tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-26 10:09:41 -04:00
Code Issues Packages Projects Releases Wiki Activity

graph : use F32 accumulators for gpt-oss

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-08-14 16:08:31 +03:00
parent d32e03f449
commit 220860aa0c
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llama-graph.cpp
View File

@@ -1566,6 +1566,11 @@ ggml_tensor * llm_graph_context::build_attn_with_sinks(
if (wo) { if (wo) {
cur = build_lora_mm(wo, cur); cur = build_lora_mm(wo, cur);
if (arch == LLM_ARCH_OPENAI_MOE) {
// similar the original build_attn
// TODO: this is tmp until we refactor and remove the build_attn_with_sinks() path
ggml_mul_mat_set_prec(cur, GGML_PREC_F32);
}
} }
if (wo_b) { if (wo_b) {