From 0d7c1435fad3d5ebd29887390a134bef7f7a2761 Mon Sep 17 00:00:00 2001 From: Xuan Son Nguyen Date: Tue, 2 Dec 2025 11:17:40 +0100 Subject: [PATCH] convert: add error message for mistral3 quantized weight --- convert_hf_to_gguf.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py index a54cce887bb..8ddb6d04cd9 100755 --- a/convert_hf_to_gguf.py +++ b/convert_hf_to_gguf.py @@ -2842,6 +2842,10 @@ def set_gguf_parameters(self): self.gguf_writer.add_attn_temperature_scale(rope_params["llama_4_scaling_beta"]) def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None): + # TODO: probably not worth supporting quantized weight, as official BF16 is also available + if name.endswith("weight_scale_inv"): + raise ValueError("This is a quantized weight, please use BF16 weight instead") + name = name.replace("language_model.", "") if "multi_modal_projector" in name or "vision_tower" in name: return []