ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-21 13:44:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

qwen3next: warn when forcing fused decode mode

Browse Source
This commit is contained in:
yurko
2026-02-08 00:08:33 -08:00
parent 64099e71c0
commit 343e335ff0
1 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llama-build-context.cpp
View File

@@ -4208,6 +4208,13 @@ ggml_cgraph * llm_build_context::build_qwen3next() {
return qwen3next_fused_delta_mode::off;
}
}();
if (fused_delta_mode == qwen3next_fused_delta_mode::all_tokens) {
static bool warned_all_tokens = false;
if (!warned_all_tokens) {
LLAMA_LOG_WARN("%s: LLAMA_QWEN3NEXT_FUSED_DELTA=2 enables fused single-token decode; quality regression is known in this mode\n", __func__);
warned_all_tokens = true;
}
}
auto get_slice_2d = [&](ggml_tensor * t, int64_t c) -> ggml_tensor * {
return ggml_view_4d(ctx0, t, t->ne[0], t->ne[1], 1, t->ne[3],