Skip to content

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

llama.cpp update (#16960)

Daniel Hiltgen1c5ebbf5f4a4
diff --git a/LLAMA_CPP_VERSION b/LLAMA_CPP_VERSION
index 8d1d9a867..151c77373 100644
--- a/LLAMA_CPP_VERSION
+++ b/LLAMA_CPP_VERSION
@@ -1 +1 @@
-b9781
+b9840
diff --git a/llama/compat/models/003-llama-cpp-laguna.patch b/llama/compat/models/003-llama-cpp-laguna.patch
index 13e162bf1..e6536a731 100644
--- a/llama/compat/models/003-llama-cpp-laguna.patch
+++ b/llama/compat/models/003-llama-cpp-laguna.patch
@@ -5,7 +5,7 @@ diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
{ LLM_ARCH_MELLUM, "mellum" },
+ { LLM_ARCH_LAGUNA, "laguna" },
{ LLM_ARCH_UNKNOWN, "(unknown)" },
-@@ -380,2 +381,3 @@ static const std::map<llm_tensor, const char *> LLM_TENSOR_NAMES = {
+@@ -398,2 +399,3 @@ static const std::map<llm_tensor, const char *> LLM_TENSOR_NAMES = {
{ LLM_TENSOR_ATTN_GATE, "blk.%d.attn_gate" },
+ { LLM_TENSOR_ATTN_GATE_LAGUNA, "blk.%d.attn_g" },
{ LLM_TENSOR_FFN_POST_NORM, "blk.%d.post_ffw_norm" },
@@ -17,7 +17,7 @@ diff --git a/src/llama-arch.h b/src/llama-arch.h
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -145,2 +145,3 @@ enum llm_arch {
- LLM_ARCH_EAGLE3,
+ LLM_ARCH_DFLASH,
+ LLM_ARCH_LAGUNA,
LLM_ARCH_UNKNOWN,
@@ -382,2 +383,3 @@ enum llm_tensor {
@@ -34,10 +34,10 @@ diff --git a/src/llama-model.cpp b/src/llama-model.cpp
+ return new llama_model_laguna(params);
case LLM_ARCH_DECI:
return new llama_model_deci(params);
-@@ -2409,2 +2411,3 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
- case LLM_ARCH_TALKIE:
-+ case LLM_ARCH_LAGUNA:
+@@ -2525,2 +2527,3 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_MELLUM:
++ case LLM_ARCH_LAGUNA:
+ case LLM_ARCH_DFLASH:
diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
--- a/src/llama-model-loader.cpp
+++ b/src/llama-model-loader.cpp
@@ -76,7 +76,7 @@ diff --git a/src/llama-vocab.h b/src/llama-vocab.h
diff --git a/src/models/models.h b/src/models/models.h
--- a/src/models/models.h
+++ b/src/models/models.h
-@@ -1481,1 +1481,14 @@ struct llama_model_dots1 : public llama_model_base {
+@@ -1657,1 +1657,14 @@ struct llama_model_arcee : public llama_model_base {
+struct llama_model_laguna : public llama_model_base {
+ llama_model_laguna(const struct llama_model_params & params) : llama_model_base(params) {}
+ void load_arch_hparams(llama_model_loader & ml) override;