optionally allow any spaces in json schema grammars (useful for llama3 8b tool outputs)

Author: ochafik

common/chat.cpp

@@ -451,8 +451,7 @@ std::string common_chat_format_name(common_chat_format format) {
 
 const common_grammar_options grammar_options {
     /* .dotall = */ false,
-    /* .compact_spaces = */ false,
-    // /* .compact_spaces = */ true,
+    /* .any_spaces = */ true,
 };
 
 static std::optional<json> parse_json(std::string::const_iterator & it, const std::string::const_iterator & end) {

common/json-schema-to-grammar.cpp

@@ -765,10 +765,10 @@ class SchemaConverter {
     SchemaConverter(
         const std::function<json(const std::string &)> & fetch_json,
         bool dotall,
-        bool compact_spaces)
+        bool any_spaces)
           : _fetch_json(fetch_json), _dotall(dotall)
     {
-        _rules["space"] = compact_spaces ? "\" \"?" : SPACE_RULE;
+        _rules["space"] = any_spaces ? "[ \\t\\n]*" : SPACE_RULE;
     }
 
     void resolve_refs(json & schema, const std::string & url) {
@@ -1007,7 +1007,7 @@ std::string json_schema_to_grammar(const json & schema, bool force_gbnf) {
 }
 
 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options) {
-    SchemaConverter converter([&](const std::string &) { return json(); }, options.dotall, options.compact_spaces);
+    SchemaConverter converter([&](const std::string &) { return json(); }, options.dotall, options.any_spaces);
     common_grammar_builder builder {
         /* .add_rule = */ [&](const std::string & name, const std::string & rule) {
             return converter._add_rule(name, rule);

common/json-schema-to-grammar.h

@@ -16,7 +16,7 @@ struct common_grammar_builder {
 
 struct common_grammar_options {
     bool dotall = false;
-    bool compact_spaces = false;
+    bool any_spaces = false;
 };
 
 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options = {});

scripts/tool_bench.sh

@@ -8,7 +8,6 @@ cmake --build build -j
 
 export LLAMA_CACHE=${LLAMA_CACHE:-$HOME/Library/Caches/llama.cpp}
 export LLAMA_SERVER_BIN_PATH=$PWD/build/bin/llama-server
-export LLAMA_SERVER_BASELINE=$(which llama-server)
 
 if [ ! -x "$LLAMA_SERVER_BIN_PATH" ]; then
     echo "Could not find llama-server binary at $LLAMA_SERVER_BIN_PATH"
@@ -20,7 +19,7 @@ if [ ! -d "$LLAMA_CACHE" ]; then
 fi
 
 export ARGS=(
-    --llama-baseline="$LLAMA_SERVER_BASELINE"
+    --llama-baseline="$(which llama-server)"
     --n 30
     --temp -1  # Leaves temperature parameter unset (use the server's default, e.g. 0.6 for ollama)
     --temp 0