Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-19 01:04:55 +02:00
Code Issues Packages Projects Releases Wiki Activity

test-model-random : configurable model n_ctx, and smaller seq lengths

Browse Source
This commit is contained in:
Francis Couture-Harpin
2025-07-08 16:34:44 -04:00
parent 6b38c7a04c
commit 362cf5429c
2 changed files with 13 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+3 -1
View File
@@ -598,9 +598,11 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_INTERLEAVE_MOE_LAYER_STEP, hparams.n_moe_layer_step);
hparams.swa_type = LLAMA_SWA_TYPE_CHUNKED;
hparams.n_swa = 8192; // should this be a gguf kv? currently it's the same for Scout and Maverick
hparams.n_swa = 8192; // currently it's the same for Scout and Maverick
hparams.set_swa_pattern(4); // pattern: 3 chunked - 1 full
ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false); // for overrides in tests
switch (hparams.n_expert) {
case 16: type = LLM_TYPE_17B_16E; break;
case 128: type = LLM_TYPE_17B_128E; break;