File size: 37,286 Bytes
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
1e8c453
1e6e9f4
 
 
1e8c453
 
 
edf6dca
1e8c453
 
 
 
 
 
 
eeb7ca1
edf6dca
eeb7ca1
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
1e8c453
edf6dca
 
 
1e8c453
edf6dca
2ce9a1a
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
 
 
 
 
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
 
eeb7ca1
 
 
 
 
1e8c453
 
 
 
 
 
 
 
 
 
 
 
 
 
eeb7ca1
54f4f91
eeb7ca1
 
 
 
 
1e8c453
eeb7ca1
 
 
 
1e8c453
eeb7ca1
1e8c453
 
 
 
eeb7ca1
 
 
1e8c453
 
 
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
 
eeb7ca1
1e8c453
eeb7ca1
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
1e8c453
 
eeb7ca1
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
eeb7ca1
 
 
 
 
1e8c453
edf6dca
1e8c453
 
 
 
 
 
 
 
 
 
 
 
 
 
 
edf6dca
1e8c453
eeb7ca1
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
eeb7ca1
 
 
 
 
 
1e8c453
 
eeb7ca1
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
1e8c453
 
eeb7ca1
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
1e8c453
 
eeb7ca1
 
 
 
 
 
1e8c453
 
 
 
 
 
 
 
 
 
 
 
 
eeb7ca1
 
 
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
1e8c453
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
edf6dca
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b368114
 
 
 
 
 
 
 
 
 
 
 
 
 
2ce9a1a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eeb7ca1
 
 
1e8c453
 
 
 
eeb7ca1
1e8c453
 
 
 
 
 
eeb7ca1
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
1e8c453
 
 
eeb7ca1
1e8c453
 
eeb7ca1
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
 
1e8c453
eeb7ca1
 
 
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
 
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
 
1e8c453
 
eeb7ca1
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eeb7ca1
1e8c453
 
 
 
 
 
 
 
eeb7ca1
 
 
1e8c453
eeb7ca1
 
 
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
1e8c453
 
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54f4f91
 
 
eeb7ca1
1e8c453
eeb7ca1
 
1e8c453
eeb7ca1
1e8c453
eeb7ca1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
eeb7ca1
 
 
 
 
 
 
1e8c453
eeb7ca1
1e8c453
eeb7ca1
 
 
 
 
 
b368114
eeb7ca1
 
 
 
1e8c453
eeb7ca1
b368114
 
 
 
 
 
 
 
 
 
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
import os
import ast
import time
from enums import PromptType  # also supports imports from this file from other files

non_hf_types = ['gpt4all_llama', 'llama', 'gptj']

prompt_type_to_model_name = {
    'plain': [
        'EleutherAI/gpt-j-6B',
        'EleutherAI/pythia-6.9b',
        'EleutherAI/pythia-12b',
        'EleutherAI/pythia-12b-deduped',
        'EleutherAI/gpt-neox-20b',
        'openlm-research/open_llama_7b_700bt_preview',
        'decapoda-research/llama-7b-hf',
        'decapoda-research/llama-13b-hf',
        'decapoda-research/llama-30b-hf',
        'decapoda-research/llama-65b-hf',
        'facebook/mbart-large-50-many-to-many-mmt',
        'philschmid/bart-large-cnn-samsum',
        'philschmid/flan-t5-base-samsum',
        'gpt2',
        'distilgpt2',
        'mosaicml/mpt-7b-storywriter',
    ],
    'gptj': ['gptj', 'gpt4all_llama'],
    'prompt_answer': [
        'h2oai/h2ogpt-gm-oasst1-en-1024-20b',
        'h2oai/h2ogpt-gm-oasst1-en-1024-12b',
        'h2oai/h2ogpt-gm-oasst1-multilang-1024-20b',
        'h2oai/h2ogpt-gm-oasst1-multilang-2048-falcon-7b',
        'h2oai/h2ogpt-gm-oasst1-multilang-2048-falcon-7b-v2',
        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v3',
        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b',
        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v2',
        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-40b-v1',
        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-40b-v2',
        'h2oai/h2ogpt-gm-oasst1-en-xgen-7b-8k',
        'h2oai/h2ogpt-gm-oasst1-multilang-xgen-7b-8k',
        'TheBloke/h2ogpt-gm-oasst1-en-2048-falcon-40b-v2-GPTQ',
    ],
    'prompt_answer_openllama': [
        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt',
        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt-v2',
        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-700bt',
        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b',
        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-13b',
    ],
    'instruct': ['TheBloke/llama-30b-supercot-SuperHOT-8K-fp16'],  # https://huggingface.co/TheBloke/llama-30b-supercot-SuperHOT-8K-fp16#prompting
    'instruct_with_end': ['databricks/dolly-v2-12b'],
    'quality': [],
    'human_bot': [
        'h2oai/h2ogpt-oasst1-512-12b',
        'h2oai/h2ogpt-oasst1-512-20b',
        'h2oai/h2ogpt-oig-oasst1-256-6_9b',
        'h2oai/h2ogpt-oig-oasst1-512-6_9b',
        'h2oai/h2ogpt-oig-oasst1-256-6.9b',  # legacy
        'h2oai/h2ogpt-oig-oasst1-512-6.9b',  # legacy
        'h2oai/h2ogpt-research-oasst1-512-30b',
        'h2oai/h2ogpt-research-oasst1-llama-65b',
        'h2oai/h2ogpt-oasst1-falcon-40b',
        'h2oai/h2ogpt-oig-oasst1-falcon-40b',
    ],
    'dai_faq': [],
    'summarize': [],
    'simple_instruct': ['t5-small', 't5-large', 'google/flan-t5', 'google/flan-t5-xxl', 'google/flan-ul2'],
    'instruct_vicuna': ['AlekseyKorshuk/vicuna-7b', 'TheBloke/stable-vicuna-13B-HF', 'junelee/wizard-vicuna-13b'],
    'human_bot_orig': ['togethercomputer/GPT-NeoXT-Chat-Base-20B'],
    "open_assistant": ['OpenAssistant/oasst-sft-7-llama-30b-xor', 'oasst-sft-7-llama-30b'],
    "wizard_lm": ['ehartford/WizardLM-7B-Uncensored', 'ehartford/WizardLM-13B-Uncensored'],
    "wizard_mega": ['openaccess-ai-collective/wizard-mega-13b'],
    "instruct_simple": ['JosephusCheung/Guanaco'],
    "wizard_vicuna": ['ehartford/Wizard-Vicuna-13B-Uncensored'],
    "wizard2": ['llama'],
    "mptinstruct": ['mosaicml/mpt-30b-instruct', 'mosaicml/mpt-7b-instruct', 'mosaicml/mpt-30b-instruct'],
    "mptchat": ['mosaicml/mpt-7b-chat', 'mosaicml/mpt-30b-chat', 'TheBloke/mpt-30B-chat-GGML'],
    "vicuna11": ['lmsys/vicuna-33b-v1.3'],
    "falcon": ['tiiuae/falcon-40b-instruct', 'tiiuae/falcon-40b', 'tiiuae/falcon-7b-instruct', 'tiiuae/falcon-7b'],
    "llama2": [
        'meta-llama/Llama-2-7b-chat-hf',
        'meta-llama/Llama-2-13b-chat-hf',
        'meta-llama/Llama-2-34b-chat-hf',
        'meta-llama/Llama-2-70b-chat-hf',
    ],
    # could be plain, but default is correct prompt_type for default TheBloke model ggml-wizardLM-7B.q4_2.bin
}
if os.getenv('OPENAI_API_KEY'):
    prompt_type_to_model_name.update({
        "openai": ["text-davinci-003", "text-curie-001", "text-babbage-001", "text-ada-001"],
        "openai_chat": ["gpt-3.5-turbo", "gpt-3.5-turbo-16k"],
    })

inv_prompt_type_to_model_name = {v.strip(): k for k, l in prompt_type_to_model_name.items() for v in l}
inv_prompt_type_to_model_lower = {v.strip().lower(): k for k, l in prompt_type_to_model_name.items() for v in l}

prompt_types_strings = []
for p in PromptType:
    prompt_types_strings.extend([p.name])

prompt_types = []
for p in PromptType:
    prompt_types.extend([p.name, p.value, str(p.value)])


def get_prompt(prompt_type, prompt_dict, chat, context, reduced, making_context, return_dict=False):
    prompt_dict_error = ''
    generates_leading_space = False

    if prompt_type == PromptType.custom.name and not isinstance(prompt_dict, dict):
        try:
            prompt_dict = ast.literal_eval(prompt_dict)
        except BaseException as e:
            prompt_dict_error = str(e)
    if prompt_dict_error:
        promptA = None
        promptB = None
        PreInstruct = None
        PreInput = ''
        PreResponse = ''
        terminate_response = None
        chat_sep = ''
        chat_turn_sep = ''
        humanstr = ''
        botstr = ''
        generates_leading_space = False
    elif prompt_type in [PromptType.custom.value, str(PromptType.custom.value),
                         PromptType.custom.name]:
        promptA = prompt_dict.get('promptA', '')
        promptB = prompt_dict.get('promptB', '')
        PreInstruct = prompt_dict.get('PreInstruct', '')
        PreInput = prompt_dict.get('PreInput', '')
        PreResponse = prompt_dict.get('PreResponse', '')
        terminate_response = prompt_dict.get('terminate_response', None)
        chat_sep = prompt_dict.get('chat_sep', '\n')
        chat_turn_sep = prompt_dict.get('chat_turn_sep', '\n')
        humanstr = prompt_dict.get('humanstr', '')
        botstr = prompt_dict.get('botstr', '')
    elif prompt_type in [PromptType.plain.value, str(PromptType.plain.value),
                         PromptType.plain.name]:
        promptA = promptB = PreInstruct = PreInput = PreResponse = None
        terminate_response = []
        chat_turn_sep = chat_sep = ''
        # plain should have None for human/bot, so nothing truncated out, not '' that would truncate after first token
        humanstr = None
        botstr = None
    elif prompt_type == 'simple_instruct':
        promptA = promptB = PreInstruct = PreInput = PreResponse = None
        terminate_response = []
        chat_turn_sep = chat_sep = '\n'
        humanstr = None
        botstr = None
    elif prompt_type in [PromptType.instruct.value, str(PromptType.instruct.value),
                         PromptType.instruct.name] + [PromptType.instruct_with_end.value,
                                                      str(PromptType.instruct_with_end.value),
                                                      PromptType.instruct_with_end.name]:
        promptA = 'Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n' if not (
                chat and reduced) else ''
        promptB = 'Below is an instruction that describes a task. Write a response that appropriately completes the request.\n' if not (
                chat and reduced) else ''

        PreInstruct = """
### Instruction:
"""

        PreInput = """
### Input:
"""

        PreResponse = """
### Response:
"""
        if prompt_type in [PromptType.instruct_with_end.value, str(PromptType.instruct_with_end.value),
                           PromptType.instruct_with_end.name]:
            terminate_response = ['### End']
        else:
            terminate_response = None
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.quality.value, str(PromptType.quality.value),
                         PromptType.quality.name]:
        promptA = 'Write a detailed high-quality, accurate, fair, Response with about 100 words by following the Instruction as applied on the Input.\n' if not (
                chat and reduced) else ''
        promptB = 'Write a detailed high-quality, accurate, fair, Response with about 100 words by following the Instruction.\n' if not (
                chat and reduced) else ''

        PreInstruct = """
### Instruction:
"""

        PreInput = """
### Input:
"""

        PreResponse = """
### Response:
"""
        terminate_response = None
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct  # first thing human says
        botstr = PreResponse  # first thing bot says
    elif prompt_type in [PromptType.human_bot.value, str(PromptType.human_bot.value),
                         PromptType.human_bot.name] + [PromptType.human_bot_orig.value,
                                                       str(PromptType.human_bot_orig.value),
                                                       PromptType.human_bot_orig.name]:
        human = '<human>:'
        bot = "<bot>:"
        if reduced or context or prompt_type in [PromptType.human_bot.value, str(PromptType.human_bot.value),
                                                 PromptType.human_bot.name]:
            preprompt = ''
        else:
            cur_date = time.strftime('%Y-%m-%d')
            cur_time = time.strftime('%H:%M:%S %p %Z')

            PRE_PROMPT = """\
Current Date: {}
Current Time: {}

"""
            preprompt = PRE_PROMPT.format(cur_date, cur_time)
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)

        PreInstruct = human + ' '

        PreInput = None

        if making_context:
            # when making context, want it to appear as-if LLM generated, which starts with space after :
            PreResponse = bot + ' '
        else:
            # normally LLM adds space after this, because was how trained.
            # if add space here, non-unique tokenization will often make LLM produce wrong output
            PreResponse = bot

        terminate_response = ['\n' + human, '\n' + bot, human, bot, PreResponse]
        chat_turn_sep = chat_sep = '\n'
        humanstr = human  # tag before human talks
        botstr = bot  # tag before bot talks
        generates_leading_space = True
    elif prompt_type in [PromptType.dai_faq.value, str(PromptType.dai_faq.value),
                         PromptType.dai_faq.name]:
        promptA = ''
        promptB = 'Answer the following Driverless AI question.\n'

        PreInstruct = """
### Driverless AI frequently asked question:
"""

        PreInput = None

        PreResponse = """
### Driverless AI documentation answer:
"""
        terminate_response = ['\n\n']
        chat_turn_sep = chat_sep = terminate_response
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.summarize.value, str(PromptType.summarize.value),
                         PromptType.summarize.name]:
        promptA = promptB = PreInput = ''
        PreInstruct = '## Main Text\n\n'
        PreResponse = '\n\n## Summary\n\n'
        terminate_response = None
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.instruct_vicuna.value, str(PromptType.instruct_vicuna.value),
                         PromptType.instruct_vicuna.name]:
        promptA = promptB = "A chat between a curious human and an artificial intelligence assistant. " \
                            "The assistant gives helpful, detailed, and polite answers to the human's questions." if not (
                chat and reduced) else ''

        PreInstruct = """
### Human:
"""

        PreInput = None

        PreResponse = """
### Assistant:
"""
        terminate_response = [
            '### Human:']  # but only allow terminate after prompt is found correctly, else can't terminate
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.prompt_answer.value, str(PromptType.prompt_answer.value),
                         PromptType.prompt_answer.name]:
        preprompt = ''
        prompt_tokens = "<|prompt|>"
        answer_tokens = "<|answer|>"
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = prompt_tokens
        PreInput = None
        PreResponse = answer_tokens
        eos = '<|endoftext|>'  # neox eos
        humanstr = prompt_tokens
        botstr = answer_tokens
        terminate_response = [humanstr, PreResponse, eos]
        chat_sep = eos
        chat_turn_sep = eos
    elif prompt_type in [PromptType.prompt_answer_openllama.value, str(PromptType.prompt_answer_openllama.value),
                         PromptType.prompt_answer_openllama.name]:
        preprompt = ''
        prompt_tokens = "<|prompt|>"
        answer_tokens = "<|answer|>"
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = prompt_tokens
        PreInput = None
        PreResponse = answer_tokens
        eos = '</s>'  # llama eos
        humanstr = prompt_tokens
        botstr = answer_tokens
        terminate_response = [humanstr, PreResponse, eos]
        chat_sep = eos
        chat_turn_sep = eos
    elif prompt_type in [PromptType.open_assistant.value, str(PromptType.open_assistant.value),
                         PromptType.open_assistant.name]:
        # From added_tokens.json
        preprompt = ''
        prompt_tokens = "<|prompter|>"
        answer_tokens = "<|assistant|>"
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = prompt_tokens
        PreInput = None
        PreResponse = answer_tokens
        pend = "<|prefix_end|>"
        eos = "</s>"
        humanstr = prompt_tokens
        botstr = answer_tokens
        terminate_response = [humanstr, PreResponse, pend, eos]
        chat_turn_sep = chat_sep = eos
    elif prompt_type in [PromptType.wizard_lm.value, str(PromptType.wizard_lm.value),
                         PromptType.wizard_lm.name]:
        # https://github.com/ehartford/WizardLM/blob/main/src/train_freeform.py
        preprompt = ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = ""
        PreInput = None
        PreResponse = "\n\n### Response\n"
        eos = "</s>"
        terminate_response = [PreResponse, eos]
        chat_turn_sep = chat_sep = eos
        humanstr = promptA
        botstr = PreResponse
    elif prompt_type in [PromptType.wizard_mega.value, str(PromptType.wizard_mega.value),
                         PromptType.wizard_mega.name]:
        preprompt = ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = """
### Instruction:
"""
        PreInput = None
        PreResponse = """
### Assistant:
"""
        terminate_response = [PreResponse]
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.instruct_vicuna2.value, str(PromptType.instruct_vicuna2.value),
                         PromptType.instruct_vicuna2.name]:
        promptA = promptB = "" if not (chat and reduced) else ''

        PreInstruct = """
HUMAN:
"""

        PreInput = None

        PreResponse = """
ASSISTANT:
"""
        terminate_response = [
            'HUMAN:']  # but only allow terminate after prompt is found correctly, else can't terminate
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.instruct_vicuna3.value, str(PromptType.instruct_vicuna3.value),
                         PromptType.instruct_vicuna3.name]:
        promptA = promptB = "" if not (chat and reduced) else ''

        PreInstruct = """
### User:
"""

        PreInput = None

        PreResponse = """
### Assistant:
"""
        terminate_response = [
            '### User:']  # but only allow terminate after prompt is found correctly, else can't terminate
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.wizard2.value, str(PromptType.wizard2.value),
                         PromptType.wizard2.name]:
        # https://huggingface.co/TheBloke/WizardLM-7B-uncensored-GGML
        preprompt = """Below is an instruction that describes a task. Write a response that appropriately completes the request.""" if not (
                chat and reduced) else ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = """
### Instruction:
"""
        PreInput = None
        PreResponse = """
### Response:
"""
        terminate_response = [PreResponse]
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.wizard3.value, str(PromptType.wizard3.value),
                         PromptType.wizard3.name]:
        # https://huggingface.co/TheBloke/wizardLM-13B-1.0-GGML
        preprompt = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.""" if not (
                chat and reduced) else ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = """USER: """
        PreInput = None
        PreResponse = """ASSISTANT: """
        terminate_response = [PreResponse]
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.wizard_vicuna.value, str(PromptType.wizard_vicuna.value),
                         PromptType.wizard_vicuna.name]:
        preprompt = ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = """USER: """
        PreInput = None
        PreResponse = """ASSISTANT: """
        terminate_response = [PreResponse]
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse

    elif prompt_type in [PromptType.instruct_simple.value, str(PromptType.instruct_simple.value),
                         PromptType.instruct_simple.name]:
        promptB = promptA = '' if not (chat and reduced) else ''

        PreInstruct = """
### Instruction:
"""

        PreInput = """
### Input:
"""

        PreResponse = """
### Response:
"""
        terminate_response = None
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.openai.value, str(PromptType.openai.value),
                         PromptType.openai.name]:
        preprompt = """The following is a conversation with an AI assistant. The assistant is helpful, creative, clever, and very friendly.""" if not (
                chat and reduced) else ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = "\nHuman: "
        PreInput = None
        PreResponse = "\nAI:"
        terminate_response = [PreResponse] + [" Human:", " AI:"]
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.gptj.value, str(PromptType.gptj.value),
                         PromptType.gptj.name]:
        preprompt = "### Instruction:\n The prompt below is a question to answer, a task to complete, or a conversation to respond to; decide which and write an appropriate response." if not (
                chat and reduced) else ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = "\n### Prompt: "
        PreInput = None
        PreResponse = "\n### Response: "
        terminate_response = [PreResponse] + ["Prompt:", "Response:"]
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.openai_chat.value, str(PromptType.openai_chat.value),
                         PromptType.openai_chat.name]:
        # prompting and termination all handled by endpoint
        preprompt = """"""
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        PreInstruct = ""
        PreInput = None
        PreResponse = ""
        terminate_response = []
        chat_turn_sep = chat_sep = '\n'
        humanstr = None
        botstr = None
    elif prompt_type in [PromptType.vicuna11.value, str(PromptType.vicuna11.value),
                         PromptType.vicuna11.name]:
        preprompt = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. """ if not (
                chat and reduced) else ''
        start = ''
        promptB = promptA = '%s%s' % (preprompt, start)
        eos = '</s>'
        PreInstruct = """USER: """
        PreInput = None
        PreResponse = """ASSISTANT:"""
        terminate_response = [PreResponse]
        chat_sep = ' '
        chat_turn_sep = eos
        humanstr = PreInstruct
        botstr = PreResponse

        if making_context:
            # when making context, want it to appear as-if LLM generated, which starts with space after :
            PreResponse = PreResponse + ' '
        else:
            # normally LLM adds space after this, because was how trained.
            # if add space here, non-unique tokenization will often make LLM produce wrong output
            PreResponse = PreResponse
    elif prompt_type in [PromptType.mptinstruct.value, str(PromptType.mptinstruct.value),
                         PromptType.mptinstruct.name]:
        # https://huggingface.co/mosaicml/mpt-30b-instruct#formatting
        promptA = promptB = 'Below is an instruction that describes a task. Write a response that appropriately completes the request.\n' if not (
                chat and reduced) else ''

        PreInstruct = """
### Instruction
"""

        PreInput = """
### Input
"""

        PreResponse = """
### Response
"""
        terminate_response = None
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.mptchat.value, str(PromptType.mptchat.value),
                         PromptType.mptchat.name]:
        # https://huggingface.co/TheBloke/mpt-30B-chat-GGML#prompt-template
        promptA = promptB = """<|im_start|>system\nA conversation between a user and an LLM-based AI assistant. The assistant gives helpful and honest answers.\n<|im_end|>""" if not (
                chat and reduced) else ''

        PreInstruct = """<|im_start|>user
"""

        PreInput = None

        PreResponse = """<|im_end|><|im_start|>assistant
"""
        terminate_response = ['<|im_end|>']
        chat_sep = ''
        chat_turn_sep = '<|im_end|>'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.falcon.value, str(PromptType.falcon.value),
                         PromptType.falcon.name]:
        promptA = promptB = "" if not (chat and reduced) else ''

        PreInstruct = """User: """

        PreInput = None

        PreResponse = """Assistant:"""
        terminate_response = ['\nUser', "<|endoftext|>"]
        chat_sep = '\n\n'
        chat_turn_sep = '\n\n'
        humanstr = PreInstruct
        botstr = PreResponse
        if making_context:
            # when making context, want it to appear as-if LLM generated, which starts with space after :
            PreResponse = 'Assistant: '
        else:
            # normally LLM adds space after this, because was how trained.
            # if add space here, non-unique tokenization will often make LLM produce wrong output
            PreResponse = PreResponse
        # generates_leading_space = True
    elif prompt_type in [PromptType.guanaco.value, str(PromptType.guanaco.value),
                         PromptType.guanaco.name]:
        # https://huggingface.co/TheBloke/guanaco-65B-GPTQ
        promptA = promptB = "" if not (chat and reduced) else ''

        PreInstruct = """### Human: """

        PreInput = None

        PreResponse = """### Assistant:"""
        terminate_response = ['### Human:']  # but only allow terminate after prompt is found correctly, else can't terminate
        chat_turn_sep = chat_sep = '\n'
        humanstr = PreInstruct
        botstr = PreResponse
    elif prompt_type in [PromptType.llama2.value, str(PromptType.llama2.value),
                         PromptType.llama2.name]:
        PreInstruct = ""
        llama2_sys = "<<SYS>>\nYou are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe.  Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n\nIf a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\n<</SYS>>\n\n"
        prompt = "<s>[INST] "
        enable_sys = False  # too much safety, hurts accuracy
        if not (chat and reduced):
            if enable_sys:
                promptA = promptB = prompt + llama2_sys
            else:
                promptA = promptB = prompt
        else:
            promptA = promptB = ''
        PreInput = None
        PreResponse = ""
        terminate_response = ["[INST]", "</s>"]
        chat_sep = ' [/INST]'
        chat_turn_sep = ' </s><s>[INST] '
        humanstr = PreInstruct
        botstr = PreResponse
        if making_context:
            PreResponse += " "
    else:
        raise RuntimeError("No such prompt_type=%s" % prompt_type)

    if isinstance(terminate_response, (tuple, list)):
        assert '' not in terminate_response, "Bad terminate_response"

    ret_dict = dict(promptA=promptA, promptB=promptB, PreInstruct=PreInstruct, PreInput=PreInput,
                    PreResponse=PreResponse, terminate_response=terminate_response, chat_sep=chat_sep,
                    chat_turn_sep=chat_turn_sep,
                    humanstr=humanstr, botstr=botstr,
                    generates_leading_space=generates_leading_space)

    if return_dict:
        return ret_dict, prompt_dict_error
    else:
        return tuple(list(ret_dict.values()))


def generate_prompt(data_point, prompt_type, prompt_dict, chat, reduced, making_context):
    context = data_point.get('context')
    if context is None:
        context = ''
    instruction = data_point.get('instruction')
    input = data_point.get('input')
    output = data_point.get('output')
    prompt_type = data_point.get('prompt_type', prompt_type)
    prompt_dict = data_point.get('prompt_dict', prompt_dict)
    assert prompt_type in prompt_types, "Bad prompt type: %s" % prompt_type
    promptA, promptB, PreInstruct, PreInput, PreResponse, \
        terminate_response, chat_sep, chat_turn_sep, humanstr, botstr, \
        generates_leading_space = get_prompt(prompt_type, prompt_dict, chat,
                                             context, reduced, making_context)

    # could avoid if reduce=True, but too complex for parent functions to handle
    prompt = context

    if input and promptA:
        prompt += f"""{promptA}"""
    elif promptB:
        prompt += f"""{promptB}"""

    if instruction and PreInstruct is not None and input and PreInput is not None:
        prompt += f"""{PreInstruct}{instruction}{PreInput}{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif instruction and input and PreInstruct is None and PreInput is not None:
        prompt += f"""{PreInput}{instruction}
{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif input and instruction and PreInput is None and PreInstruct is not None:
        prompt += f"""{PreInstruct}{instruction}
{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif instruction and PreInstruct is not None:
        prompt += f"""{PreInstruct}{instruction}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif input and PreInput is not None:
        prompt += f"""{PreInput}{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif input and instruction and PreInput is not None:
        prompt += f"""{PreInput}{instruction}{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif input and instruction and PreInstruct is not None:
        prompt += f"""{PreInstruct}{instruction}{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif input and instruction:
        # i.e. for simple_instruct
        prompt += f"""{instruction}: {input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif input:
        prompt += f"""{input}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
    elif instruction:
        prompt += f"""{instruction}"""
        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)

    if PreResponse is not None:
        prompt += f"""{PreResponse}"""
        pre_response = PreResponse  # Don't use strip
    else:
        pre_response = ''

    if output:
        prompt += f"""{output}"""

    return prompt, pre_response, terminate_response, chat_sep, chat_turn_sep


def inject_chatsep(prompt_type, prompt, chat_sep=None):
    if chat_sep:
        # only add new line if structured prompt, while 'plain' is just generation of next tokens from input
        prompt += chat_sep
    return prompt


class Prompter(object):
    def __init__(self, prompt_type, prompt_dict, debug=False, chat=False, stream_output=False, repeat_penalty=True,
                 allowed_repeat_line_length=10):
        self.prompt_type = prompt_type
        self.prompt_dict = prompt_dict
        self.debug = debug
        self.chat = chat
        self.stream_output = stream_output
        self.repeat_penalty = repeat_penalty
        self.allowed_repeat_line_length = allowed_repeat_line_length
        self.prompt = None
        context = ""  # not for chat context
        reduced = False  # not for chat context
        making_context = False  # not for chat context
        self.promptA, self.promptB, self.PreInstruct, self.PreInput, self.PreResponse, \
            self.terminate_response, self.chat_sep, self.chat_turn_sep, self.humanstr, self.botstr, \
            self.generates_leading_space = \
            get_prompt(self.prompt_type, self.prompt_dict, chat, context, reduced, making_context)
        self.pre_response = self.PreResponse

    def generate_prompt(self, data_point, reduced=None):
        """
        data_point['context'] is assumed to be like a system prompt or pre-conversation, not inserted after user prompt
        :param data_point:
        :param reduced:
        :return:
        """
        reduced = data_point.get('context') not in ['', None] if reduced is None else reduced
        making_context = False  # whether really making final prompt or just generating context
        prompt, _, _, _, _ = generate_prompt(data_point, self.prompt_type, self.prompt_dict, self.chat, reduced,
                                             making_context)
        if self.debug:
            print("prompt: %s" % prompt, flush=True)
        # if have context, should have always reduced and only preappend promptA/B here
        if data_point.get('context'):
            if data_point.get('input') and self.promptA:
                prompt = self.promptA + prompt
            elif self.promptB:
                prompt = self.promptB + prompt

        self.prompt = prompt
        return prompt

    def get_response(self, outputs, prompt=None, sanitize_bot_response=False):
        if isinstance(outputs, str):
            outputs = [outputs]
        if self.debug:
            print("output:\n%s" % '\n\n'.join(outputs), flush=True)
        if prompt is not None:
            self.prompt = prompt

        def clean_response(response):
            meaningless_words = ['<pad>', '</s>', '<|endoftext|>']
            for word in meaningless_words:
                response = response.replace(word, "")
            if sanitize_bot_response:
                from better_profanity import profanity
                response = profanity.censor(response)
            if self.generates_leading_space and isinstance(response, str) and len(response) > 0 and response[0] == ' ':
                response = response[1:]
            return response

        def clean_repeats(response):
            lines = response.split('\n')
            new_lines = []
            [new_lines.append(line) for line in lines if
             line not in new_lines or len(line) < self.allowed_repeat_line_length]
            if self.debug and len(lines) != len(new_lines):
                print("cleaned repeats: %s %s" % (len(lines), len(new_lines)), flush=True)
            response = '\n'.join(new_lines)
            return response

        multi_output = len(outputs) > 1

        for oi, output in enumerate(outputs):
            if self.prompt_type in [PromptType.plain.value, str(PromptType.plain.value), PromptType.plain.name]:
                output = clean_response(output)
            elif prompt is None:
                # then use most basic parsing like pipeline
                if not self.botstr:
                    pass
                elif self.botstr in output:
                    if self.humanstr:
                        output = clean_response(output.split(self.botstr)[1].split(self.humanstr)[0])
                    else:
                        # i.e. use after bot but only up to next bot
                        output = clean_response(output.split(self.botstr)[1].split(self.botstr)[0])
                else:
                    # output = clean_response(output)
                    # assume just not printed yet
                    output = ""
            else:
                # find first instance of prereponse
                # prompt sometimes has odd characters, that mutate length,
                # so can't go by length alone
                if self.pre_response:
                    outputi = output.find(prompt)
                    if outputi >= 0:
                        output = output[outputi + len(prompt):]
                        allow_terminate = True
                    else:
                        # subtraction is risky due to space offsets sometimes, so only do if necessary
                        output = output[len(prompt) - len(self.pre_response):]
                        # [1] to avoid repeated pre_response, just take first (after prompt - pre_response for chat)
                        if self.pre_response in output:
                            output = output.split(self.pre_response)[1]
                            allow_terminate = True
                        else:
                            if output:
                                print("Failure of parsing or not enough output yet: %s" % output, flush=True)
                            allow_terminate = False
                else:
                    allow_terminate = True
                    output = output[len(prompt):]
                # clean after subtract prompt out, so correct removal of pre_response
                output = clean_response(output)
                if self.repeat_penalty:
                    output = clean_repeats(output)
                if self.terminate_response and allow_terminate:
                    finds = []
                    for term in self.terminate_response:
                        finds.append(output.find(term))
                    finds = [x for x in finds if x >= 0]
                    if len(finds) > 0:
                        termi = finds[0]
                        output = output[:termi]
                    else:
                        output = output
            if multi_output:
                # prefix with output counter
                output = "\n=========== Output %d\n\n" % (1 + oi) + output
                if oi > 0:
                    # post fix outputs with seperator
                    output += '\n'
            output = self.fix_text(self.prompt_type, output)
            outputs[oi] = output
        # join all outputs, only one extra new line between outputs
        output = '\n'.join(outputs)
        if self.debug:
            print("outputclean:\n%s" % '\n\n'.join(outputs), flush=True)
        return output

    @staticmethod
    def fix_text(prompt_type1, text1):
        if prompt_type1 == 'human_bot':
            # hack bug in vLLM with stopping, stops right, but doesn't return last token
            hfix = '<human'
            if text1.endswith(hfix):
                text1 = text1[:-len(hfix)]
        return text1