Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

tests : add support for qwen3 SSM archs (#24031)

Browse Source 
* tests : add support for qwen3 SSM archs

* arch : add LLM_KV_ATTENTION_RECURRENT_LAYERS

* cont : naming + TODOs
This commit is contained in:
Georgi Gerganov
2026-06-03 10:15:27 +03:00
committed by GitHub
parent d545a2a993
commit 06938ac129
25 changed files with 109 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/plamo2.cpp
+3 -3
View File
@@ -12,7 +12,7 @@ void llama_model_plamo2::load_arch_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_SSM_GROUP_COUNT, hparams.ssm_n_group);
for (uint32_t i = 0; i < hparams.n_layer; ++i) {
hparams.recurrent_layer_arr[i] = hparams.n_head_kv(i) == 0;
hparams.is_recr_impl[i] = hparams.n_head_kv(i) == 0;
}
switch (hparams.n_layer) {
@@ -54,7 +54,7 @@ void llama_model_plamo2::load_arch_tensors(llama_model_loader &) {
for (int i = 0; i < n_layer; ++i) {
auto & layer = layers[i];
bool is_mamba_layer = hparams.is_recurrent(i);
bool is_mamba_layer = hparams.is_recr(i);
layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
@@ -128,7 +128,7 @@ llama_model_plamo2::graph::graph(const llama_model & model, const llm_graph_para
cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
// check if this layer is Mamba or Attention
const bool is_mamba_layer = hparams.is_recurrent(il);
const bool is_mamba_layer = hparams.is_recr(il);
if (is_mamba_layer) {
// PLaMo-2 Mamba layer