From f99e1e456eaf69cc38c1982a2693ce41c0f897ef Mon Sep 17 00:00:00 2001
From: Haoxiang Fei <tonyfettes@tonyfettes.com>
Date: Sat, 11 May 2024 16:12:06 +0800
Subject: llama : lookup word in vocab before doing BPE merges (#7193)

* fix: llama-3 ignore_merges

* test: add test for llama-3 bpe ignore_merges

* fix: set ignore_merges only for llama-3

* fix: test-tokenizer-1-bpe --ingore-merges detection

* fix: copy to fix fallthrough

* fix: change ignore_merges to bool

* fix: add ignore merges tests to cmake

* llama : alternative merge ignore logic

---------

Co-authored-by: Haoxiang Fei <feihaoxiang@idea.edu.cn>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---
 llama.cpp | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

(limited to 'llama.cpp')

diff --git a/llama.cpp b/llama.cpp
index cdff28cd..e91ad728 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -12253,13 +12253,14 @@ struct llm_tokenizer_bpe {
 
     void tokenize(const std::string & text, std::vector<llama_vocab::id> & output) {
         int final_prev_index = -1;
+        bool ignore_merges = false;
 
         std::vector<std::string> word_collection;
         switch (vocab.type) {
             case LLAMA_VOCAB_TYPE_BPE:
                 switch (vocab.type_pre) {
                     case LLAMA_VOCAB_PRE_TYPE_LLAMA3:
-                    case LLAMA_VOCAB_PRE_TYPE_DBRX:
+                        ignore_merges = true;
                         word_collection = unicode_regex_split(text, {
                             // original regex from tokenizer.json
                             //"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
@@ -12268,6 +12269,12 @@ struct llm_tokenizer_bpe {
                             "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
                         });
                         break;
+                    case LLAMA_VOCAB_PRE_TYPE_DBRX:
+                        word_collection = unicode_regex_split(text, {
+                            // same as llama3
+                            "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+                        });
+                        break;
                     case LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM:
                         word_collection = unicode_regex_split(text, {
                             "[\r\n]",
@@ -12351,6 +12358,11 @@ struct llm_tokenizer_bpe {
             int index = 0;
             size_t offset = 0;
 
+            if (ignore_merges && vocab.token_to_id.find(word) != vocab.token_to_id.end()) {
+                symbols.emplace_back(llm_symbol{-1, -1, word.c_str(), word.size()});
+                offset = word.size();
+            }
+
             while (offset < word.size()) {
                 llm_symbol sym;
                 size_t char_len = std::min(word.size() - offset, (size_t) ::utf8_len(word[offset]));
-- 
cgit v1.2.3