tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 20:05:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

hparams : support models for which all layers use SWA (#13682)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-05-21 20:00:49 +03:00
committed by GitHub
parent 5fbfe384d4
commit 8e186ef0e7
2 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-hparams.cpp
View File

@ -72,7 +72,7 @@ uint32_t llama_hparams::n_embd_v_s() const {
bool llama_hparams::is_swa(uint32_t il) const { bool llama_hparams::is_swa(uint32_t il) const {
if (il < n_layer) { if (il < n_layer) {
return n_swa > 0 && n_swa_pattern > 0 && il % n_swa_pattern < (n_swa_pattern - 1); return n_swa_pattern == 0 || (il % n_swa_pattern < (n_swa_pattern - 1));
} }
GGML_ABORT("fatal error"); GGML_ABORT("fatal error");

13
src/llama-hparams.h
View File

@ -104,7 +104,18 @@ struct llama_hparams {
llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE; llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE;
uint32_t n_swa = 0; // the size of the sliding window (0 - no SWA) uint32_t n_swa = 0; // the size of the sliding window (0 - no SWA)
uint32_t n_swa_pattern = 1; // by default, all layers use non-sliding-window attention uint32_t n_swa_pattern = 1; // this value n means that every nth layer is dense (i.e. non-SWA)
// by default n == 1, all layers are dense
// note that if n_swa_pattern == 0, all layers are SWA
// example: n_swa_pattern = 3
// il == 0: swa
// il == 1: swa
// il == 2: dense
// il == 3: swa
// il == 4: swa
// il == 5: dense
// il == 6: swa
// etc ...
// for State Space Models // for State Space Models
uint32_t ssm_d_conv = 0; uint32_t ssm_d_conv = 0;