mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 32 Packages Projects Releases Wiki Activity

lora: count lora nodes in graph_max_nodes (#18469)

Browse Source 
* lora: count lora nodes in graph_max_nodes

* 3 nodes per weight

* 4 nodes

* keep track n_lora_nodes from llama_model

* fix assert

* rm redundant header

* common: load adapters before context creation

* use 6 nodes
This commit is contained in:
Xuan-Son Nguyen
2025-12-30 15:53:12 +01:00
committed by GitHub
parent c32fa21db8
commit cd78e57c3a
6 changed files with 46 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-adapter.cpp
+12 -3
View File
@@ -146,9 +146,11 @@ llama_adapter_lora_weight * llama_adapter_lora::get_weight(ggml_tensor * w) {
return nullptr;
}
static void llama_adapter_lora_init_impl(llama_model & model, const char * path_lora, llama_adapter_lora & adapter) {
static void llama_adapter_lora_init_impl(const char * path_lora, llama_adapter_lora & adapter) {
LLAMA_LOG_INFO("%s: loading lora adapter from '%s' ...\n", __func__, path_lora);
llama_model & model = adapter.model;
ggml_context * ctx_init;
gguf_init_params meta_gguf_params = {
/* .no_alloc = */ true,
@@ -411,14 +413,17 @@ static void llama_adapter_lora_init_impl(llama_model & model, const char * path_
}
}
// update number of nodes used
model.n_lora_nodes += adapter.get_n_nodes();
LLAMA_LOG_INFO("%s: loaded %zu tensors from lora file\n", __func__, adapter.ab_map.size()*2);
}
llama_adapter_lora * llama_adapter_lora_init(llama_model * model, const char * path_lora) {
llama_adapter_lora * adapter = new llama_adapter_lora();
llama_adapter_lora * adapter = new llama_adapter_lora(*model);
try {
llama_adapter_lora_init_impl(*model, path_lora, *adapter);
llama_adapter_lora_init_impl(path_lora, *adapter);
return adapter;
} catch (const std::exception & err) {
LLAMA_LOG_ERROR("%s: failed to apply lora adapter: %s\n", __func__, err.what());
@@ -469,6 +474,10 @@ int32_t llama_adapter_meta_val_str_by_index(const llama_adapter_lora * adapter,
}
void llama_adapter_lora_free(llama_adapter_lora * adapter) {
// update number of nodes used
GGML_ASSERT(adapter->model.n_lora_nodes >= adapter->get_n_nodes());
adapter->model.n_lora_nodes -= adapter->get_n_nodes();
delete adapter;
}