cui-llama.rn 1.2.1 → 1.2.3

package/cpp/llama.h

@@ -103,6 +103,7 @@ extern "C" {
         LLAMA_VOCAB_PRE_TYPE_BLOOM          = 23,
         LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH   = 24,
         LLAMA_VOCAB_PRE_TYPE_EXAONE         = 25,
+        LLAMA_VOCAB_PRE_TYPE_CHAMELEON      = 26,
     };
 
     enum llama_rope_type {
@@ -193,6 +194,7 @@ extern "C" {
         LLAMA_POOLING_TYPE_MEAN = 1,
         LLAMA_POOLING_TYPE_CLS  = 2,
         LLAMA_POOLING_TYPE_LAST = 3,
+        LLAMA_POOLING_TYPE_RANK = 4, // used by reranking models to attach the classification head to the graph
     };
 
     enum llama_attention_type {
@@ -202,9 +204,9 @@ extern "C" {
     };
 
     enum llama_split_mode {
-        LLAMA_SPLIT_MODE_NONE    = 0, // single GPU
-        LLAMA_SPLIT_MODE_LAYER   = 1, // split layers and KV across GPUs
-        LLAMA_SPLIT_MODE_ROW     = 2, // split rows across GPUs
+        LLAMA_SPLIT_MODE_NONE  = 0, // single GPU
+        LLAMA_SPLIT_MODE_LAYER = 1, // split layers and KV across GPUs
+        LLAMA_SPLIT_MODE_ROW   = 2, // split rows across GPUs
     };
 
     // TODO: simplify (https://github.com/ggerganov/llama.cpp/pull/9294#pullrequestreview-2286561979)
@@ -872,7 +874,8 @@ extern "C" {
 
     // Get the embeddings for a sequence id
     // Returns NULL if pooling_type is LLAMA_POOLING_TYPE_NONE
-    // shape: [n_embd] (1-dimensional)
+    // when pooling_type == LLAMA_POOLING_TYPE_RANK, returns float[1] with the rank of the sequence
+    // otherwise: float[n_embd] (1-dimensional)
     LLAMA_API float * llama_get_embeddings_seq(struct llama_context * ctx, llama_seq_id seq_id);
 
     //
@@ -911,6 +914,8 @@ extern "C" {
     //
     // Tokenization
     //
+    // The API is thread-safe.
+    //
 
     /// @details Convert the provided text into tokens.
     /// @param tokens The tokens pointer must be large enough to hold the resulting tokens.

package/cpp/unicode-data.cpp

@@ -7,7 +7,7 @@
 #include <unordered_map>
 #include <unordered_set>
 
-const std::vector<std::pair<uint32_t, uint16_t>> unicode_ranges_flags = {  // start, flags // last=next_start-1
+const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags = {  // start, flags // last=next_start-1
 {0x000000, 0x0080},
 {0x000020, 0x0008},
 {0x000021, 0x0020},
@@ -2311,7 +2311,8 @@ const std::unordered_set<uint32_t> unicode_set_whitespace = {
 0x003000,
 };
 
-const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase = {
+// list is always in ascending order, to enable binary search
+const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase = {
 {0x000041, 0x000061},
 {0x000042, 0x000062},
 {0x000043, 0x000063},
@@ -3747,7 +3748,8 @@ const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase = {
 {0x01E921, 0x01E943},
 };
 
-const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase = {
+// list is always in ascending order, to enable binary search
+const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase = {
 {0x000061, 0x000041},
 {0x000062, 0x000042},
 {0x000063, 0x000043},
@@ -5200,7 +5202,7 @@ const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase = {
 {0x01E943, 0x01E921},
 };
 
-const std::vector<range_nfd> unicode_ranges_nfd = {  // start, last, nfd
+const std::initializer_list<range_nfd> unicode_ranges_nfd = {  // start, last, nfd
 {0x000000, 0x000000, 0x000000},
 {0x0000C0, 0x0000C5, 0x000041},
 {0x0000C7, 0x0000C7, 0x000043},

package/cpp/unicode-data.h

@@ -13,8 +13,8 @@ struct range_nfd {
 
 static const uint32_t MAX_CODEPOINTS = 0x110000;
 
-extern const std::vector<std::pair<uint32_t, uint16_t>> unicode_ranges_flags;
+extern const std::initializer_list<std::pair<uint32_t, uint16_t>> unicode_ranges_flags;
 extern const std::unordered_set<uint32_t> unicode_set_whitespace;
-extern const std::unordered_map<uint32_t, uint32_t> unicode_map_lowercase;
-extern const std::unordered_map<uint32_t, uint32_t> unicode_map_uppercase;
-extern const std::vector<range_nfd> unicode_ranges_nfd;
+extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase;
+extern const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase;
+extern const std::initializer_list<range_nfd> unicode_ranges_nfd;

package/cpp/unicode.cpp

@@ -123,11 +123,11 @@ uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset) {
 static std::vector<codepoint_flags> unicode_cpt_flags_array() {
     std::vector<codepoint_flags> cpt_flags(MAX_CODEPOINTS, codepoint_flags::UNDEFINED);
 
-    assert (unicode_ranges_flags.front().first == 0);
-    assert (unicode_ranges_flags.back().first == MAX_CODEPOINTS);
+    assert (unicode_ranges_flags.begin()[0].first == 0);
+    assert (unicode_ranges_flags.begin()[unicode_ranges_flags.size()-1].first == MAX_CODEPOINTS);
     for (size_t i = 1; i < unicode_ranges_flags.size(); ++i) {
-        const auto range_ini = unicode_ranges_flags[i-1];  // codepoint_ini, flags
-        const auto range_end = unicode_ranges_flags[i];    // codepoint_end, flags
+        const auto range_ini = unicode_ranges_flags.begin()[i-1];  // codepoint_ini, flags
+        const auto range_end = unicode_ranges_flags.begin()[i];    // codepoint_end, flags
         for (uint32_t cpt = range_ini.first; cpt < range_end.first; ++cpt) {
             cpt_flags[cpt] = range_ini.second;
         }
@@ -597,7 +597,7 @@ std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & c
     std::vector<uint32_t> result(cpts.size());
     for (size_t i = 0; i < cpts.size(); ++i) {
         const uint32_t cpt = cpts[i];
-        auto it = std::upper_bound(unicode_ranges_nfd.cbegin(), unicode_ranges_nfd.cend(), cpt, comp) - 1;
+        auto it = std::upper_bound(unicode_ranges_nfd.begin(), unicode_ranges_nfd.end(), cpt, comp) - 1;
         result[i] = (it->first <= cpt && cpt <= it->last) ? it->nfd : cpt;
     }
     return result;
@@ -639,8 +639,15 @@ uint8_t unicode_utf8_to_byte(const std::string & utf8) {
 }
 
 uint32_t unicode_tolower(uint32_t cp) {
-    auto it = unicode_map_lowercase.find(cp);
-    return it == unicode_map_lowercase.end() ? cp : it->second;
+    // binary search
+    auto it = std::lower_bound(unicode_map_lowercase.begin(), unicode_map_lowercase.end(), cp,
+        [](const std::pair<uint32_t, uint32_t> & pair, uint32_t value) {
+            return pair.first < value;
+        });
+    if (it != unicode_map_lowercase.end() && it->first == cp) {
+        return it->second;
+    }
+    return cp;  // Return the original code point if no lowercase mapping is found
 }
 
 std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cui-llama.rn",
-  "version": "1.2.1",
+  "version": "1.2.3",
   "description": "Fork of llama.rn for ChatterUI",
   "main": "lib/commonjs/index",
   "module": "lib/module/index",