Skip to content
Navigation Menu
{{ message }}
forked from ggml-org/llama.cpp
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathchat.cpp
More file actions
2711 lines (2337 loc) · 115 KB
/
Copy pathchat.cpp
File metadata and controls
2711 lines (2337 loc) · 115 KB
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 "chat.h"
#include "chat-auto-parser-helpers.h"
#include "chat-auto-parser.h"
#include "chat-peg-parser.h"
#include "common.h"
#include "ggml.h"
#include "json-schema-to-grammar.h"
#include "log.h"
#include "jinja/value.h"
#include "jinja/runtime.h"
#include "jinja/caps.h"
#include "peg-parser.h"
#include "nlohmann/json.hpp"
#include <cstdio>
#include <cstdlib>
#include <ctime>
#include <exception>
#include <functional>
#include <optional>
#include <sstream>
#include <stdexcept>
#include <string>
#include <utility>
#include <vector>
using json = nlohmann::ordered_json;
static std::string format_time(const std::chrono::system_clock::time_point & now, const std::string & format) {
auto time = std::chrono::system_clock::to_time_t(now);
auto local_time = *std::localtime(&time);
std::ostringstream ss;
ss << std::put_time(&local_time, format.c_str());
auto res = ss.str();
return res;
}
static json safe_args_parse(const std::string & to_parse) {
std::string stripped = to_parse;
if (to_parse.at(0) == '"' && to_parse.at(to_parse.length() - 1) == '"') {
stripped = to_parse.substr(1, to_parse.length() - 1);
}
try {
return json::parse(stripped);
} catch (json::exception & e) {
return stripped;
}
}
static std::string string_diff(const std::string & last, const std::string & current) {
if (last.empty()) {
return current;
}
if (!string_starts_with(current, last)) {
if (string_starts_with(last, current)) {
// This happens if the last generation ended on a partial stop word (not erased),
// and the current ended on a stop word (erased).
return "";
}
throw std::runtime_error("Invalid diff: '" + last + "' not found at start of '" + current + "'");
}
return current.substr(last.size());
}
static bool has_content_or_tool_calls(const common_chat_msg & msg) {
return !msg.content.empty() || !msg.tool_calls.empty();
}
std::string common_chat_msg::render_content(const std::string & delimiter) const {
if (!content.empty() && !content_parts.empty()) {
throw std::runtime_error("Cannot specify both content and content_parts");
}
if (!content.empty()) {
return content;
}
std::string text;
for (const auto & part : content_parts) {
if (part.type == "text") {
if (!text.empty()) {
text += delimiter;
}
text += part.text;
}
}
return text;
}
std::vector<common_chat_msg_span> common_chat_split_by_role(const std::string & prompt, const std::vector<common_chat_msg_delimiter> & delims) {
if (delims.empty() || prompt.empty()) {
return {};
}
auto parser = build_peg_parser([&](common_peg_parser_builder & p) {
std::vector<std::string> all_delims;
std::vector<common_peg_parser> tagged_messages;
all_delims.reserve(delims.size());
tagged_messages.reserve(delims.size());
for (const auto & d : delims) {
all_delims.push_back(d.delimiter);
}
auto any_delim = p.until_one_of(all_delims);
for (const auto & d : delims) {
tagged_messages.push_back(p.tag(d.role, p.literal(d.delimiter) + any_delim));
}
return any_delim + p.zero_or_more(p.choice(tagged_messages)) + p.end();
});
common_peg_parse_context ctx(prompt);
const auto result = parser.parse(ctx);
if (!result.success()) {
return {};
}
std::vector<common_chat_msg_span> spans;
ctx.ast.visit(result, [&](const common_peg_ast_node & node) {
if (!node.tag.empty()) {
spans.push_back({ node.tag, node.start, node.end - node.start });
}
});
return spans;
}
json common_chat_msg::to_json_oaicompat(bool concat_typed_text) const {
if (!content.empty() && !content_parts.empty()) {
throw std::runtime_error("Cannot specify both content and content_parts");
}
json jmsg {
{"role", role},
};
if (!content.empty()) {
jmsg["content"] = content;
} else if (!content_parts.empty()) {
if (concat_typed_text || contains_media()) {
std::string text;
bool last_was_media_marker = false;
// join parts with newline, do not add newline before or after media markers
for (const auto & part : content_parts) {
bool add_new_line = true;
if (part.type == "text") {
add_new_line = !last_was_media_marker && !text.empty();
last_was_media_marker = false;
} else if (part.type == "media_marker") {
add_new_line = false;
last_was_media_marker = true;
} else {
LOG_WRN("Ignoring content part type: %s\n", part.type.c_str());
continue;
}
if (add_new_line) {
text += '\n';
}
text += part.text;
}
jmsg["content"] = text;
} else {
auto & parts = jmsg["content"] = json::array();
for (const auto & part : content_parts) {
parts.push_back({
{"type", part.type},
{"text", part.text},
});
}
}
} else {
jmsg["content"] = "";
}
if (!reasoning_content.empty()) {
jmsg["reasoning_content"] = reasoning_content;
}
if (!tool_name.empty()) {
jmsg["name"] = tool_name;
}
if (!tool_call_id.empty()) {
jmsg["tool_call_id"] = tool_call_id;
}
if (!tool_calls.empty()) {
jmsg["tool_calls"] = json::array();
auto & jtool_calls = jmsg["tool_calls"];
for (const auto & tool_call : tool_calls) {
json tc {
{"type", "function"},
{"function", {
{"name", tool_call.name},
{"arguments", json(tool_call.arguments)},
}},
};
if (!tool_call.id.empty()) {
tc["id"] = tool_call.id;
}
// Some templates generate and require an id (sometimes in a very specific format, e.g. Mistral Nemo).
// We only generate a random id for the ones that don't generate one by themselves
// (they also won't get to see it as their template likely doesn't use it, so it's all for the client)
// {"id", tc.id.empty() ? gen_tool_call_id() : tc.id},
jtool_calls.push_back(tc);
}
}
return jmsg;
}
std::vector<common_chat_msg_diff> common_chat_msg_diff::compute_diffs(const common_chat_msg & msg_prv,
const common_chat_msg & msg_new) {
std::vector<common_chat_msg_diff> diffs;
if (msg_new.tool_calls.size() > msg_prv.tool_calls.size()) {
diffs.reserve(msg_new.tool_calls.size() - msg_prv.tool_calls.size() + 3);
} else {
diffs.reserve(3);
}
// TODO: these can become expensive for long messages - how to optimize?
if (msg_prv.reasoning_content != msg_new.reasoning_content) {
auto & diff = diffs.emplace_back();
diff.reasoning_content_delta = string_diff(msg_prv.reasoning_content, msg_new.reasoning_content);
}
if (msg_prv.content != msg_new.content) {
auto & diff = diffs.emplace_back();
diff.content_delta = string_diff(msg_prv.content, msg_new.content);
}
if (msg_new.tool_calls.size() < msg_prv.tool_calls.size()) {
std::string err = "Invalid diff: now finding less tool calls!\n";
err += " Previous (" + std::to_string(msg_prv.tool_calls.size()) + "):\n";
for (const auto & tc : msg_prv.tool_calls) {
err += " - name: '" + tc.name + "', args: '" + tc.arguments + "'\n";
}
err += " Current (" + std::to_string(msg_new.tool_calls.size()) + "):\n";
for (const auto & tc : msg_new.tool_calls) {
err += " - name: '" + tc.name + "', args: '" + tc.arguments + "'\n";
}
err += " Current msg text content:\n" + msg_new.content + "\n";
throw std::runtime_error(err);
}
if (!msg_prv.tool_calls.empty()) {
const auto idx = msg_prv.tool_calls.size() - 1;
const auto & pref = msg_prv.tool_calls[idx];
const auto & newf = msg_new.tool_calls[idx];
// Allow tool name to change during incremental parsing:
// - empty -> non-empty (initial discovery)
// - prefix -> longer string (name grows as more input is parsed)
if (pref.name != newf.name && !pref.name.empty() && !newf.name.empty()) {
// Check if one is a prefix of the other (for incremental parsing where names grow or shrink)
bool is_prefix = (newf.name.rfind(pref.name, 0) == 0);
if (!is_prefix) {
LOG_ERR("Tool call mismatch: prev='%s' new='%s'\n", pref.name.c_str(), newf.name.c_str());
throw std::runtime_error("Invalid diff: tool call mismatch!");
}
}
const auto args_diff = string_diff(pref.arguments, newf.arguments);
if (!args_diff.empty() || pref.id != newf.id || pref.name != newf.name) {
auto & diff = diffs.emplace_back();
diff.tool_call_index = idx;
if (pref.id != newf.id || pref.name != newf.name) {
diff.tool_call_delta.id = newf.id;
diff.tool_call_delta.name = newf.name;
}
diff.tool_call_delta.arguments = args_diff;
}
}
for (size_t idx = msg_prv.tool_calls.size(); idx < msg_new.tool_calls.size(); ++idx) {
auto & diff = diffs.emplace_back();
diff.tool_call_index = idx;
diff.tool_call_delta = msg_new.tool_calls[idx];
}
return diffs;
}
using chat_template_caps = jinja::caps;
struct common_chat_templates {
bool add_bos;
bool add_eos;
bool has_explicit_template; // Model had builtin template or template overridden was specified.
std::unique_ptr<common_chat_template> template_default; // always set (defaults to chatml)
std::unique_ptr<common_chat_template> template_tool_use;
};
common_chat_tool_choice common_chat_tool_choice_parse_oaicompat(const std::string & tool_choice) {
if (tool_choice == "auto") {
return COMMON_CHAT_TOOL_CHOICE_AUTO;
}
if (tool_choice == "none") {
return COMMON_CHAT_TOOL_CHOICE_NONE;
}
if (tool_choice == "required") {
return COMMON_CHAT_TOOL_CHOICE_REQUIRED;
}
throw std::invalid_argument("Invalid tool_choice: " + tool_choice);
}
bool common_chat_templates_support_enable_thinking(const common_chat_templates * chat_templates) {
common_chat_templates_inputs inputs;
inputs.reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
common_chat_msg msg;
msg.role = "user";
msg.content = "test";
inputs.messages = { msg };
inputs.enable_thinking = true;
inputs.add_generation_prompt = true;
inputs.reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
auto params = common_chat_templates_apply(chat_templates, inputs);
return params.supports_thinking;
}
std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messages) {
std::vector<common_chat_msg> msgs;
try {
if (!messages.is_array()) {
throw std::invalid_argument("Expected 'messages' to be an array, got " + messages.dump());
}
for (const auto & message : messages) {
if (!message.is_object()) {
throw std::invalid_argument("Expected 'message' to be an object, got " + message.dump());
}
common_chat_msg msg;
if (!message.contains("role")) {
throw std::invalid_argument("Missing 'role' in message: " + message.dump());
}
msg.role = message.at("role");
auto has_content = message.contains("content");
auto has_tool_calls = message.contains("tool_calls");
if (has_content) {
const auto & content = message.at("content");
if (content.is_string()) {
msg.content = content;
} else if (content.is_array()) {
for (const auto & part : content) {
if (!part.contains("type")) {
throw std::invalid_argument("Missing content part type: " + part.dump());
}
const auto & type = part.at("type");
if (type != "text" && type != "media_marker") {
throw std::invalid_argument("Unsupported content part type: " + type.dump());
}
common_chat_msg_content_part msg_part;
msg_part.type = type;
msg_part.text = part.at("text");
msg.content_parts.push_back(msg_part);
}
} else if (!content.is_null()) {
throw std::invalid_argument("Invalid 'content' type: expected string or array, got " +
content.dump() +
" (ref: https://github.com/ggml-org/llama.cpp/issues/8367)");
}
}
if (has_tool_calls) {
for (const auto & tool_call : message.at("tool_calls")) {
common_chat_tool_call tc;
if (!tool_call.contains("type")) {
throw std::invalid_argument("Missing tool call type: " + tool_call.dump());
}
const auto & type = tool_call.at("type");
if (type != "function") {
throw std::invalid_argument("Unsupported tool call type: " + tool_call.dump());
}
if (!tool_call.contains("function")) {
throw std::invalid_argument("Missing tool call function: " + tool_call.dump());
}
const auto & fc = tool_call.at("function");
if (!fc.contains("name")) {
throw std::invalid_argument("Missing tool call name: " + tool_call.dump());
}
tc.name = fc.at("name");
const auto & args = fc.at("arguments");
if (args.is_string()) {
tc.arguments = args;
} else {
tc.arguments = args.dump();
}
if (tool_call.contains("id")) {
tc.id = tool_call.at("id");
}
msg.tool_calls.push_back(tc);
}
}
if (!has_content && !has_tool_calls) {
throw std::invalid_argument(
"Expected 'content' or 'tool_calls' (ref: https://github.com/ggml-org/llama.cpp/issues/8367 & "
"https://github.com/ggml-org/llama.cpp/issues/12279)");
}
if (message.contains("reasoning_content")) {
msg.reasoning_content = message.at("reasoning_content");
}
if (message.contains("name")) {
msg.tool_name = message.at("name");
}
if (message.contains("tool_call_id")) {
msg.tool_call_id = message.at("tool_call_id");
}
msgs.push_back(msg);
}
} catch (const std::exception & e) {
// @ngxson : disable otherwise it's bloating the API response
// printf("%s\n", std::string("; messages = ") + messages.dump(2));
throw std::runtime_error("Failed to parse messages: " + std::string(e.what()));
}
return msgs;
}
static json render_message_to_json(const std::vector<common_chat_msg> & msgs, const jinja::caps & c) {
if (!c.supports_string_content && !c.supports_typed_content) {
LOG_WRN("%s: Neither string content nor typed content is supported by the template. This is unexpected and may lead to issues.\n", __func__);
}
bool only_string_accepted = c.supports_string_content && !c.supports_typed_content;
bool only_typed_accepted = !c.supports_string_content && c.supports_typed_content;
json messages = json::array();
for (const auto & msg : msgs) {
if (only_string_accepted) {
json jmsg = msg.to_json_oaicompat(/* concat_typed_text= */ true);
messages.push_back(jmsg);
} else if (only_typed_accepted) {
json jmsg = msg.to_json_oaicompat(/* concat_typed_text= */ false);
if (jmsg.at("content").is_string()) {
jmsg["content"] = json::array({
json{
{"type", "text"},
{"text", jmsg.at("content").get<std::string>()},
}
});
}
messages.push_back(jmsg);
} else {
json jmsg = msg.to_json_oaicompat(/* concat_typed_text= */ false);
messages.push_back(jmsg);
}
}
return messages;
}
// DEPRECATED: only used in tests
json common_chat_msgs_to_json_oaicompat(const std::vector<common_chat_msg> & msgs, bool concat_typed_text) {
jinja::caps c;
c.supports_string_content = true;
c.supports_typed_content = !concat_typed_text;
return render_message_to_json(msgs, c);
}
json common_chat_tools_to_json_oaicompat(const std::vector<common_chat_tool> & tools) {
if (tools.empty()) {
return json();
}
auto result = json::array();
for (const auto & tool : tools) {
result.push_back({
{ "type", "function" },
{ "function", {
{ "name", tool.name },
{ "description", tool.description },
{ "parameters", json::parse(tool.parameters) },
}},
});
}
return result;
}
std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const json & tools) {
std::vector<common_chat_tool> result;
try {
if (!tools.is_null()) {
if (!tools.is_array()) {
throw std::invalid_argument("Expected 'tools' to be an array, got " + tools.dump());
}
for (const auto & tool : tools) {
if (!tool.contains("type")) {
throw std::invalid_argument("Missing tool type: " + tool.dump());
}
const auto & type = tool.at("type");
if (!type.is_string() || type != "function") {
throw std::invalid_argument("Unsupported tool type: " + tool.dump());
}
if (!tool.contains("function")) {
throw std::invalid_argument("Missing tool function: " + tool.dump());
}
const auto & function = tool.at("function");
result.push_back({
/* .name = */ function.at("name"),
/* .description = */ function.value("description", ""),
/* .parameters = */ function.value("parameters", json::object()).dump(),
});
}
}
} catch (const std::exception & e) {
throw std::runtime_error("Failed to parse tools: " + std::string(e.what()) + "; tools = " + tools.dump(2));
}
return result;
}
common_chat_continuation common_chat_continuation_parse(const nlohmann::ordered_json & value) {
if (value.is_boolean() && value.get<bool>()) {
return COMMON_CHAT_CONTINUATION_AUTO;
}
if (value.is_string()) {
auto value_str = value.get<std::string>();
if (value_str == "reasoning_content") {
return COMMON_CHAT_CONTINUATION_REASONING;
}
if (value_str == "content") {
return COMMON_CHAT_CONTINUATION_CONTENT;
}
}
return COMMON_CHAT_CONTINUATION_NONE;
}
bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) {
if (use_jinja) {
try {
common_chat_msg msg;
msg.role = "user";
msg.content = "test";
auto tmpls = common_chat_templates_init(/* model= */ nullptr, tmpl);
common_chat_templates_inputs inputs;
inputs.messages = { msg };
common_chat_templates_apply(tmpls.get(), inputs);
return true;
} catch (const std::exception & e) {
LOG_ERR("%s: failed to apply template: %s\n", __func__, e.what());
return false;
}
}
llama_chat_message chat[] = {
{ "user", "test" }
};
const int res = llama_chat_apply_template(tmpl.c_str(), chat, 1, true, nullptr, 0);
return res >= 0;
}
std::string common_chat_format_single(const struct common_chat_templates * tmpls,
const std::vector<common_chat_msg> & past_msg,
const common_chat_msg & new_msg,
bool add_ass,
bool use_jinja) {
common_chat_templates_inputs inputs;
inputs.use_jinja = use_jinja;
inputs.add_bos = tmpls->add_bos;
inputs.add_eos = tmpls->add_eos;
std::string fmt_past_msg;
if (!past_msg.empty()) {
inputs.messages = past_msg;
inputs.add_generation_prompt = false;
fmt_past_msg = common_chat_templates_apply(tmpls, inputs).prompt;
}
std::ostringstream ss;
// if the past_msg ends with a newline, we must preserve it in the formatted version
if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') {
ss << "\n";
};
// format chat with new_msg
inputs.messages.push_back(new_msg);
inputs.add_generation_prompt = add_ass;
auto fmt_new_msg = common_chat_templates_apply(tmpls, inputs).prompt;
// get the diff part
ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
return ss.str();
}
std::string common_chat_format_example(const struct common_chat_templates * tmpls,
bool use_jinja,
const std::map<std::string, std::string> & chat_template_kwargs) {
common_chat_templates_inputs inputs;
inputs.use_jinja = use_jinja;
inputs.add_bos = tmpls->add_bos;
inputs.add_eos = tmpls->add_eos;
inputs.chat_template_kwargs = chat_template_kwargs;
auto add_simple_msg = [&](auto role, auto content) {
common_chat_msg msg;
msg.role = role;
msg.content = content;
inputs.messages.push_back(msg);
};
add_simple_msg("system", "You are a helpful assistant");
add_simple_msg("user", "Hello");
add_simple_msg("assistant", "Hi there");
add_simple_msg("user", "How are you?");
return common_chat_templates_apply(tmpls, inputs).prompt;
}
#define CHATML_TEMPLATE_SRC \
"{%- for message in messages -%}\n" \
" {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>\n' -}}\n" \
"{%- endfor -%}\n" \
"{%- if add_generation_prompt -%}\n" \
" {{- '<|im_start|>assistant\n' -}}\n" \
"{%- endif -%}"
void common_chat_templates_free(struct common_chat_templates * tmpls) {
delete tmpls;
}
bool common_chat_templates_was_explicit(const struct common_chat_templates * tmpls) {
return tmpls->has_explicit_template;
}
// LFM2 format detection: template uses <|tool_list_start|>[...]<|tool_list_end|> around the tool list
// and <|tool_call_start|>[...]<|tool_call_end|> around each tool call
static bool is_lfm2_template(const std::string & src) {
return src.find("<|tool_list_start|>") != std::string::npos &&
src.find("<|tool_list_end|>") != std::string::npos;
}
common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates) {
common_chat_prompt_preset asr_preset;
asr_preset.system = "";
asr_preset.user = "Transcribe audio to text";
if (chat_templates && chat_templates->template_default && is_lfm2_template(chat_templates->template_default->source())) {
asr_preset.system = "Perform ASR.";
asr_preset.user = "";
}
return asr_preset;
}
std::string common_chat_templates_source(const struct common_chat_templates * tmpls, const std::string & variant) {
if (!variant.empty()) {
if (variant == "tool_use") {
if (tmpls->template_tool_use) {
return tmpls->template_tool_use->source();
}
return "";
}
LOG_DBG("%s: unknown template variant: %s\n", __func__, variant.c_str());
}
return tmpls->template_default->source();
}
common_chat_templates_ptr common_chat_templates_init(const struct llama_model * model,
const std::string & chat_template_override,
const std::string & bos_token_override,
const std::string & eos_token_override) {
std::string default_template_src;
std::string template_tool_use_src;
bool has_explicit_template = !chat_template_override.empty();
if (chat_template_override.empty()) {
GGML_ASSERT(model != nullptr);
const auto * str = llama_model_chat_template(model, /* name */ nullptr);
if (str) {
default_template_src = str;
has_explicit_template = true;
}
str = llama_model_chat_template(model, /* name */ "tool_use");
if (str) {
template_tool_use_src = str;
has_explicit_template = true;
}
} else {
default_template_src = chat_template_override;
}
if (default_template_src.empty() || default_template_src == "chatml") {
if (!template_tool_use_src.empty()) {
default_template_src = template_tool_use_src;
} else {
default_template_src = CHATML_TEMPLATE_SRC;
}
}
// TODO @ngxson : this is a temporary hack to prevent chat template from throwing an error
// Ref: https://github.com/ggml-org/llama.cpp/pull/15230#issuecomment-3173959633
if (default_template_src.find("<|channel|>") != std::string::npos
// search for the error message and patch it
&& default_template_src.find("in message.content or") != std::string::npos) {
string_replace_all(default_template_src,
"{%- if \"<|channel|>analysis<|message|>\" in message.content or "
"\"<|channel|>final<|message|>\" in message.content %}",
"{%- if false %}");
}
// TODO @aldehir : this is a temporary fix, pending Minja changes
// Ref: https://github.com/ggml-org/llama.cpp/pull/17713#issuecomment-3631342664
if (default_template_src.find("[TOOL_CALLS]") != std::string::npos
// search for the error message and patch it
&& default_template_src.find("if (message['content'] is none or") != std::string::npos) {
string_replace_all(default_template_src,
"{%- if (message['content'] is none or message['content'] == '' or "
"message['content']|length == 0) and (message['tool_calls'] is not defined or "
"message['tool_calls'] is none or message['tool_calls']|length == 0) %}",
"{%- if false %}");
}
std::string token_bos = bos_token_override;
std::string token_eos = eos_token_override;
bool add_bos = false;
bool add_eos = false;
if (model) {
const auto * vocab = llama_model_get_vocab(model);
const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) {
if (token == LLAMA_TOKEN_NULL) {
if (default_template_src.find(jinja_variable_name) != std::string::npos ||
template_tool_use_src.find(jinja_variable_name) != std::string::npos) {
LOG_WRN(
"common_chat_templates_init: warning: vocab does not have a %s token, jinja template won't "
"work as intended.\n",
name);
}
return std::string();
}
return common_token_to_piece(vocab, token, true);
};
token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token");
token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token");
add_bos = llama_vocab_get_add_bos(vocab);
add_eos = llama_vocab_get_add_eos(vocab);
}
common_chat_templates_ptr tmpls(new common_chat_templates());
tmpls->has_explicit_template = has_explicit_template;
tmpls->add_bos = add_bos;
tmpls->add_eos = add_eos;
try {
tmpls->template_default = std::make_unique<common_chat_template>(default_template_src, token_bos, token_eos);
} catch (const std::exception & e) {
LOG_ERR("%s: error: %s\n", __func__, e.what());
LOG_ERR("%s: failed to initialize chat template\n", __func__);
LOG_ERR("%s: please consider disabling jinja via --no-jinja, or using another chat template\n", __func__);
throw e;
}
if (!template_tool_use_src.empty()) {
try {
tmpls->template_tool_use = std::make_unique<common_chat_template>(template_tool_use_src, token_bos, token_eos);
} catch (const std::exception & e) {
LOG_ERR("%s: failed to parse tool use chat template (ignoring it): %s\n", __func__, e.what());
}
}
return tmpls;
}
const char * common_chat_format_name(common_chat_format format) {
switch (format) {
case COMMON_CHAT_FORMAT_CONTENT_ONLY:
return "Content-only";
case COMMON_CHAT_FORMAT_PEG_SIMPLE:
return "peg-simple";
case COMMON_CHAT_FORMAT_PEG_NATIVE:
return "peg-native";
case COMMON_CHAT_FORMAT_PEG_GEMMA4:
return "peg-gemma4";
default:
throw std::runtime_error("Unknown chat format");
}
}
const char * common_reasoning_format_name(common_reasoning_format format) {
switch (format) {
case COMMON_REASONING_FORMAT_NONE:
return "none";
case COMMON_REASONING_FORMAT_AUTO:
return "auto";
case COMMON_REASONING_FORMAT_DEEPSEEK:
return "deepseek";
case COMMON_REASONING_FORMAT_DEEPSEEK_LEGACY:
return "deepseek-legacy";
default:
throw std::runtime_error("Unknown reasoning format");
}
}
common_reasoning_format common_reasoning_format_from_name(const std::string & format) {
if (format == "none") {
return COMMON_REASONING_FORMAT_NONE;
}
if (format == "auto") {
return COMMON_REASONING_FORMAT_AUTO;
}
if (format == "deepseek") {
return COMMON_REASONING_FORMAT_DEEPSEEK;
}
if (format == "deepseek-legacy") {
return COMMON_REASONING_FORMAT_DEEPSEEK_LEGACY;
}
throw std::runtime_error("Unknown reasoning format: " + format);
}
static void foreach_function(const json & tools, const std::function<void(const json &)> & fn) {
for (const auto & tool : tools) {
if (!tool.contains("type") || tool.at("type") != "function" || !tool.contains("function")) {
LOG_INF("Skipping tool without function: %s", tool.dump(2).c_str());
continue;
}
fn(tool);
}
}
static void foreach_parameter(const json & function,
const std::function<void(const std::string &, const json &, bool)> & fn) {
if (!function.contains("parameters") || !function.at("parameters").is_object()) {
return;
}
const auto & params = function.at("parameters");
if (!params.contains("properties") || !params.at("properties").is_object()) {
return;
}
const auto & props = params.at("properties");
std::set<std::string> required;
if (params.contains("required") && params.at("required").is_array()) {
params.at("required").get_to(required);
}
for (const auto & [name, prop] : props.items()) {
bool is_required = (required.find(name) != required.end());
fn(name, prop, is_required);
}
}
static std::string common_chat_template_direct_apply_impl(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs,
const std::optional<json> & messages_override = std::nullopt,
const std::optional<json> & tools_override = std::nullopt,
const std::optional<json> & additional_context = std::nullopt) {
jinja::context ctx(tmpl.source());
nlohmann::ordered_json inp = nlohmann::ordered_json{
{"messages", messages_override.has_value() ? *messages_override : inputs.messages},
{"bos_token", tmpl.bos_token()},
{"eos_token", tmpl.eos_token()},
{"enable_thinking", inputs.enable_thinking},
};
if (tools_override.has_value() || !inputs.tools.empty()) {
inp["tools"] = tools_override.has_value() ? *tools_override : inputs.tools;
}
if (inputs.extra_context.is_object()) {
// TODO: do we need to merge, or replacing is fine?
for (const auto & [k, v] : inputs.extra_context.items()) {
inp[k] = v;
}
}
if (additional_context.has_value()) {
// TODO: merge properly instead of overwriting (matching old behavior)
for (const auto & [k, v] : additional_context->items()) {
inp[k] = v;
}
}
if (inputs.add_generation_prompt) {
inp["add_generation_prompt"] = true;
}
jinja::global_from_json(ctx, inp, inputs.mark_input);
// render
jinja::runtime runtime(ctx);
const jinja::value results = runtime.execute(tmpl.prog);
auto parts = jinja::runtime::gather_string_parts(results);
std::string result = parts->as_string().str();
// TODO: improve this later
if (inputs.add_bos && string_starts_with(result, tmpl.bos_token())) {
result = result.substr(tmpl.bos_token().size());
}
if (inputs.add_eos && string_ends_with(result, tmpl.eos_token())) {
result = result.substr(0, result.size() - tmpl.eos_token().size());
}
return result;
}
std::string common_chat_template_direct_apply(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs) {
return common_chat_template_direct_apply_impl(tmpl, inputs, std::nullopt, std::nullopt, std::nullopt);
}
static std::string common_chat_template_generation_prompt_impl(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs,
const std::optional<json> & messages_override = std::nullopt,
const std::optional<json> & tools_override = std::nullopt,
const std::optional<json> & additional_context = std::nullopt) {
auto adjusted_messages = messages_override ? *messages_override : inputs.messages;
autoparser::generation_params params = inputs;
params.add_generation_prompt = false;
params.continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
std::string no_gen_prompt = common_chat_template_direct_apply_impl(tmpl, params, adjusted_messages, tools_override, additional_context);
params.add_generation_prompt = true;
std::string gen_prompt = common_chat_template_direct_apply_impl(tmpl, params, adjusted_messages, tools_override, additional_context);
size_t prefix_len = 0;
size_t min_size = std::min(no_gen_prompt.size(), gen_prompt.size());
while (prefix_len < min_size && no_gen_prompt[prefix_len] == gen_prompt[prefix_len]) {
prefix_len++;
}
return gen_prompt.substr(prefix_len);
}
std::string common_chat_template_generation_prompt(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs) {
return common_chat_template_generation_prompt_impl(tmpl, inputs, std::nullopt, std::nullopt, std::nullopt);
}
static common_chat_params common_chat_params_init_ministral_3(const common_chat_template & tmpl,
const autoparser::generation_params & inputs) {
common_chat_params data;
// Build up messages to follow the format: https://huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512/blob/main/chat_template.jinja
auto adjusted_messages = json::array();
for (const auto & msg : inputs.messages) {
auto role = msg.value("role", "");
if (role != "system" && role != "assistant") {
// Only adjust system and assistant messages. Interestingly, the system message may contain thinking.
adjusted_messages.push_back(msg);
continue;
}
auto content = json::array();
// If message contains `reasoning_content`, add it as a block of type `thinking`
if (msg.contains("reasoning_content") && msg.at("reasoning_content").is_string()) {
content.push_back({
{ "type", "thinking" },
{ "thinking", msg.at("reasoning_content").get<std::string>() },
});
}
// If message contains `content`, add it as a block of type `text`
if (msg.contains("content")) {
if (msg.at("content").is_string()) {
content.push_back({
{ "type", "text" },
{ "text", msg.at("content").get<std::string>() },
});
} else if (msg.at("content").is_array()) {
auto blocks = msg.at("content");
content.insert(content.end(), blocks.begin(), blocks.end());
}
}
auto adjusted = msg;
adjusted["content"] = content;
adjusted.erase("reasoning_content");
adjusted_messages.push_back(adjusted);
}
auto has_tools = inputs.tools.is_array() && !inputs.tools.empty();
auto has_response_format = inputs.json_schema.is_object() && !inputs.json_schema.empty();
auto extract_reasoning = inputs.reasoning_format != COMMON_REASONING_FORMAT_NONE;
auto include_grammar = true;
data.supports_thinking = true;
data.thinking_start_tag = "[THINK]";
data.thinking_end_tag = "[/THINK]";
data.prompt = common_chat_template_direct_apply_impl(tmpl, inputs, /* messages_override = */ adjusted_messages);
data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs, /* messages_override = */ adjusted_messages);
data.format = COMMON_CHAT_FORMAT_PEG_NATIVE;
data.preserved_tokens = {
"[THINK]",
"[/THINK]",
"[TOOL_CALLS]",
"[ARGS]",
};
if (inputs.has_continuation()) {
const auto & msg = inputs.continue_msg;
data.generation_prompt = "[THINK]" + msg.reasoning_content;
if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
data.generation_prompt += "[/THINK]" + msg.render_content();
}
data.prompt += data.generation_prompt;
}
auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
auto generation_prompt = p.eps();
auto reasoning =
extract_reasoning ? p.optional("[THINK]" + p.reasoning(p.until("[/THINK]")) + "[/THINK]") : p.eps();
// Response format parser
if (has_response_format) {
// Ministral wants to emit json surrounded by code fences
return generation_prompt + (reasoning << "```json" << p.content(p.schema(p.json(), "response-format", inputs.json_schema)) << "```");
}
You can’t perform that action at this time.