From 3fc60e57486e1e3b972ad048d61295936e6f871a Mon Sep 17 00:00:00 2001
From: Le Zhang <damahua@users.noreply.github.com>
Date: Thu, 14 May 2026 09:56:42 -0700
Subject: [PATCH] [MLX] bench_one_batch: thread --quantization through to
 MlxModelRunner (#25221)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 python/sglang/bench_one_batch.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/python/sglang/bench_one_batch.py b/python/sglang/bench_one_batch.py
index 6924b3036..61fa3ba5a 100644
--- a/python/sglang/bench_one_batch.py
+++ b/python/sglang/bench_one_batch.py
@@ -531,6 +531,7 @@ class _MlxBenchRunner:
             trust_remote_code=server_args.trust_remote_code,
             disable_radix_cache=True,
             mem_fraction_static=server_args.mem_fraction_static,
+            quantization=server_args.quantization,
         )
         if server_args.max_total_tokens is not None:
             init_kwargs["pool_size"] = server_args.max_total_tokens