mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

model : add EXAONE MoE (#18543)

Browse Source 
* Add EXAONE MoE implementations

Co-authored-by: Junwon Hwang <nuclear1221@gmail.com>

* Address PR feedback

* Address PR feedback

* [WIP] Add MTP for EXAONE-MoE

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

---------

Co-authored-by: LG-AI-EXAONE <exaonemodels@lgresearch.ai>
This commit is contained in:
Junwon Hwang
2026-01-14 07:28:38 +09:00
committed by GitHub
parent e4832e3ae4
commit 60591f01d4
17 changed files with 656 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-arch.cpp
+33
View File
@@ -81,6 +81,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_NEMOTRON_H_MOE, "nemotron_h_moe" },
{ LLM_ARCH_EXAONE, "exaone" },
{ LLM_ARCH_EXAONE4, "exaone4" },
{ LLM_ARCH_EXAONE_MOE, "exaone-moe" },
{ LLM_ARCH_RWKV6, "rwkv6" },
{ LLM_ARCH_RWKV6QWEN2, "rwkv6qwen2" },
{ LLM_ARCH_RWKV7, "rwkv7" },
@@ -1728,6 +1729,38 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
LLM_TENSOR_FFN_UP,
LLM_TENSOR_FFN_POST_NORM,
};
case LLM_ARCH_EXAONE_MOE:
return {
LLM_TENSOR_TOKEN_EMBD,
LLM_TENSOR_OUTPUT_NORM,
LLM_TENSOR_OUTPUT,
LLM_TENSOR_ROPE_FREQS,
LLM_TENSOR_ATTN_NORM,
LLM_TENSOR_ATTN_Q,
LLM_TENSOR_ATTN_Q_NORM,
LLM_TENSOR_ATTN_K,
LLM_TENSOR_ATTN_K_NORM,
LLM_TENSOR_ATTN_V,
LLM_TENSOR_ATTN_OUT,
LLM_TENSOR_FFN_NORM,
LLM_TENSOR_FFN_GATE,
LLM_TENSOR_FFN_DOWN,
LLM_TENSOR_FFN_UP,
LLM_TENSOR_FFN_GATE_INP,
LLM_TENSOR_FFN_GATE_EXPS,
LLM_TENSOR_FFN_DOWN_EXPS,
LLM_TENSOR_FFN_UP_EXPS,
LLM_TENSOR_FFN_GATE_SHEXP,
LLM_TENSOR_FFN_UP_SHEXP,
LLM_TENSOR_FFN_DOWN_SHEXP,
LLM_TENSOR_FFN_EXP_PROBS_B,
LLM_TENSOR_NEXTN_EH_PROJ,
LLM_TENSOR_NEXTN_EMBED_TOKENS,
LLM_TENSOR_NEXTN_ENORM,
LLM_TENSOR_NEXTN_HNORM,
LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
};
case LLM_ARCH_RWKV6:
return {
LLM_TENSOR_TOKEN_EMBD,