From 3907d55b00f6ca5697a8e000d6ff640d7509b4d2 Mon Sep 17 00:00:00 2001 From: yuyazhua Date: Sat, 22 Nov 2025 11:31:38 +0800 Subject: [PATCH] fix eval_llama_qnn custom annotation --- examples/qualcomm/oss_scripts/llama/eval_llama_qnn.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/examples/qualcomm/oss_scripts/llama/eval_llama_qnn.py b/examples/qualcomm/oss_scripts/llama/eval_llama_qnn.py index 9af9cdf9549..a21c45c2017 100644 --- a/examples/qualcomm/oss_scripts/llama/eval_llama_qnn.py +++ b/examples/qualcomm/oss_scripts/llama/eval_llama_qnn.py @@ -316,7 +316,9 @@ def eval_llm(args): if args.ptq is not None: quant_dtype = getattr(QuantDtype, f"use_{args.ptq}") decoder_model_config = SUPPORTED_LLM_MODELS[args.decoder_model] - custom_annotations = decoder_model_config.custom_annotation + custom_annotations = ( + decoder_model_config.quant_recipe().recipe.custom_quant_annotations + ) quantizer = make_custom_quantizer( quant_dtype, args.range_setting, custom_annotations, args.quant_linear_only