convert : force patch_merger tensors to f16/f32 (#18124)

2025-12-17 22:15:53 +01:00
parent 6ce3d85796
commit 5166aaf868
1 changed files with 1 additions and 1 deletions
@@ -1838,7 +1838,7 @@ class MmprojModel(ModelBase):

    def tensor_force_quant(self, name, new_name, bid, n_dims):
        del bid, name, n_dims  # unused
-        if ".patch_embd.weight" in new_name:
+        if ".patch_embd.weight" in new_name or ".patch_merger.weight" in new_name:
            return gguf.GGMLQuantizationType.F16 if self.ftype == gguf.LlamaFileType.MOSTLY_F16 else gguf.GGMLQuantizationType.F32
        return False