llama_cpp 0.16.1 → 0.16.2

data/vendor/tmp/llama.cpp/unicode.cpp

@@ -226,8 +226,9 @@ static std::vector<size_t> unicode_regex_split_custom_gpt2(const std::string & t
         assert(offset_end <= cpts.size());
         start = offset_end;
 
-        auto _get_cpt = [&] (const size_t pos) -> char32_t {
-            return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : 0;
+        static const uint32_t OUT_OF_RANGE = 0xFFFFFFFF;
+        auto _get_cpt = [&] (const size_t pos) -> uint32_t {
+            return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : OUT_OF_RANGE;
         };
 
         auto _get_flags = [&] (const size_t pos) -> codepoint_flags {
@@ -253,18 +254,18 @@ static std::vector<size_t> unicode_regex_split_custom_gpt2(const std::string & t
         };
 
         for (size_t pos = offset_ini; pos < offset_end; /*pos++*/ ) {
-            const char32_t cpt = _get_cpt(pos);
+            const uint32_t cpt = _get_cpt(pos);
             const auto flags = _get_flags(pos);
 
             // regex: 's|'t|'re|'ve|'m|'ll|'d
             if (cpt == '\'' && pos+1 < offset_end) {
-                char32_t cpt_next = _get_cpt(pos+1);
+                uint32_t cpt_next = _get_cpt(pos+1);
                 if (cpt_next == 's' || cpt_next == 't' || cpt_next == 'm' || cpt_next == 'd') {
                     pos += _add_token(pos+2);
                     continue;
                 }
                 if (pos+2 < offset_end) {
-                    char32_t cpt_next_next = _get_cpt(pos+2);
+                    uint32_t cpt_next_next = _get_cpt(pos+2);
                     if ((cpt_next == 'r' && cpt_next_next == 'e') ||
                         (cpt_next == 'v' && cpt_next_next == 'e') ||
                         (cpt_next == 'l' && cpt_next_next == 'l')) {
@@ -309,7 +310,7 @@ static std::vector<size_t> unicode_regex_split_custom_gpt2(const std::string & t
             }
 
             // regex: \s+(?!\S)
-            if (num_whitespaces > 1 && _get_cpt(pos+num_whitespaces) != 0) {
+            if (num_whitespaces > 1 && _get_cpt(pos+num_whitespaces) != OUT_OF_RANGE) {
                 pos += num_whitespaces - 1;
                 _add_token(pos);
                 continue;
@@ -344,8 +345,9 @@ static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string &
         assert(offset_end <= cpts.size());
         start = offset_end;
 
-        auto _get_cpt = [&] (const size_t pos) -> char32_t {
-            return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : 0;
+        static const uint32_t OUT_OF_RANGE = 0xFFFFFFFF;
+        auto _get_cpt = [&] (const size_t pos) -> uint32_t {
+            return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : OUT_OF_RANGE;
         };
 
         auto _get_flags = [&] (const size_t pos) -> codepoint_flags {
@@ -371,18 +373,18 @@ static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string &
         };
 
         for (size_t pos = offset_ini; pos < offset_end; /*pos++*/ ) {
-            const char32_t cpt = _get_cpt(pos);
+            const uint32_t cpt = _get_cpt(pos);
             const auto flags = _get_flags(pos);
 
             // regex: (?i:'s|'t|'re|'ve|'m|'ll|'d) // case insensitive
             if (cpt == '\'' && pos+1 < offset_end) {
-                char32_t cpt_next = unicode_tolower(_get_cpt(pos+1));
+                uint32_t cpt_next = unicode_tolower(_get_cpt(pos+1));
                 if (cpt_next == 's' || cpt_next == 't' || cpt_next == 'm' || cpt_next == 'd') {
                     pos += _add_token(pos+2);
                     continue;
                 }
                 if (pos+2 < offset_end) {
-                    char32_t cpt_next_next = unicode_tolower(_get_cpt(pos+2));
+                    uint32_t cpt_next_next = unicode_tolower(_get_cpt(pos+2));
                     if ((cpt_next == 'r' && cpt_next_next == 'e') ||
                         (cpt_next == 'v' && cpt_next_next == 'e') ||
                         (cpt_next == 'l' && cpt_next_next == 'l')) {
@@ -424,7 +426,7 @@ static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string &
                 while (!(flags2.is_whitespace || flags2.is_letter || flags2.is_number || flags2.is_undefined)) {
                     flags2 = _get_flags(++pos);
                 }
-                char32_t cpt2 = _get_cpt(pos);
+                uint32_t cpt2 = _get_cpt(pos);
                 while (cpt2 == '\r' || cpt2 == '\n') {
                     cpt2 = _get_cpt(++pos);
                 }
@@ -435,7 +437,7 @@ static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string &
             size_t num_whitespaces = 0;
             size_t last_end_r_or_n = 0;
             while (_get_flags(pos+num_whitespaces).is_whitespace) {
-                char32_t cpt2 = _get_cpt(pos+num_whitespaces);
+                uint32_t cpt2 = _get_cpt(pos+num_whitespaces);
                 if (cpt2 == '\r' || cpt2 == '\n') {
                     last_end_r_or_n = pos + num_whitespaces + 1;
                 }
@@ -450,7 +452,7 @@ static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string &
             }
 
             // regex: \s+(?!\S)
-            if (num_whitespaces > 1 && _get_cpt(pos+num_whitespaces) != 0) {
+            if (num_whitespaces > 1 && _get_cpt(pos+num_whitespaces) != OUT_OF_RANGE) {
                 pos += num_whitespaces - 1;
                 _add_token(pos);
                 continue;
@@ -594,6 +596,7 @@ std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & c
 
 std::vector<uint32_t> unicode_cpts_from_utf8(const std::string & utf8) {
     std::vector<uint32_t> result;
+    result.reserve(utf8.size());
     size_t offset = 0;
     while (offset < utf8.size()) {
         result.push_back(unicode_cpt_from_utf8(utf8, offset));
@@ -626,7 +629,7 @@ uint8_t unicode_utf8_to_byte(const std::string & utf8) {
     return map.at(utf8);
 }
 
-char32_t unicode_tolower(char32_t cp) {
+uint32_t unicode_tolower(uint32_t cp) {
     auto it = unicode_map_lowercase.find(cp);
     return it == unicode_map_lowercase.end() ? cp : it->second;
 }
@@ -679,10 +682,14 @@ std::vector<std::string> unicode_regex_split(const std::string & text, const std
                 continue;
             }
 
-            const int cpt_flag = unicode_cpt_flags(cpts[i]).category_flag();
+            const auto flags = unicode_cpt_flags(cpts[i]);
 
-            if (k_ucat_cpt.find(cpt_flag) != k_ucat_cpt.end()) {
-                text_collapsed[i] = k_ucat_cpt.at(cpt_flag);
+            if (flags.is_whitespace) {
+                //NOTE: C++ std::regex \s does not mach 0x85, Rust and Python regex does.
+                //text_collapsed[i] = (char) 0x85;  // <Next Line> as whitespace fallback
+                text_collapsed[i] = (char) 0x0B;    // <vertical tab> as whitespace fallback
+            } else if (k_ucat_cpt.find(flags.category_flag()) != k_ucat_cpt.end()) {
+                text_collapsed[i] = k_ucat_cpt.at(flags.category_flag());
             } else {
                 text_collapsed[i] = (char) 0xD0; // fallback
             }
@@ -766,9 +773,16 @@ std::vector<std::string> unicode_regex_split(const std::string & text, const std
                 bpe_offsets = unicode_regex_split_stl(text_collapsed, regex_expr_collapsed, bpe_offsets);
             } else {
                 // no unicode category used, we can use std::wregex directly
-                const std::wstring wtext       = unicode_wstring_from_utf8(text);
                 const std::wstring wregex_expr = unicode_wstring_from_utf8(regex_expr);
 
+                // std::wregex \s does not mach non-ASCII whitespaces, using 0x0B as fallback
+                std::wstring wtext(cpts.begin(), cpts.end());
+                for (size_t i = 0; i < wtext.size(); ++i) {
+                    if (wtext[i] > 0x7F && unicode_cpt_flags(wtext[i]).is_whitespace) {
+                        wtext[i] = 0x0B;
+                    }
+                }
+
                 //printf("text: %s\n", text.c_str());
                 //printf("regex_expr: %s\n", regex_expr.c_str());
                 bpe_offsets = unicode_regex_split_stl(wtext, wregex_expr, bpe_offsets);

data/vendor/tmp/llama.cpp/unicode.h

@@ -58,6 +58,6 @@ codepoint_flags unicode_cpt_flags(const std::string & utf8);
 std::string unicode_byte_to_utf8(uint8_t byte);
 uint8_t unicode_utf8_to_byte(const std::string & utf8);
 
-char32_t unicode_tolower(char32_t cp);
+uint32_t unicode_tolower(uint32_t cp);
 
 std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs);

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: llama_cpp
 version: !ruby/object:Gem::Version
-  version: 0.16.1
+  version: 0.16.2
 platform: ruby
 authors:
 - yoshoku
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2024-06-15 00:00:00.000000000 Z
+date: 2024-06-22 00:00:00.000000000 Z
 dependencies: []
 description: llama_cpp.rb provides Ruby bindings for the llama.cpp.
 email: