model : add Kanana-2 model support (#19803)

* model: Add Kanana-2 model support

* lint: adjust spacing

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index 31acd5bb48546b5aeb4a43d32ad97cd3eeda9f9a..e038109599a578a6dc4d3f4083a03926e5858115 100755 (executable)
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -1274,6 +1274,9 @@ class TextModel(ModelBase):
         if chkhsh == "b4b8ca1f9769494fbd956ebc4c249de6131fb277a4a3345a7a92c7dd7a55808d":
             # ref: https://huggingface.co/jdopensource/JoyAI-LLM-Flash
             res = "joyai-llm"
+        if chkhsh == "e4d54df1ebc1f2b91acd986c5b51aa50837d5faf7c7398e73c1f9e9ee5d19869":
+            # ref: https://huggingface.co/kakaocorp/kanana-2-30b-a3b-instruct-2601
+            res = "kanana2"
 
         if res is None:
             logger.warning("\n")

diff --git a/convert_hf_to_gguf_update.py b/convert_hf_to_gguf_update.py
index 8f7443d1b579a3a195bf4d449a2c694dc91d2f32..53a73759ec14937d3fefe112172a47a23cd458e1 100755 (executable)
--- a/convert_hf_to_gguf_update.py
+++ b/convert_hf_to_gguf_update.py
@@ -152,6 +152,7 @@ models = [
     {"name": "exaone-moe",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B", },
     {"name": "qwen35",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen3.5-9B-Instruct", },
     {"name": "joyai-llm",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jdopensource/JoyAI-LLM-Flash", },
+    {"name": "kanana2",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/kakaocorp/kanana-2-30b-a3b-instruct-2601", },
 ]
 
 # some models are known to be broken upstream, so we will skip them as exceptions

diff --git a/src/llama-model.cpp b/src/llama-model.cpp
index 764839b9bc86445f3816fdb54af039342e3adac3..c93e29555b39d8628b2b7ec4b9c29840da7bb0ae 100644 (file)
--- a/src/llama-model.cpp
+++ b/src/llama-model.cpp
@@ -1703,8 +1703,8 @@ void llama_model::load_hparams(llama_model_loader & ml) {
             } break;
         case LLM_ARCH_DEEPSEEK2:
             {
-                // lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
-                const bool is_lite = (hparams.n_layer == 27 || hparams.n_layer == 26);
+                // lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B, Kanana-2-30B-A3B
+                const bool is_lite = (hparams.n_layer == 27 || hparams.n_layer == 26 || (hparams.n_layer == 48 && n_vocab == 128256));
 
                 ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
                 ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT,   hparams.n_layer_dense_lead);

diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp
index 69b25a1bf9f59f4db9b65e67d6577f4b7a9347a7..9c118eab7e08706f8db9ad728f9cd990c5a6e969 100644 (file)
--- a/src/llama-vocab.cpp
+++ b/src/llama-vocab.cpp
@@ -2027,7 +2027,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_QWEN2;
             } else if (
                 tokenizer_pre == "gpt-4o" ||
-                tokenizer_pre == "llama4") {
+                tokenizer_pre == "llama4" ||
+                tokenizer_pre == "kanana2") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_GPT4O;
                 clean_spaces = false;
             } else if (