk in raw_response.iter_lines(
                    chunk_size=8192, decode_unicode=False, delimiter=b'\n'):
                if chunk:
                    decoded = chunk.decode('utf-8')
                    if decoded.startswith('data: [DONE]'):
                        return
                    if decoded[:5] == 'data:':
                        decoded = decoded[5:]
                        if decoded[0] == ' ':
                            decoded = decoded[1:]
                    else:
                        print(decoded)
                        continue
                    try:
                        response = json.loads(decoded)
                        if 'code' in response and response['code'] == -20003:
                            # Context exceeds maximum length
                            yield ''
                            return
                        if self.model_type.lower().startswith('qwen'):
                            choice = response['output']['choices'][0]
                            yield choice['message']['content']
                            if choice['finish_reason'] == 'stop':
                                return
                        else:
                            choice = response['choices'][0]
                            if choice['finish_reason'] == 'stop':
                                return
                            yield choice['delta'].get('content', '')
                    except Exception as exc:
                        msg = f'response {decoded} lead to exception of {str(exc)}'
                        self.logger.error(msg)
                        raise Exception(msg) from exc

        assert isinstance(messages, list)

        header, data = self.generate_request_data(
            model_type=self.model_type,
            messages=messages,
            gen_params=gen_params,
            json_mode=self.json_mode)

        max_num_retries, errmsg = 0, ''
        while max_num_retries < self.retry:
            if len(self.invalid_keys) == len(self.keys):
                raise RuntimeError('All keys have insufficient quota.')

            # find the next valid key
            while True:
                self.key_ctr += 1
                if self.key_ctr == len(self.keys):
                    self.key_ctr = 0

                if self.keys[self.key_ctr] not in self.invalid_keys:
                    break

            key = self.keys[self.key_ctr]
            header['Authorization'] = f'Bearer {key}'

            if self.orgs:
                self.org_ctr += 1
                if self.org_ctr == len(self.orgs):
                    self.org_ctr = 0
                header['OpenAI-Organization'] = self.orgs[self.org_ctr]

            response = dict()
            try:
                raw_response = requests.post(
                    self.url,
                    headers=header,
                    data=json.dumps(data),
                    proxies=self.proxies)
                return streaming(raw_response)
            except requests.ConnectionError:
                errmsg = 'Got connection error ' + str(traceback.format_exc())
                self.logger.error(errmsg)
                continue
            except requests.JSONDecodeError:
                errmsg = 'JsonDecode error, got ' + str(raw_response.content)
                self.logger.error(errmsg)
                continue
            except KeyError:
                if 'error' in response:
                    if response['error']['code'] == 'rate_limit_exceeded':
                        time.sleep(1)
                        continue
                    elif response['error']['code'] == 'insufficient_quota':
                        self.invalid_keys.add(key)
                        self.logger.warn(f'insufficient_quota key: {key}')
                        continue

                    errmsg = 'Find error message in response: ' + str(
                        response['error'])
                    self.logger.error(errmsg)
            except Exception as error:
                errmsg = str(error) + '\n' + str(traceback.format_exc())
                self.logger.error(errmsg)
            max_num_retries += 1

        raise RuntimeError('Calling OpenAI failed after retrying for '
                           f'{max_num_retries} times. Check the logs for '
                           f'details. errmsg: {errmsg}')

    def generate_request_data(self,
                              model_type,
                              messages,
                              gen_params,
                              json_mode=False):
        """
        Generates the request data for different model types.

        Args:
            model_type (str): The type of the model (e.g., 'gpt', 'internlm', 'qwen').
            messages (list): The list of messages to be sent to the model.
            gen_params (dict): The generation parameters.
            json_mode (bool): Flag to determine if the response format should be JSON.

        Returns:
            tuple: A tuple containing the header and the request data.
        """
        # Copy generation parameters to avoid modifying the original dictionary
        gen_params = gen_params.copy()

        # Hold out 100 tokens due to potential errors in token calculation
        max_tokens = min(gen_params.pop('max_new_tokens'), 4096)
        if max_tokens <= 0:
            return '', ''

        # Initialize the header
        header = {
            'content-type': 'application/json',
        }

        # Common parameters processing
        gen_params['max_tokens'] = max_tokens
        if 'stop_words' in gen_params:
            gen_params['stop'] = gen_params.pop('stop_words')
        if 'repetition_penalty' in gen_params:
            gen_params['frequency_penalty'] = gen_params.pop(
                'repetition_penalty')

        # Model-specific processing
        data = {}
        if model_type.lower().startswith('gpt'):
            if 'top_k' in gen_params:
                warnings.warn(
                    '`top_k` parameter is deprecated in OpenAI APIs.',
                    DeprecationWarning)
                gen_params.pop('top_k')
            gen_params.pop('skip_special_tokens', None)
            gen_params.pop('session_id', None)
            data = {
                'model': model_type,
                'messages': messages,
                'n': 1,
                **gen_params
            }
            if json_mode:
                data['response_format'] = {'type': 'json_object'}
        elif model_type.lower().startswith('internlm'):
            data = {
                'model': model_type,
                'messages': messages,
                'n': 1,
                **gen_params
            }
            if json_mode:
                data['response_format'] = {'type': 'json_object'}
        elif model_type.lower().startswith('qwen'):
            header['X-DashScope-SSE'] = 'enable'
            gen_params.pop('skip_special_tokens', None)
            gen_params.pop('session_id', None)
            if 'frequency_penalty' in gen_params:
                gen_params['repetition_penalty'] = gen_params.pop(
                    'frequency_penalty')
            gen_params['result_format'] = 'message'
            data = {
                'model': model_type,
                'input': {
                    'messages': messages
                },
                'parameters': {
                    **gen_params
                }
            }
        else:
            raise NotImplementedError(
                f'Model type {model_type} is not supported')

        return header, data

    def tokenize(self, prompt: str) -> list:
        """Tokenize the input prompt.

        Args:
            prompt (str): Input string.

        Returns:
            list: token ids
        """
        import tiktoken
        self.tiktoken = tiktoken
        enc = self.tiktoken.encoding_for_model(self.model_type)
        return enc.encode(prompt)


class AsyncGPTAPI(AsyncBaseAPILLM):
    """Model wrapper around OpenAI's models.

    Args:
        model_type (str): The name of OpenAI's model.
        retry (int): Number of retires if the API call fails. Defaults to 2.
        key (str or List[str]): OpenAI key(s). In particular, when it
            is set to "ENV", the key will be fetched from the environment
            variable $OPENAI_API_KEY, as how openai defaults to be. If it's a
            list, the keys will be used in round-robin manner. Defaults to
            'ENV'.
        org (str or List[str], optional): OpenAI organization(s). If not
            specified, OpenAI uses the default organization bound to each API
            key. If specified, the orgs will be posted with each request in
            round-robin manner. Defaults to None.
        meta_template (Dict, optional): The model's meta prompt
            template if needed, in case the requirement of injecting or
            wrapping of any meta instructions.
        api_base (str): The base url of OpenAI's API. Defaults to
            'https://api.openai.com/v1/chat/completions'.
        gen_params: Default generation configuration which could be overridden
            on the fly of generation.
    """

    is_api: bool = True

    def __init__(self,
                 model_type: str = 'gpt-3.5-turbo',
                 retry: int = 2,
                 json_mode: bool = False,
                 key: Union[str, List[str]] = 'ENV',
                 org: Optional[Union[str, List[str]]] = None,
                 meta_template: Optional[Dict] = [
                     dict(role='system', api_role='system'),
                     dict(role='user', api_role='user'),
                     dict(role='assistant', api_role='assistant')
                 ],
                 api_base: str = OPENAI_API_BASE,
                 proxies: Optional[Dict] = None,
                 **gen_params):
        if 'top_k' in gen_params:
            warnings.warn('`top_k` parameter is deprecated in OpenAI APIs.',
                          DeprecationWarning)
            gen_params.pop('top_k')
        super().__init__(
            model_type=model_type,
            meta_template=meta_template,
            retry=retry,
            **gen_params)
        self.gen_params.pop('top_k')
        self.logger = getLogger(__name__)

        if isinstance(key, str):
            self.keys = [os.getenv('OPENAI_API_KEY') if key == 'ENV' else key]
        else:
            self.keys = key

        # record invalid keys and skip them when requesting API
        # - keys have insufficient_quota
        self.invalid_keys = set()

        self.key_ctr = 0
        if isinstance(org, str):
            self.orgs = [org]
        else:
            self.orgs = org
        self.org_ctr = 0
        self.url = api_base
        self.model_type = model_type
        self.proxies = proxies or {}
        self.json_mode = json_mode

    async def chat(
        self,
        inputs: Union[List[dict], List[List[dict]]],
        session_ids: Union[int, List[int]] = None,
        **gen_params,
    ) -> Union[str, List[str]]:
        """Generate responses given the contexts.

        Args:
            inputs (Union[List[dict], List[List[dict]]]): a list of messages
                or list of lists of messages
            gen_params: additional generation configuration

        Returns:
            Union[str, List[str]]: generated string(s)
        """
        assert isinstance(inputs, list)
        if 'max_tokens' in gen_params:
            raise NotImplementedError('unsupported parameter: max_tokens')
        gen_params = {**self.gen_params, **gen_params}
        tasks = [
            self._chat(messages, **gen_params) for messages in (
                [inputs] if isinstance(inputs[0], dict) else inputs)
        ]
        ret = await asyncio.gather(*tasks)
        return ret[0] if isinstance(inputs[0], dict) else ret

    async def stream_chat(
        self,
        inputs: List[dict],
        **gen_params,
    ):
        """Generate responses given the contexts.

        Args:
            inputs (List[dict]): a list of messages
            gen_params: additional generation configuration

        Returns:
            str: generated string
        """
        assert isinstance(inputs, list)
        if 'max_tokens' in gen_params:
            raise NotImplementedError('unsupported parameter: max_tokens')
        gen_params = self.update_gen_params(**gen_params)
        gen_params['stream'] = True

        resp = ''
        finished = False
        stop_words = gen_params.get('stop_words')
        if stop_words is None:
            stop_words = []
        # mapping to role that openai supports
        messages = self.template_parser._prompt2api(inputs)
        async for text in self._stream_chat(messages, **gen_params):
            if self.model_type.lower().startswith('qwen'):
                resp = text
            else:
                resp += text
            if not resp:
                continue
            # remove stop_words
            for sw in stop_words:
                if sw in resp:
                    resp = filter_suffix(resp, stop_words)
                    finished = True
                    break
            yield ModelStatusCode.STREAM_ING, resp, None
            if finished:
                break
        yield ModelStatusCode.END, resp, None

    async def _chat(self, messages: List[dict], **gen_params) -> str:
        """Generate completion from a list of templates.

        Args:
            messages (List[dict]): a list of prompt dictionaries
            gen_params: additional generation configuration

        Returns:
            str: The generated string.
        """
        assert isinstance(messages, list)

        header, data = self.generate_request_data(
            model_type=self.model_type,
            messages=messages,
            gen_params=gen_params,
            json_mode=self.json_mode)

        max_num_retries, errmsg = 0, ''
        while max_num_retries < self.retry:
            if len(self.invalid_keys) == len(self.keys):
                raise RuntimeError('All keys have insufficient quota.')

            # find the next valid key
            while True:
                self.key_ctr += 1
                if self.key_ctr == len(self.keys):
                    self.key_ctr = 0

                if self.keys[self.key_ctr] not in self.invalid_keys:
                    break

            key = self.keys[self.key_ctr]
            header['Authorization'] = f'Bearer {key}'

            if self.orgs:
                self.org_ctr += 1
                if self.org_ctr == len(self.orgs):
                    self.org_ctr = 0
                header['OpenAI-Organization'] = self.orgs[self.org_ctr]

            response = dict()
            try:
                async with aiohttp.ClientSession() as session:
                    async with session.post(
                            self.url,
                            headers=header,
                            json=data,
                            proxy=self.proxies.get(
                                'https', self.proxies.get('http'))) as resp:
                        response = await resp.json()
                        return response['choices'][0]['message'][
                            'content'].strip()
            except aiohttp.ClientConnectionError:
                errmsg = 'Got connection error ' + str(traceback.format_exc())
                self.logger.error(errmsg)
                continue
            except aiohttp.ClientResponseError as e:
                errmsg = 'Response error, got ' + str(e)
                self.logger.error(errmsg)
                continue
            except json.JSONDecodeError:
                errmsg = 'JsonDecode error, got ' + (await resp.text(
                    errors='replace'))
                self.logger.error(errmsg)
                continue
            except KeyError:
                if 'error' in response:
                    if response['error']['code'] == 'rate_limit_exceeded':
                        time.sleep(1)
                        continue
                    elif response['error']['code'] == 'insufficient_quota':
                        self.invalid_keys.add(key)
                        self.logger.warn(f'insufficient_quota key: {key}')
                        continue

                    errmsg = 'Find error message in response: ' + str(
                        response['error'])
                    self.logger.error(errmsg)
            except Exception as error:
                errmsg = str(error) + '\n' + str(traceback.format_exc())
                self.logger.error(errmsg)
            max_num_retries += 1

        raise RuntimeError('Calling OpenAI failed after retrying for '
                           f'{max_num_retries} times. Check the logs for '
                           f'details. errmsg: {errmsg}')

    async def _stream_chat(self, messages: List[dict],
                           **gen_params) -> AsyncGenerator[str, None]:
        """Generate completion from a list of templates.

        Args:
            messages (List[dict]): a list of prompt dictionaries
            gen_params: additional generation configuration

        Returns:
            str: The generated string.
        """

        async def streaming(raw_response):
            async for chunk in raw_response.content:
                if chunk:
                    decoded = chunk.decode('utf-8')
                    if decoded.startswith('data: [DONE]'):
                        return
                    if decoded[:5] == 'data:':
                        decoded = decoded[5:]
                        if decoded[0] == ' ':
                            decoded = decoded[1:]
                    else:
                        print(decoded)
                        continue
                    try:
                        response = json.loads(decoded)
                        if 'code' in response and response['code'] == -20003:
                            # Context exceeds maximum length
                            yield ''
                            return
                        if self.model_type.lower().startswith('qwen'):
                            choice = response['output']['choices'][0]
                            yield choice['message']['content']
                            if choice['finish_reason'] == 'stop':
                                return
                        else:
                            choice = response['choices'][0]
                            if choice['finish_reason'] == 'stop':
                                return
                            yield choice['delta'].get('content', '')
                    except Exception as exc:
                        msg = f'response {decoded} lead to exception of {str(exc)}'
                        self.logger.error(msg)
                        raise Exception(msg) from exc

        assert isinstance(messages, list)

        header, data = self.generate_request_data(
            model_type=self.model_type,
            messages=messages,
            gen_params=gen_params,
            json_mode=self.json_mode)

        max_num_retries, errmsg = 0, ''
        while max_num_retries < self.retry:
            if len(self.invalid_keys) == len(self.keys):
                raise RuntimeError('All keys have insufficient quota.')

            # find the next valid key
            while True:
                self.key_ctr += 1
                if self.key_ctr == len(self.keys):
                    self.key_ctr = 0

                if self.keys[self.key_ctr] not in self.invalid_keys:
                    break

            key = self.keys[self.key_ctr]
            header['Authorization'] = f'Bearer {key}'

            if self.orgs:
                self.org_ctr += 1
                if self.org_ctr == len(self.orgs):
                    self.org_ctr = 0
                header['OpenAI-Organization'] = self.orgs[self.org_ctr]

            response = dict()
            try:
                async with aiohttp.ClientSession() as session:
                    async with session.post(
                            self.url,
                            headers=header,
                            json=data,
                            proxy=self.proxies.get(
                                'https',
                                self.proxies.get('http'))) as raw_response:
                        async for msg in streaming(raw_response):
                            yield msg
                        return
            except aiohttp.ClientConnectionError:
                errmsg = 'Got connection error ' + str(traceback.format_exc())
                self.logger.error(errmsg)
                continue
            except aiohttp.ClientResponseError as e:
                errmsg = 'Response error, got ' + str(e)
                self.logger.error(errmsg)
                continue
            except KeyError:
                if 'error' in response:
                    if response['error']['code'] == 'rate_limit_exceeded':
                        time.sleep(1)
                        continue
                    elif response['error']['code'] == 'insufficient_quota':
                        self.invalid_keys.add(key)
                        self.logger.warn(f'insufficient_quota key: {key}')
                        continue

                    errmsg = 'Find error message in response: ' + str(
                        response['error'])
                    self.logger.error(errmsg)
            except Exception as error:
                errmsg = str(error) + '\n' + str(traceback.format_exc())
                self.logger.error(errmsg)
            max_num_retries += 1

        raise RuntimeError('Calling OpenAI failed after retrying for '
                           f'{max_num_retries} times. Check the logs for '
                           f'details. errmsg: {errmsg}')

    def generate_request_data(self,
                              model_type,
                              messages,
                              gen_params,
                              json_mode=False):
        """
        Generates the request data for different model types.

        Args:
            model_type (str): The type of the model (e.g., 'gpt', 'internlm', 'qwen').
            messages (list): The list of messages to be sent to the model.
            gen_params (dict): The generation parameters.
            json_mode (bool): Flag to determine if the response format should be JSON.

        Returns:
            tuple: A tuple containing the header and the request data.
        """
        # Copy generation parameters to avoid modifying the original dictionary
        gen_params = gen_params.copy()

        # Hold out 100 tokens due to potential errors in token calculation
        max_tokens = min(gen_params.pop('max_new_tokens'), 4096)
        if max_tokens <= 0:
            return '', ''

        # Initialize the header
        header = {
            'content-type': 'application/json',
        }

        # Common parameters processing
        gen_params['max_tokens'] = max_tokens
        if 'stop_words' in gen_params:
            gen_params['stop'] = gen_params.pop('stop_words')
        if 'repetition_penalty' in gen_params:
            gen_params['frequency_penalty'] = gen_params.pop(
                'repetition_penalty')

        # Model-specific processing
        data = {}
        if model_type.lower().startswith('gpt'):
            if 'top_k' in gen_params:
                warnings.warn(
                    '`top_k` parameter is deprecated in OpenAI APIs.',
                    DeprecationWarning)
                gen_params.pop('top_k')
            gen_params.pop('skip_special_tokens', None)
            gen_params.pop('session_id', None)
            data = {
                'model': model_type,
                'messages': messages,
                'n': 1,
                **gen_params
            }
            if json_mode:
                data['response_format'] = {'type': 'json_object'}
        elif model_type.lower().startswith('internlm'):
            data = {
                'model': model_type,
                'messages': messages,
                'n': 1,
                **gen_params
            }
            if json_mode:
                data['response_format'] = {'type': 'json_object'}
        elif model_type.lower().startswith('qwen'):
            header['X-DashScope-SSE'] = 'enable'
            gen_params.pop('skip_special_tokens', None)
            gen_params.pop('session_id', None)
            if 'frequency_penalty' in gen_params:
                gen_params['repetition_penalty'] = gen_params.pop(
                    'frequency_penalty')
            gen_params['result_format'] = 'message'
            data = {
                'model': model_type,
                'input': {
                    'messages': messages
                },
                'parameters': {
                    **gen_params
                }
            }
        else:
            raise NotImplementedError(
                f'Model type {model_type} is not supported')

        return header, data

    def tokenize(self, prompt: str) -> list:
        """Tokenize the input prompt.

        Args:
            prompt (str): Input string.

        Returns:
            list: token ids
        """
        import tiktoken
        self.tiktoken = tiktoken
        enc = self.tiktoken.encoding_for_model(self.model_type)
        return enc.encode(prompt)