CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama-graph : use ggml_repeat_4d (#13998)

Browse Source
This commit is contained in:
Xuan-Son Nguyen
2025-06-04 10:11:26 +02:00
committed by GitHub
parent 0b4be4c435
commit 3ac67535c8
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llama-graph.cpp
View File

@@ -769,9 +769,8 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
cur = ggml_reshape_3d(ctx0, cur, n_embd, 1, n_tokens); cur = ggml_reshape_3d(ctx0, cur, n_embd, 1, n_tokens);
if (weight_before_ffn) { if (weight_before_ffn) {
// TODO: this is a workaround as we don't yet have a repeat op that takes custom dim (ggml_repeat_4d) // repeat cur to [n_embd, n_expert_used, n_tokens]
ggml_tensor * repeated = ggml_new_tensor_3d(ctx0, cur->type, n_embd, n_expert_used, n_tokens); ggml_tensor * repeated = ggml_repeat_4d(ctx0, cur, n_embd, n_expert_used, n_tokens, 1);
repeated = ggml_repeat(ctx0, cur, repeated); // [n_embd, n_expert_used, n_tokens]
cur = ggml_mul(ctx0, repeated, weights); cur = ggml_mul(ctx0, repeated, weights);
cb(cur, "ffn_moe_weighted", il); cb(cur, "ffn_moe_weighted", il);
} }