jdelony/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-06-28 04:30:15 -05:00
Code Issues Packages Projects Releases Wiki Activity

Fix Minimax M3 crash when -muge merges up/gate experts

Browse Source 
The graph builder for Minimax M3 (build_minimaxm3.cpp) was not passing
model.layers[il].ffn_up_gate_exps to llm_build_std_moe_ffn, unlike
Minimax M2 and all other MoE model graph builders.

When -muge (merge_up_gate_experts) is enabled, the merge creates a single
ffn_up_gate_exps tensor with ffn_up_exps and ffn_gate_exps as views.
Only the parent merged tensor gets the split 'extra' pointer set.
Without passing it as up_gate_exps parameter, the function sees null
split pointers for up/gate (the views) while split_down_exps is valid,
causing the assertion at llama-build-context.cpp:1453 to fail.
This commit is contained in:
Nexesenex 2026-06-15 14:36:14 +02:00
parent f81673c7db
commit 3c9680fd3c
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/graphs/build_minimaxm3.cpp
View File

@ -52,7 +52,7 @@ ggml_cgraph* llm_build_context::build_minimaxm3() {
hparams.expert_weights_scale != 0.0f, hparams.expert_weights_scale, hparams.expert_weights_scale != 0.0f, hparams.expert_weights_scale,
(llm_expert_gating_func_type) hparams.expert_gating_func, (llm_expert_gating_func_type) hparams.expert_gating_func,
LLM_FFN_SWIGLU_OAI, LLM_FFN_SWIGLU_OAI,
cb, il, gf, true); cb, il, gf, true, model.layers[il].ffn_up_gate_exps);
} }
cur = lctx.cvec.apply_to(ctx0, cur, il); cur = lctx.cvec.apply_to(ctx0, cur, il);