root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-13 04:00:16 +00:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

llama : fix hard-coded number of experts

Browse Source
This commit is contained in:
Georgi Gerganov 2023-12-11 08:55:16 +02:00
parent b0029815e4
commit 8cbaed1d9a
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llama.cpp
View File

@ -3079,7 +3079,7 @@ static void llm_load_tensors(
GGML_ASSERT(hparams.n_expert_used > 0);
// MoE branch
for (int x = 0; x < 8; ++x) {
for (uint32_t x = 0; x < hparams.n_expert; ++x) {
layer.ffn_gate_exp[x] = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_GATE_EXP, "weight", i, x), {n_embd, n_ff}, backend_split);
layer.ffn_down_exp[x] = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_DOWN_EXP, "weight", i, x), { n_ff, n_embd}, backend_split);
layer.ffn_up_exp[x] = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_UP_EXP, "weight", i, x), {n_embd, n_ff}, backend_split);