From 9df0b73c707900b7277a46b5ae52ff7e7e567d13 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Tue, 20 Oct 2026 15:58:09 +0300 Subject: [PATCH] model : drop the redundant precision overrides in the Kolibri-1 router (#30865) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * model : drop the redundant precision overrides in the Kolibri-1 router Follow-up to the review in #30673: the router matmul works with the default precision, so both ggml_prec_set_* calls can go away. Assisted-by: pi:llama.cpp/ggml-org/Qwen3.8-Flash-Next-GGUF (MacBook, M5 Max, 128GB) * cont : fix comment Co-authored-by: Sigbjørn Skjæret --------- Co-authored-by: Sigbjørn Skjæret --- src/models/kolibri1.cpp | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/src/models/kolibri1.cpp b/src/models/kolibri1.cpp index acc1dd159..2301ef272 100644 --- a/src/models/kolibri1.cpp +++ b/src/models/kolibri1.cpp @@ -131,11 +131,8 @@ llama_model_kolibri1::graph::graph(const llama_model & model, const llm_graph_pa cur = build_norm(ffn_inp, layer.ffn_norm, NULL, LLM_NORM_RMS, il); cb(cur, "ffn_norm", il); - // The router runs in f32: the selection adds the bias to the logits, the weights - // are the sigmoid of the unbiased logits, so the same logits feed both. + // The selection adds the bias to the logits, the weights are the sigmoid of the unbiased logits, so the same logits feed both. ggml_tensor * logits = build_lora_mm(layer.ffn_gate_inp, cur); - ggml_prec_set_acc(logits, GGML_PREC_F32); - ggml_prec_set_src(logits, GGML_PREC_F32, 1); cb(logits, "ffn_moe_logits", il); ggml_tensor * selection = ggml_add(ctx0, logits, layer.ffn_exp_probs_b);