From 50a64f075e5ad69fc477ac7d567d1f4a124af191 Mon Sep 17 00:00:00 2001
From: Adrian Ciprian Iancu <adrian.iancu@keytech.srl>
Date: Sat, 24 Oct 2026 21:02:28 +0200
Subject: [PATCH] fix: skip metadata alpha for LoRAs with baked_scale (#2202)

---
 src/model/adapter/lora.hpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/model/adapter/lora.hpp b/src/model/adapter/lora.hpp
index 1e5377cc..b2657a36 100644
--- a/src/model/adapter/lora.hpp
+++ b/src/model/adapter/lora.hpp
@@ -69,7 +69,8 @@ struct LoraModel : public GGMLRunner {
         metadata_alpha      = -1.f;
         const auto metadata = model_manager->loader().file_metadata(source_file);
         const auto alpha    = metadata.find("alpha");
-        if (alpha != metadata.end()) {
+        // Converters that write "baked_scale" have already folded alpha / rank into lora_up.
+        if (alpha != metadata.end() && metadata.find("baked_scale") == metadata.end()) {
             try {
                 size_t parsed  = 0;
                 metadata_alpha = std::stof(alpha->second, &parsed);