Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

Update llama.cpp submodule to `a8db2a9c` (#5530)

Jeffrey Morgan571dc61955ce
diff --git a/llm/llama.cpp b/llm/llama.cpp
index d7fd29fff..a8db2a9ce 160000
--- a/llm/llama.cpp
+++ b/llm/llama.cpp
@@ -1 +1 @@
-Subproject commit d7fd29fff16456ce9c3a23fd2d09a66256b05aff
+Subproject commit a8db2a9ce64cd4417f6a312ab61858f17f0f8584
diff --git a/llm/patches/05-default-pretokenizer.diff b/llm/patches/05-default-pretokenizer.diff
index f4eaced72..341a6f590 100644
--- a/llm/patches/05-default-pretokenizer.diff
+++ b/llm/patches/05-default-pretokenizer.diff
@@ -1,11 +1,11 @@
diff --git a/src/llama.cpp b/src/llama.cpp
-index 73f52435..2b81b4bd 100644
+index 2b9ace28..172640e2 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
-@@ -5092,16 +5092,7 @@ static void llm_load_vocab(
-
- // for now, only BPE models have pre-tokenizers
+@@ -5357,16 +5357,7 @@ static void llm_load_vocab(
if (vocab.type == LLAMA_VOCAB_TYPE_BPE) {
+ vocab.tokenizer_add_space_prefix = false;
+ vocab.tokenizer_clean_spaces = true;
- if (tokenizer_pre.empty()) {
- LLAMA_LOG_WARN("%s: missing pre-tokenizer type, using: 'default'\n", __func__);
- LLAMA_LOG_WARN("%s: \n", __func__);
@@ -20,7 +20,7 @@ index 73f52435..2b81b4bd 100644
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
} else if (
tokenizer_pre == "llama3" ||
-@@ -5164,7 +5155,8 @@ static void llm_load_vocab(
+@@ -5439,7 +5430,8 @@ static void llm_load_vocab(
tokenizer_pre == "jais") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_JAIS;
} else {