ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-28 17:14:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Graph parallel for dense Qwen-3.5 models (#1331)

Browse Source 
* Graph parallel for idense Qwen-3.5 models

* Cleanup
This commit is contained in:
Kawrakow
2026-02-27 07:03:25 +01:00
committed by GitHub
parent facc8fdc44
commit 1e6d36b1b4
3 changed files with 28 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llama.cpp
View File

@@ -661,7 +661,7 @@ llama_context::~llama_context() {
// kv cache helpers
//
static inline bool llama_qwen3next_is_recurrent_layer(const llama_hparams & hparams, uint32_t il) {
static inline bool llama_is_recurrent_layer(const llama_hparams & hparams, uint32_t il) {
return hparams.is_recurrent(il);
}
@@ -836,7 +836,7 @@ static bool llama_kv_cache_init(
int n_mla = 0;
for (int i = 0; i < (int) n_layer; i++) {
const bool qnext_recurrent = llama_qwen3next_is_recurrent_layer(hparams, i);
const bool qnext_recurrent = llama_is_recurrent_layer(hparams, i);
const uint32_t n_embd_v_row = llama_kv_v_row_embd(model, hparams, i);
const uint32_t n_head_kv = hparams.n_head_kv(i);
const uint32_t n_embd_head_k= hparams.n_embd_head_k;
@@ -1937,6 +1937,7 @@ static bool is_model_split_supported(const llama_model & model) {
LLM_ARCH_SEED_OSS,
LLM_ARCH_STEP35,
LLM_ARCH_QWEN3NEXT,
LLM_ARCH_QWEN35,
};
auto it = k_supported.find(model.arch);
return it != k_supported.end();