nicolas
/
ik_llama_opt
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix MiniMax-M2 KV-cache loading/saving

This commit is contained in:
Kawrakow 2026-01-30 13:38:07 +00:00
parent 811f8c3393
commit 098b1a2e04
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llama.cpp
View File

@ -6131,7 +6131,9 @@ struct llama_data_write_buffer : llama_data_write {
throw std::runtime_error(std::string{"Split cache for type "} + ggml_type_name(tensor->type) + " is not supported");
}
GGML_ASSERT(il >= 0 && il < int(model.layers.size()));
auto kv = tensor->ne[1] > 1 ? model.layers[il].wk : model.layers[il].wv;
const auto & l = model.layers[il];
bool use_V_for_K = l.attn_k_norm && l.attn_k_norm->ne[0] == l.wk->ne[1] ? true : false;
auto kv = tensor->ne[1] > 1 && !use_V_for_K ? l.wk : l.wv;
get_tensor_data_split(ptr, tensor, kv, aux_buffer, offset, size);
}