From 16e6bc20b0915b004c7573a4634efc00f7fdb936 Mon Sep 17 00:00:00 2001 From: mmdbhs <48610427+mmdbhs@users.noreply.github.com> Date: Mon, 15 Dec 2025 15:50:39 +0800 Subject: [PATCH] fix CompressedTensorsW8A8Int8 min_capability (#13914) Co-authored-by: Fan Yin <1106310035@qq.com> Co-authored-by: Peng Zhang --- .../schemes/compressed_tensors_w8a8_int8.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py index 9bca2834d..2448210be 100644 --- a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +++ b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py @@ -35,8 +35,8 @@ class CompressedTensorsW8A8Int8(CompressedTensorsScheme): @classmethod def get_min_capability(cls) -> int: - # lovelace and up - return 89 + # ampere and up + return 80 def process_weights_after_loading(self, layer) -> None: # If per tensor, when we have a fused module (e.g. QKV) with per