diff --git a/include/llama.h b/include/llama.h
index 34b42dc1f45..e424d5c4010 100644
--- a/include/llama.h
+++ b/include/llama.h
@@ -157,6 +157,7 @@ extern "C" {
         LLAMA_FTYPE_MOSTLY_Q1_0          = 40, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_Q2_0          = 41, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_PQ2_0     = 141, // except 1d tensors (Prism group-128 Q2_0; matches published PQ2_0 ggufs)
+        LLAMA_FTYPE_MOSTLY_PQ2_0_LEGACY = 142, // pre-rename value for the same format, still found in published ggufs
 
         LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
     };
diff --git a/src/llama-model-loader.cpp b/src/llama-model-loader.cpp
index 7e7eb5cca66..c43499aa234 100644
--- a/src/llama-model-loader.cpp
+++ b/src/llama-model-loader.cpp
@@ -40,6 +40,9 @@ const char * llama_ftype_name(llama_ftype ftype) {
         case LLAMA_FTYPE_MOSTLY_Q1_0:      name = LLAMA_FTYPE_PREFIX "Q1_0"; break;
         case LLAMA_FTYPE_MOSTLY_Q2_0:      name = LLAMA_FTYPE_PREFIX "Q2_0"; break;
         case LLAMA_FTYPE_MOSTLY_PQ2_0: name = LLAMA_FTYPE_PREFIX "PQ2_0 - 2.13 bpw (group 128)"; break;
+        // ggufs packed before the Q2_0_G128 -> PQ2_0 rename carry the old ftype value.
+        // They load and compute correctly; name it so it does not report as unknown.
+        case LLAMA_FTYPE_MOSTLY_PQ2_0_LEGACY: name = LLAMA_FTYPE_PREFIX "PQ2_0 - 2.13 bpw (group 128, legacy ftype)"; break;
         case LLAMA_FTYPE_MOSTLY_Q4_0:      name = LLAMA_FTYPE_PREFIX "Q4_0"; break;
         case LLAMA_FTYPE_MOSTLY_Q4_1:      name = LLAMA_FTYPE_PREFIX "Q4_1"; break;
         case LLAMA_FTYPE_MOSTLY_Q5_0:      name = LLAMA_FTYPE_PREFIX "Q5_0"; break;