tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 20:25:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

model : disable SWA for Phi models (#13676)

Browse Source 
* model : disable SWA for Phi models

ggml-ci

* model : update warning message

* model : print warning only if n_swa > 0

* model : fix typo
This commit is contained in:
Georgi Gerganov
2025-05-21 13:09:21 +03:00
committed by GitHub
parent 33983057d0
commit b44890df2e
2 changed files with 30 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llama-graph.cpp
View File

@ -1236,8 +1236,7 @@ llm_graph_input_attn_kv_unified * llm_graph_context::build_attn_inp_kv_unified()
auto inp = std::make_unique<llm_graph_input_attn_kv_unified>(hparams, cparams, kv_self);
{
GGML_ASSERT(hparams.n_swa_pattern == 1 && "Use llama_kv_cache_unified_iswa for SWA");
GGML_ASSERT(hparams.n_swa == 0 && "Use llama_kv_cache_unified_iswa for SWA");
GGML_ASSERT(hparams.swa_type == LLAMA_SWA_TYPE_NONE && "Use llama_kv_cache_unified_iswa for SWA");
const auto n_kv = kv_self->get_n();
@ -1312,8 +1311,8 @@ llm_graph_input_attn_kv_unified_iswa * llm_graph_context::build_attn_inp_kv_unif
inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
}
if (hparams.n_swa_pattern > 1) {
GGML_ASSERT(hparams.n_swa > 0 && "Use llama_kv_cache_unified for non-SWA");
{
GGML_ASSERT(hparams.swa_type != LLAMA_SWA_TYPE_NONE && "Use llama_kv_cache_unified for non-SWA");
const auto n_kv = kv_self->get_kv_swa()->get_n();