update llama.cpp submodule to d7fd29f (#5475)

This commit is contained in:
Jeffrey Morgan
2024-07-05 13:25:58 -04:00
committed by GitHub
parent d89454de80
commit 8f8e736b13
15 changed files with 150 additions and 422 deletions

View File

@@ -1,8 +1,8 @@
diff --git a/llama.cpp b/llama.cpp
index 61948751..4b72a293 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -4824,16 +4824,7 @@ static void llm_load_vocab(
diff --git a/src/llama.cpp b/src/llama.cpp
index 73f52435..2b81b4bd 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
if (vocab.type == LLAMA_VOCAB_TYPE_BPE) {
@@ -20,13 +20,13 @@ index 61948751..4b72a293 100644
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
} else if (
tokenizer_pre == "llama3" ||
@@ -4888,7 +4879,8 @@ static void llm_load_vocab(
tokenizer_pre == "poro-chat") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_PORO;
@@ -5164,7 +5155,8 @@ static void llm_load_vocab(
tokenizer_pre == "jais") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_JAIS;
} else {
- throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
+ LLAMA_LOG_WARN("%s: missing or unrecognized pre-tokenizer type, using: 'default'\n", __func__);
+ vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
}
} else {
} else if (vocab.type == LLAMA_VOCAB_TYPE_SPM) {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;