thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

model : add GroveMoE support (#15510)

Browse source 
* add GroveMoE support

* remove constexpr that fails on certain compilers

* revert crude scalar div implementation, use cast

* build_attn_inp_kv_unified -> build_attn_inp_kv

* fix build_attn

* re-apply ffn_exps regex changes
This commit is contained in:
Sigbjørn Skjæret 2025-09-25 19:50:28 +02:00 • committed by GitHub
parent b05a9d650f
commit 835b2b915c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 451 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

2
common/common.h
View file

@ -738,7 +738,7 @@ const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
// MoE utils
//
const char * const LLM_FFN_EXPS_REGEX = "\\.ffn_(up|down|gate)_exps";
const char * const LLM_FFN_EXPS_REGEX = "\\.ffn_(up|down|gate)_(ch|)exps";
static std::string llm_ffn_exps_block_regex(int idx) {
return string_format("blk\\.%d%s", idx, LLM_FFN_EXPS_REGEX);