common_grammar_trigger: always use string value (+ optional token)

ochafik · Olivier Chafik · commit e6e9c138c6d2 · 2025-03-04T19:07:42.000Z
diff --git a/common/common.h b/common/common.h
@@ -120,7 +120,8 @@ enum common_grammar_trigger_type {
 
 struct common_grammar_trigger {
     common_grammar_trigger_type type;
-    std::variant<llama_token, std::string> value;
+    std::string value;
+    llama_token token = LLAMA_TOKEN_NULL;
 };
 
 // sampling parameters
diff --git a/common/sampling.cpp b/common/sampling.cpp
@@ -166,20 +166,20 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
             switch (trigger.type) {
                 case COMMON_GRAMMAR_TRIGGER_TYPE_WORD:
                 {
-                    const auto & word = std::get<std::string>(trigger.value);
+                    const auto & word = trigger.value;
                     patterns_anywhere.push_back(regex_escape(word));
                     break;
                 }
                 case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN:
                 case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START:
                 {
-                    const auto & pattern = std::get<std::string>(trigger.value);
+                    const auto & pattern = trigger.value;
                     (trigger.type == COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START ? patterns_at_start : patterns_anywhere).push_back(pattern);
                     break;
                 }
                 case COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN:
                 {
-                    const auto & token = std::get<llama_token>(trigger.value);
+                    const auto token = trigger.token;
                     trigger_tokens.push_back(token);
                     break;
                 }
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
@@ -135,16 +135,16 @@ struct slot_params {
         for (const auto & trigger : sampling.grammar_triggers) {
             switch (trigger.type) {
                 case COMMON_GRAMMAR_TRIGGER_TYPE_WORD:
-                    grammar_triggers.push_back({{"word", std::get<std::string>(trigger.value)}});
+                    grammar_triggers.push_back({{"word", trigger.value}});
                     break;
                 case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN:
-                    grammar_triggers.push_back({{"pattern", std::get<std::string>(trigger.value)}});
+                    grammar_triggers.push_back({{"pattern", trigger.value}});
                     break;
                 case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START:
-                    grammar_triggers.push_back({{"pattern_start", std::get<std::string>(trigger.value)}});
+                    grammar_triggers.push_back({{"pattern_start", trigger.value}});
                     break;
                 case COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN:
-                    grammar_triggers.push_back({{"token", std::get<llama_token>(trigger.value)}});
+                    grammar_triggers.push_back({{"token", trigger.token}});
                     break;
             }
         }
diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp
@@ -623,9 +623,7 @@ static json oaicompat_completion_params_parse(
     for (const auto & trigger : chat_params.grammar_triggers) {
         grammar_triggers.push_back({
             {"type", (int) trigger.type},
-            {"value", trigger.type == COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN
-                ? json((int) std::get<llama_token>(trigger.value))
-                : json(std::get<std::string>(trigger.value))},
+            {"value", trigger.token},
         });
     }
     llama_params["grammar_triggers"] = grammar_triggers;
diff --git a/tests/test-chat.cpp b/tests/test-chat.cpp
@@ -242,21 +242,21 @@ static void test_templates(const struct common_chat_templates * tmpls, const std
                 switch (trigger.type) {
                     case COMMON_GRAMMAR_TRIGGER_TYPE_WORD:
                     {
-                        const auto & word = std::get<std::string>(trigger.value);
+                        const auto & word = trigger.value;
                         pos = constrained.find(word);
                         break;
                     }
                     case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN:
                     {
-                        const auto & pattern = std::get<std::string>(trigger.value);
+                        const auto & pattern = trigger.value;
                         if (std::regex_search(constrained, match, std::regex(pattern))) {
                             pos = match.position();
                         }
                         break;
                     }
                     case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START:
                     {
-                        const auto & pattern = std::get<std::string>(trigger.value);
+                        const auto & pattern = trigger.value;
                         if (std::regex_search(constrained, match, std::regex(pattern)) && match.position() == 0) {
                             pos = 0;
                         }

Original file line number	Diff line number	Diff line change
`@@ -166,20 +166,20 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co`
`166`	`166`	`switch (trigger.type) {`
`167`	`167`	`case COMMON_GRAMMAR_TRIGGER_TYPE_WORD:`
`168`	`168`	`{`
`169`		`- const auto & word = std::get<std::string>(trigger.value);`
	`169`	`+ const auto & word = trigger.value;`
`170`	`170`	`patterns_anywhere.push_back(regex_escape(word));`
`171`	`171`	`break;`
`172`	`172`	`}`
`173`	`173`	`case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN:`
`174`	`174`	`case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START:`
`175`	`175`	`{`
`176`		`- const auto & pattern = std::get<std::string>(trigger.value);`
	`176`	`+ const auto & pattern = trigger.value;`
`177`	`177`	`(trigger.type == COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START ? patterns_at_start : patterns_anywhere).push_back(pattern);`
`178`	`178`	`break;`
`179`	`179`	`}`
`180`	`180`	`case COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN:`
`181`	`181`	`{`
`182`		`- const auto & token = std::get<llama_token>(trigger.value);`
	`182`	`+ const auto token = trigger.token;`
`183`	`183`	`trigger_tokens.push_back(token);`
`184`	`184`	`break;`
`185`	`185`	`}`
Original file line number	Diff line number	Diff line change
`@@ -242,21 +242,21 @@ static void test_templates(const struct common_chat_templates * tmpls, const std`
`242`	`242`	`switch (trigger.type) {`
`243`	`243`	`case COMMON_GRAMMAR_TRIGGER_TYPE_WORD:`
`244`	`244`	`{`
`245`		`- const auto & word = std::get<std::string>(trigger.value);`
	`245`	`+ const auto & word = trigger.value;`
`246`	`246`	`pos = constrained.find(word);`
`247`	`247`	`break;`
`248`	`248`	`}`
`249`	`249`	`case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN:`
`250`	`250`	`{`
`251`		`- const auto & pattern = std::get<std::string>(trigger.value);`
	`251`	`+ const auto & pattern = trigger.value;`
`252`	`252`	`if (std::regex_search(constrained, match, std::regex(pattern))) {`
`253`	`253`	`pos = match.position();`
`254`	`254`	`}`
`255`	`255`	`break;`
`256`	`256`	`}`
`257`	`257`	`case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START:`
`258`	`258`	`{`
`259`		`- const auto & pattern = std::get<std::string>(trigger.value);`
	`259`	`+ const auto & pattern = trigger.value;`
`260`	`260`	`if (std::regex_search(constrained, match, std::regex(pattern)) && match.position() == 0) {`
`261`	`261`	`pos = 0;`
`262`	`262`	`}`