forked from ggml-org/llama.cpp
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathserver.cpp
4063 lines (3334 loc) · 157 KB
/
server.cpp
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
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#include "utils.hpp"
#include "arg.h"
#include "common.h"
#include "json-schema-to-grammar.h"
#include "llama.h"
#include "log.h"
#include "sampling.h"
#include "speculative.h"
// Change JSON_ASSERT from assert() to GGML_ASSERT:
#define JSON_ASSERT GGML_ASSERT
#include "json.hpp"
// mime type for sending response
#define MIMETYPE_JSON "application/json; charset=utf-8"
// auto generated files (update with ./deps.sh)
#include "index.html.gz.hpp"
#include "loading.html.hpp"
#include <atomic>
#include <condition_variable>
#include <cstddef>
#include <cinttypes>
#include <deque>
#include <memory>
#include <mutex>
#include <signal.h>
#include <thread>
#include <unordered_map>
#include <unordered_set>
using json = nlohmann::ordered_json;
enum stop_type {
STOP_TYPE_NONE,
STOP_TYPE_EOS,
STOP_TYPE_WORD,
STOP_TYPE_LIMIT,
};
// state diagram: https://github.com/ggerganov/llama.cpp/pull/9283
enum slot_state {
SLOT_STATE_IDLE,
SLOT_STATE_STARTED, // TODO: this state is only used for setting up the initial prompt processing; maybe merge it with launch_slot_with_task in the future
SLOT_STATE_PROCESSING_PROMPT,
SLOT_STATE_DONE_PROMPT,
SLOT_STATE_GENERATING,
};
enum server_state {
SERVER_STATE_LOADING_MODEL, // Server is starting up, model not fully loaded yet
SERVER_STATE_READY, // Server is ready and model is loaded
};
enum server_task_type {
SERVER_TASK_TYPE_COMPLETION,
SERVER_TASK_TYPE_EMBEDDING,
SERVER_TASK_TYPE_RERANK,
SERVER_TASK_TYPE_INFILL,
SERVER_TASK_TYPE_CANCEL,
SERVER_TASK_TYPE_NEXT_RESPONSE,
SERVER_TASK_TYPE_METRICS,
SERVER_TASK_TYPE_SLOT_SAVE,
SERVER_TASK_TYPE_SLOT_RESTORE,
SERVER_TASK_TYPE_SLOT_ERASE,
SERVER_TASK_TYPE_SET_LORA,
};
// https://community.openai.com/t/openai-chat-list-of-error-codes-and-types/357791/11
enum error_type {
ERROR_TYPE_INVALID_REQUEST,
ERROR_TYPE_AUTHENTICATION,
ERROR_TYPE_SERVER,
ERROR_TYPE_NOT_FOUND,
ERROR_TYPE_PERMISSION,
ERROR_TYPE_UNAVAILABLE, // custom error
ERROR_TYPE_NOT_SUPPORTED, // custom error
};
struct slot_params {
bool stream = true;
bool cache_prompt = true; // remember the prompt to avoid reprocessing all prompt
bool return_tokens = false;
int32_t n_keep = 0; // number of tokens to keep from initial prompt
int32_t n_discard = 0; // number of tokens after n_keep that may be discarded when shifting context, 0 defaults to half
int32_t n_predict = -1; // new tokens to predict
int32_t n_indent = 0; // mininum line indentation for the generated text in number of whitespace characters
int64_t t_max_prompt_ms = -1; // TODO: implement
int64_t t_max_predict_ms = -1; // if positive, limit the generation phase to this time limit
std::vector<std::string> antiprompt;
bool timings_per_token = false;
bool ignore_eos = false;
struct common_params_sampling sampling;
struct common_params_speculative speculative;
// OAI-compat fields
bool verbose = false;
bool oaicompat = false;
bool oaicompat_chat = true;
std::string oaicompat_model;
std::string oaicompat_cmpl_id;
json to_json() const {
std::vector<std::string> samplers;
samplers.reserve(sampling.samplers.size());
for (const auto & sampler : sampling.samplers) {
samplers.emplace_back(common_sampler_type_to_str(sampler));
}
return json {
{"n_predict", n_predict}, // Server configured n_predict
{"seed", sampling.seed},
{"temperature", sampling.temp},
{"dynatemp_range", sampling.dynatemp_range},
{"dynatemp_exponent", sampling.dynatemp_exponent},
{"top_k", sampling.top_k},
{"top_p", sampling.top_p},
{"min_p", sampling.min_p},
{"xtc_probability", sampling.xtc_probability},
{"xtc_threshold", sampling.xtc_threshold},
{"typical_p", sampling.typ_p},
{"repeat_last_n", sampling.penalty_last_n},
{"repeat_penalty", sampling.penalty_repeat},
{"presence_penalty", sampling.penalty_present},
{"frequency_penalty", sampling.penalty_freq},
{"dry_multiplier", sampling.dry_multiplier},
{"dry_base", sampling.dry_base},
{"dry_allowed_length", sampling.dry_allowed_length},
{"dry_penalty_last_n", sampling.dry_penalty_last_n},
{"dry_sequence_breakers", sampling.dry_sequence_breakers},
{"mirostat", sampling.mirostat},
{"mirostat_tau", sampling.mirostat_tau},
{"mirostat_eta", sampling.mirostat_eta},
{"stop", antiprompt},
{"max_tokens", n_predict}, // User configured n_predict
{"n_keep", n_keep},
{"n_discard", n_discard},
{"ignore_eos", sampling.ignore_eos},
{"stream", stream},
{"logit_bias", format_logit_bias(sampling.logit_bias)},
{"n_probs", sampling.n_probs},
{"min_keep", sampling.min_keep},
{"grammar", sampling.grammar},
{"samplers", samplers},
{"speculative.n_max", speculative.n_max},
{"speculative.n_min", speculative.n_min},
{"speculative.p_min", speculative.p_min},
{"timings_per_token", timings_per_token},
};
}
};
struct server_task {
int id = -1; // to be filled by server_queue
int index = -1; // used when there are multiple prompts (batch request)
server_task_type type;
// used by SERVER_TASK_TYPE_CANCEL
int id_target = -1;
// used by SERVER_TASK_TYPE_INFERENCE
slot_params params;
llama_tokens prompt_tokens;
int id_selected_slot = -1;
// used by SERVER_TASK_TYPE_SLOT_SAVE, SERVER_TASK_TYPE_SLOT_RESTORE, SERVER_TASK_TYPE_SLOT_ERASE
struct slot_action {
int slot_id;
std::string filename;
std::string filepath;
};
slot_action slot_action;
// used by SERVER_TASK_TYPE_METRICS
bool metrics_reset_bucket = false;
server_task(server_task_type type) : type(type) {}
static slot_params params_from_json_cmpl(
const llama_model * model,
const llama_context * ctx,
const common_params & params_base,
const json & data) {
slot_params params;
// Sampling parameter defaults are loaded from the global server context (but individual requests can still override them)
slot_params defaults;
defaults.sampling = params_base.sampling;
defaults.speculative = params_base.speculative;
// enabling this will output extra debug information in the HTTP responses from the server
params.verbose = params_base.verbosity > 9;
params.timings_per_token = json_value(data, "timings_per_token", false);
params.stream = json_value(data, "stream", false);
params.cache_prompt = json_value(data, "cache_prompt", true);
params.return_tokens = json_value(data, "return_tokens", false);
params.n_predict = json_value(data, "n_predict", json_value(data, "max_tokens", defaults.n_predict));
params.n_indent = json_value(data, "n_indent", defaults.n_indent);
params.n_keep = json_value(data, "n_keep", defaults.n_keep);
params.n_discard = json_value(data, "n_discard", defaults.n_discard);
//params.t_max_prompt_ms = json_value(data, "t_max_prompt_ms", defaults.t_max_prompt_ms); // TODO: implement
params.t_max_predict_ms = json_value(data, "t_max_predict_ms", defaults.t_max_predict_ms);
params.sampling.top_k = json_value(data, "top_k", defaults.sampling.top_k);
params.sampling.top_p = json_value(data, "top_p", defaults.sampling.top_p);
params.sampling.min_p = json_value(data, "min_p", defaults.sampling.min_p);
params.sampling.xtc_probability = json_value(data, "xtc_probability", defaults.sampling.xtc_probability);
params.sampling.xtc_threshold = json_value(data, "xtc_threshold", defaults.sampling.xtc_threshold);
params.sampling.typ_p = json_value(data, "typical_p", defaults.sampling.typ_p);
params.sampling.temp = json_value(data, "temperature", defaults.sampling.temp);
params.sampling.dynatemp_range = json_value(data, "dynatemp_range", defaults.sampling.dynatemp_range);
params.sampling.dynatemp_exponent = json_value(data, "dynatemp_exponent", defaults.sampling.dynatemp_exponent);
params.sampling.penalty_last_n = json_value(data, "repeat_last_n", defaults.sampling.penalty_last_n);
params.sampling.penalty_repeat = json_value(data, "repeat_penalty", defaults.sampling.penalty_repeat);
params.sampling.penalty_freq = json_value(data, "frequency_penalty", defaults.sampling.penalty_freq);
params.sampling.penalty_present = json_value(data, "presence_penalty", defaults.sampling.penalty_present);
params.sampling.dry_multiplier = json_value(data, "dry_multiplier", defaults.sampling.dry_multiplier);
params.sampling.dry_base = json_value(data, "dry_base", defaults.sampling.dry_base);
params.sampling.dry_allowed_length = json_value(data, "dry_allowed_length", defaults.sampling.dry_allowed_length);
params.sampling.dry_penalty_last_n = json_value(data, "dry_penalty_last_n", defaults.sampling.dry_penalty_last_n);
params.sampling.mirostat = json_value(data, "mirostat", defaults.sampling.mirostat);
params.sampling.mirostat_tau = json_value(data, "mirostat_tau", defaults.sampling.mirostat_tau);
params.sampling.mirostat_eta = json_value(data, "mirostat_eta", defaults.sampling.mirostat_eta);
params.sampling.seed = json_value(data, "seed", defaults.sampling.seed);
params.sampling.n_probs = json_value(data, "n_probs", defaults.sampling.n_probs);
params.sampling.min_keep = json_value(data, "min_keep", defaults.sampling.min_keep);
params.speculative.n_min = json_value(data, "speculative.n_min", defaults.speculative.n_min);
params.speculative.n_max = json_value(data, "speculative.n_max", defaults.speculative.n_max);
params.speculative.p_min = json_value(data, "speculative.p_min", defaults.speculative.p_min);
params.speculative.n_min = std::min(params.speculative.n_max, params.speculative.n_min);
params.speculative.n_min = std::max(params.speculative.n_min, 2);
params.speculative.n_max = std::max(params.speculative.n_max, 0);
// TODO: add more sanity checks for the input parameters
if (params.sampling.penalty_last_n < -1) {
throw std::runtime_error("Error: repeat_last_n must be >= -1");
}
if (params.sampling.dry_penalty_last_n < -1) {
throw std::runtime_error("Error: dry_penalty_last_n must be >= -1");
}
if (params.sampling.penalty_last_n == -1) {
// note: should be the slot's context and not the full context, but it's ok
params.sampling.penalty_last_n = llama_n_ctx(ctx);
}
if (params.sampling.dry_penalty_last_n == -1) {
params.sampling.dry_penalty_last_n = llama_n_ctx(ctx);
}
if (params.sampling.dry_base < 1.0f) {
params.sampling.dry_base = defaults.sampling.dry_base;
}
// sequence breakers for DRY
{
// Currently, this is not compatible with TextGen WebUI, Koboldcpp and SillyTavern format
// Ref: https://github.com/oobabooga/text-generation-webui/blob/d1af7a41ade7bd3c3a463bfa640725edb818ebaf/extensions/openai/typing.py#L39
if (data.contains("dry_sequence_breakers")) {
params.sampling.dry_sequence_breakers = json_value(data, "dry_sequence_breakers", std::vector<std::string>());
if (params.sampling.dry_sequence_breakers.empty()) {
throw std::runtime_error("Error: dry_sequence_breakers must be a non-empty array of strings");
}
}
}
// process "json_schema" and "grammar"
if (data.contains("json_schema") && !data.at("json_schema").is_null() && data.contains("grammar") && !data.at("grammar").is_null()) {
throw std::runtime_error("Either \"json_schema\" or \"grammar\" can be specified, but not both");
}
if (data.contains("json_schema") && !data.contains("grammar")) {
try {
auto schema = json_value(data, "json_schema", json::object());
params.sampling.grammar = json_schema_to_grammar(schema);
} catch (const std::exception & e) {
throw std::runtime_error(std::string("\"json_schema\": ") + e.what());
}
} else {
params.sampling.grammar = json_value(data, "grammar", defaults.sampling.grammar);
}
{
params.sampling.logit_bias.clear();
params.ignore_eos = json_value(data, "ignore_eos", false);
const auto & logit_bias = data.find("logit_bias");
if (logit_bias != data.end() && logit_bias->is_array()) {
const int n_vocab = llama_n_vocab(model);
for (const auto & el : *logit_bias) {
// TODO: we may want to throw errors here, in case "el" is incorrect
if (el.is_array() && el.size() == 2) {
float bias;
if (el[1].is_number()) {
bias = el[1].get<float>();
} else if (el[1].is_boolean() && !el[1].get<bool>()) {
bias = -INFINITY;
} else {
continue;
}
if (el[0].is_number_integer()) {
llama_token tok = el[0].get<llama_token>();
if (tok >= 0 && tok < n_vocab) {
params.sampling.logit_bias.push_back({tok, bias});
}
} else if (el[0].is_string()) {
auto toks = common_tokenize(model, el[0].get<std::string>(), false);
for (auto tok : toks) {
params.sampling.logit_bias.push_back({tok, bias});
}
}
}
}
}
}
{
params.antiprompt.clear();
const auto & stop = data.find("stop");
if (stop != data.end() && stop->is_array()) {
for (const auto & word : *stop) {
if (!word.empty()) {
params.antiprompt.push_back(word);
}
}
}
}
{
const auto & samplers = data.find("samplers");
if (samplers != data.end()) {
if (samplers->is_array()) {
std::vector<std::string> sampler_names;
for (const auto & name : *samplers) {
if (name.is_string()) {
sampler_names.emplace_back(name);
}
}
params.sampling.samplers = common_sampler_types_from_names(sampler_names, false);
} else if (samplers->is_string()){
std::string sampler_string;
for (const auto & name : *samplers) {
sampler_string += name;
}
params.sampling.samplers = common_sampler_types_from_chars(sampler_string);
}
} else {
params.sampling.samplers = defaults.sampling.samplers;
}
}
std::string model_name = params_base.model_alias.empty() ? DEFAULT_OAICOMPAT_MODEL : params_base.model_alias;
params.oaicompat_model = json_value(data, "model", model_name);
return params;
}
// utility function
static std::unordered_set<int> get_list_id(const std::vector<server_task> & tasks) {
std::unordered_set<int> ids(tasks.size());
for (size_t i = 0; i < tasks.size(); i++) {
ids.insert(tasks[i].id);
}
return ids;
}
};
struct result_timings {
int32_t prompt_n = -1;
double prompt_ms;
double prompt_per_token_ms;
double prompt_per_second;
int32_t predicted_n = -1;
double predicted_ms;
double predicted_per_token_ms;
double predicted_per_second;
json to_json() const {
return {
{"prompt_n", prompt_n},
{"prompt_ms", prompt_ms},
{"prompt_per_token_ms", prompt_per_token_ms},
{"prompt_per_second", prompt_per_second},
{"predicted_n", predicted_n},
{"predicted_ms", predicted_ms},
{"predicted_per_token_ms", predicted_per_token_ms},
{"predicted_per_second", predicted_per_second},
};
}
};
struct server_task_result {
int id = -1;
int id_slot = -1;
virtual bool is_error() {
// only used by server_task_result_error
return false;
}
virtual bool is_stop() {
// only used by server_task_result_cmpl_*
return false;
}
virtual int get_index() {
return -1;
}
virtual json to_json() = 0;
virtual ~server_task_result() = default;
};
// using shared_ptr for polymorphism of server_task_result
using server_task_result_ptr = std::unique_ptr<server_task_result>;
inline std::string stop_type_to_str(stop_type type) {
switch (type) {
case STOP_TYPE_EOS: return "eos";
case STOP_TYPE_WORD: return "word";
case STOP_TYPE_LIMIT: return "limit";
default: return "none";
}
}
struct completion_token_output {
llama_token tok;
std::string text_to_send;
struct token_prob {
llama_token tok;
std::string tok_str;
float prob;
};
std::vector<token_prob> probs;
json to_json() const {
json probs_for_token = json::array();
for (const auto & p : probs) {
probs_for_token.push_back(json {
{"tok_str", p.tok_str},
{"prob", p.prob},
});
}
return probs_for_token;
}
static json probs_vector_to_json(const std::vector<completion_token_output> & probs) {
json out = json::array();
for (const auto & prob : probs) {
const std::string tok_str = prob.text_to_send;
out.push_back(json {
{"content", tok_str},
{"probs", prob.to_json()},
});
}
return out;
}
};
struct server_task_result_cmpl_final : server_task_result {
int index = 0;
std::string content;
llama_tokens tokens;
bool stream;
result_timings timings;
std::string prompt;
bool truncated;
int32_t n_decoded;
int32_t n_prompt_tokens;
int32_t n_tokens_cached;
bool has_new_line;
std::string stopping_word;
stop_type stop = STOP_TYPE_NONE;
std::vector<completion_token_output> probs_output;
slot_params generation_params;
// OAI-compat fields
bool verbose = false;
bool oaicompat = false;
bool oaicompat_chat = true; // TODO: support oaicompat for non-chat
std::string oaicompat_model;
std::string oaicompat_cmpl_id;
virtual int get_index() override {
return index;
}
virtual bool is_stop() override {
return true; // in stream mode, final responses are considered stop
}
virtual json to_json() override {
return oaicompat
? (stream ? to_json_oaicompat_chat_stream() : to_json_oaicompat_chat())
: to_json_non_oaicompat();
}
json to_json_non_oaicompat() {
json res = json {
{"index", index},
{"content", stream ? "" : content}, // in stream mode, content is already in last partial chunk
{"tokens", stream ? llama_tokens {} : tokens},
{"id_slot", id_slot},
{"stop", true},
{"model", oaicompat_model},
{"tokens_predicted", n_decoded},
{"tokens_evaluated", n_prompt_tokens},
{"generation_settings", generation_params.to_json()},
{"prompt", prompt},
{"has_new_line", has_new_line},
{"truncated", truncated},
{"stop_type", stop_type_to_str(stop)},
{"stopping_word", stopping_word},
{"tokens_cached", n_tokens_cached},
{"timings", timings.to_json()},
};
if (!probs_output.empty()) {
res["completion_probabilities"] = completion_token_output::probs_vector_to_json(probs_output);
}
return res;
}
json to_json_oaicompat_chat() {
std::string finish_reason = "length";
if (stop == STOP_TYPE_WORD || stop == STOP_TYPE_EOS) {
finish_reason = "stop";
}
json choices = json::array({json{
{"finish_reason", finish_reason},
{"index", 0},
{"message", json {
{"content", content},
{"role", "assistant"}
}
}}});
std::time_t t = std::time(0);
json res = json {
{"choices", choices},
{"created", t},
{"model", oaicompat_model},
{"object", "chat.completion"},
{"usage", json {
{"completion_tokens", n_decoded},
{"prompt_tokens", n_prompt_tokens},
{"total_tokens", n_decoded + n_prompt_tokens}
}},
{"id", oaicompat_cmpl_id}
};
// extra fields for debugging purposes
if (verbose) {
res["__verbose"] = to_json_non_oaicompat();
}
if (timings.prompt_n >= 0) {
res.push_back({"timings", timings.to_json()});
}
return res;
}
json to_json_oaicompat_chat_stream() {
std::time_t t = std::time(0);
std::string finish_reason = "length";
if (stop == STOP_TYPE_WORD || stop == STOP_TYPE_EOS) {
finish_reason = "stop";
}
json choices = json::array({json{{"finish_reason", finish_reason},
{"index", 0},
{"delta", json::object()}}});
json ret = json {
{"choices", choices},
{"created", t},
{"id", oaicompat_cmpl_id},
{"model", oaicompat_model},
{"object", "chat.completion.chunk"},
{"usage", json {
{"completion_tokens", n_decoded},
{"prompt_tokens", n_prompt_tokens},
{"total_tokens", n_decoded + n_prompt_tokens},
}},
};
if (timings.prompt_n >= 0) {
ret.push_back({"timings", timings.to_json()});
}
return ret;
}
};
struct server_task_result_cmpl_partial : server_task_result {
int index = 0;
std::string content;
llama_tokens tokens;
int32_t n_decoded;
int32_t n_prompt_tokens;
std::vector<completion_token_output> probs_output;
result_timings timings;
// OAI-compat fields
bool verbose = false;
bool oaicompat = false;
bool oaicompat_chat = true; // TODO: support oaicompat for non-chat
std::string oaicompat_model;
std::string oaicompat_cmpl_id;
virtual int get_index() override {
return index;
}
virtual bool is_stop() override {
return false; // in stream mode, partial responses are not considered stop
}
virtual json to_json() override {
return oaicompat ? to_json_oaicompat() : to_json_non_oaicompat();
}
json to_json_non_oaicompat() {
// non-OAI-compat JSON
json res = json {
{"index", index},
{"content", content},
{"tokens", tokens},
{"stop", false},
{"id_slot", id_slot},
{"tokens_predicted", n_decoded},
{"tokens_evaluated", n_prompt_tokens},
};
// populate the timings object when needed (usually for the last response or with timings_per_token enabled)
if (timings.prompt_n > 0) {
res.push_back({"timings", timings.to_json()});
}
if (!probs_output.empty()) {
res["completion_probabilities"] = completion_token_output::probs_vector_to_json(probs_output);
}
return res;
}
json to_json_oaicompat() {
bool first = n_decoded == 0;
std::time_t t = std::time(0);
json choices;
if (first) {
if (content.empty()) {
choices = json::array({json{{"finish_reason", nullptr},
{"index", 0},
{"delta", json{{"role", "assistant"}}}}});
} else {
// We have to send this as two updates to conform to openai behavior
json initial_ret = json{{"choices", json::array({json{
{"finish_reason", nullptr},
{"index", 0},
{"delta", json{
{"role", "assistant"}
}}}})},
{"created", t},
{"id", oaicompat_cmpl_id},
{"model", oaicompat_model},
{"object", "chat.completion.chunk"}};
json second_ret = json{
{"choices", json::array({json{{"finish_reason", nullptr},
{"index", 0},
{"delta", json {
{"content", content}}}
}})},
{"created", t},
{"id", oaicompat_cmpl_id},
{"model", oaicompat_model},
{"object", "chat.completion.chunk"}};
return std::vector<json>({initial_ret, second_ret});
}
} else {
choices = json::array({json{
{"finish_reason", nullptr},
{"index", 0},
{"delta",
json {
{"content", content},
}},
}});
}
json ret = json {
{"choices", choices},
{"created", t},
{"id", oaicompat_cmpl_id},
{"model", oaicompat_model},
{"object", "chat.completion.chunk"}
};
if (timings.prompt_n >= 0) {
ret.push_back({"timings", timings.to_json()});
}
return std::vector<json>({ret});
}
};
struct server_task_result_embd : server_task_result {
int index = 0;
std::vector<std::vector<float>> embedding;
int32_t n_tokens;
// OAI-compat fields
bool oaicompat = false;
virtual int get_index() override {
return index;
}
virtual json to_json() override {
return oaicompat ? to_json_oaicompat() : to_json_non_oaicompat();
}
json to_json_non_oaicompat() {
return json {
{"index", index},
{"embedding", embedding},
};
}
json to_json_oaicompat() {
return json {
{"index", index},
{"embedding", embedding[0]},
{"tokens_evaluated", n_tokens},
};
}
};
struct server_task_result_rerank : server_task_result {
int index = 0;
float score = -1e6;
int32_t n_tokens;
virtual int get_index() override {
return index;
}
virtual json to_json() override {
return json {
{"index", index},
{"score", score},
{"tokens_evaluated", n_tokens},
};
}
};
// this function maybe used outside of server_task_result_error
static json format_error_response(const std::string & message, const enum error_type type) {
std::string type_str;
int code = 500;
switch (type) {
case ERROR_TYPE_INVALID_REQUEST:
type_str = "invalid_request_error";
code = 400;
break;
case ERROR_TYPE_AUTHENTICATION:
type_str = "authentication_error";
code = 401;
break;
case ERROR_TYPE_NOT_FOUND:
type_str = "not_found_error";
code = 404;
break;
case ERROR_TYPE_SERVER:
type_str = "server_error";
code = 500;
break;
case ERROR_TYPE_PERMISSION:
type_str = "permission_error";
code = 403;
break;
case ERROR_TYPE_NOT_SUPPORTED:
type_str = "not_supported_error";
code = 501;
break;
case ERROR_TYPE_UNAVAILABLE:
type_str = "unavailable_error";
code = 503;
break;
}
return json {
{"code", code},
{"message", message},
{"type", type_str},
};
}
struct server_task_result_error : server_task_result {
int index = 0;
error_type err_type = ERROR_TYPE_SERVER;
std::string err_msg;
virtual bool is_error() override {
return true;
}
virtual json to_json() override {
return format_error_response(err_msg, err_type);
}
};
struct server_task_result_metrics : server_task_result {
int n_idle_slots;
int n_processing_slots;
int n_tasks_deferred;
int64_t t_start;
int32_t kv_cache_tokens_count;
int32_t kv_cache_used_cells;
// TODO: somehow reuse server_metrics in the future, instead of duplicating the fields
uint64_t n_prompt_tokens_processed_total = 0;
uint64_t t_prompt_processing_total = 0;
uint64_t n_tokens_predicted_total = 0;
uint64_t t_tokens_generation_total = 0;
uint64_t n_prompt_tokens_processed = 0;
uint64_t t_prompt_processing = 0;
uint64_t n_tokens_predicted = 0;
uint64_t t_tokens_generation = 0;
uint64_t n_decode_total = 0;
uint64_t n_busy_slots_total = 0;
// while we can also use std::vector<server_slot> this requires copying the slot object which can be quite messy
// therefore, we use json to temporarily store the slot.to_json() result
json slots_data = json::array();
virtual json to_json() override {
return json {
{ "idle", n_idle_slots },
{ "processing", n_processing_slots },
{ "deferred", n_tasks_deferred },
{ "t_start", t_start },
{ "n_prompt_tokens_processed_total", n_prompt_tokens_processed_total },
{ "t_tokens_generation_total", t_tokens_generation_total },
{ "n_tokens_predicted_total", n_tokens_predicted_total },
{ "t_prompt_processing_total", t_prompt_processing_total },
{ "n_prompt_tokens_processed", n_prompt_tokens_processed },
{ "t_prompt_processing", t_prompt_processing },
{ "n_tokens_predicted", n_tokens_predicted },
{ "t_tokens_generation", t_tokens_generation },
{ "n_decode_total", n_decode_total },
{ "n_busy_slots_total", n_busy_slots_total },
{ "kv_cache_tokens_count", kv_cache_tokens_count },
{ "kv_cache_used_cells", kv_cache_used_cells },
{ "slots", slots_data },
};
}
};
struct server_task_result_slot_save_load : server_task_result {
std::string filename;
bool is_save; // true = save, false = load
size_t n_tokens;
size_t n_bytes;
double t_ms;
virtual json to_json() override {
if (is_save) {
return json {
{ "id_slot", id_slot },
{ "filename", filename },
{ "n_saved", n_tokens },
{ "n_written", n_bytes },
{ "timings", {
{ "save_ms", t_ms }
}},
};
} else {
return json {
{ "id_slot", id_slot },
{ "filename", filename },
{ "n_restored", n_tokens },
{ "n_read", n_bytes },
{ "timings", {
{ "restore_ms", t_ms }
}},
};
}
}
};
struct server_task_result_slot_erase : server_task_result {
size_t n_erased;
virtual json to_json() override {
return json {
{ "id_slot", id_slot },
{ "n_erased", n_erased },
};
}
};
struct server_task_result_apply_lora : server_task_result {
virtual json to_json() override {
return json {{ "success", true }};
}
};
struct server_slot {
int id;
int id_task = -1;
// only used for completion/embedding/infill/rerank
server_task_type task_type = SERVER_TASK_TYPE_COMPLETION;
llama_batch batch_spec = {};
llama_context * ctx = nullptr;
llama_context * ctx_dft = nullptr;
common_speculative * spec = nullptr;
// the index relative to completion multi-task request
size_t index = 0;
struct slot_params params;
slot_state state = SLOT_STATE_IDLE;
// used to determine the slot that has been used the longest
int64_t t_last_used = -1;
// generation props
int32_t n_ctx = 0; // context size per slot
int32_t n_past = 0;
int32_t n_decoded = 0;
int32_t n_remaining = -1;
int32_t i_batch = -1;
int32_t n_predict = -1; // TODO: disambiguate from params.n_predict
// n_prompt_tokens may not be equal to prompt_tokens.size(), because prompt maybe truncated
int32_t n_prompt_tokens = 0;
int32_t n_prompt_tokens_processed = 0;
// input prompt tokens
llama_tokens prompt_tokens;
size_t last_nl_pos = 0;
std::string generated_text;
llama_tokens generated_tokens;
llama_tokens cache_tokens;
std::vector<completion_token_output> generated_token_probs;
bool has_next_token = true;
bool has_new_line = false;
bool truncated = false;
stop_type stop;
std::string stopping_word;
// sampling
json json_schema;
struct common_sampler * smpl = nullptr;
llama_token sampled;