From 52da4a4f8c28d063378d54dd806da03614251e76 Mon Sep 17 00:00:00 2001
From: Ed Addario <eaddario@hotmail.com>
Date: Wed, 20 Aug 2025 17:26:05 +0100
Subject: [PATCH] Skip if output.weight or type is COPY

---
 src/llama-quant.cpp | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/src/llama-quant.cpp b/src/llama-quant.cpp
index c412191c8f..786adfe547 100644
--- a/src/llama-quant.cpp
+++ b/src/llama-quant.cpp
@@ -697,6 +697,9 @@ static std::unordered_map<std::string, ggml_type> target_bpw_type(
         q &= name.find("time_mix_decay_w2.weight") == std::string::npos;
         q &= name.find("time_mix_lerp_fused.weight") == std::string::npos;
         q &= name.find("attn_rel_b.weight") == std::string::npos;
+        q &= params->quantize_output_tensor || name != "output.weight";
+        q &= !params->only_copy;
+
         return q;
     };