From 03fa90ff3554fc349ac93c6502283e32d76c3ad2 Mon Sep 17 00:00:00 2001 From: Disty0 Date: Sun, 7 Jun 2026 18:02:28 +0300 Subject: [PATCH] Add warning for and skip post-load quant of pre-quants --- modules/model_quant.py | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/modules/model_quant.py b/modules/model_quant.py index baacde86e..7ab67e48e 100644 --- a/modules/model_quant.py +++ b/modules/model_quant.py @@ -356,6 +356,14 @@ def sdnq_quantize_model(model, op=None, sd_model=None, do_gc: bool = True, weigh from modules.sdnq import sdnq_post_load_quant from modules.sdnq.common import use_torch_compile as sdnq_use_torch_compile + if ( + hasattr(model, "quantization_config") + or (hasattr(model, "config") and hasattr(model.config, "quantization_config")) + or (hasattr(model, "config") and isinstance(model.config, dict) and "quantization_config" in model.config) + ): + log.warning(f'Quantization: Trying to quantize a pre-quantized model. Skipping quantization of module="{op if op is not None else model.__class__}"') + return model + if shared.opts.sdnq_use_quantized_matmul and not sdnq_use_torch_compile: log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.') shared.opts.sdnq_use_quantized_matmul = False