From 63e1825fb2689f2fe043a6e48e48adc1a2a5df69 Mon Sep 17 00:00:00 2001 From: Charles2530 <2569337619@qq.com> Date: Mon, 16 Mar 2026 11:56:09 +0800 Subject: [PATCH] debug vllm per-block --- llmc/utils/export_vllm.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/llmc/utils/export_vllm.py b/llmc/utils/export_vllm.py index 1128c3df9..87271daf4 100755 --- a/llmc/utils/export_vllm.py +++ b/llmc/utils/export_vllm.py @@ -31,7 +31,8 @@ def update_vllm_quant_config( with open(config_file, 'w') as file: json.dump(config_vllm, file, indent=4) return - elif config.quant.weight.get('granularity', 'per_block'): + # elif config.quant.weight.get('granularity', 'per_block'): + elif config.quant.weight.get('granularity') == 'per_block': quant_config = { 'activation_scheme': 'dynamic', 'fmt': 'e4m3',