diff --git a/vllm-awq-qwen36-dflash-ampere.sh b/vllm-awq-qwen36-dflash-ampere.sh
index b77b8b5..0d42b9b 100755
--- a/vllm-awq-qwen36-dflash-ampere.sh
+++ b/vllm-awq-qwen36-dflash-ampere.sh
@@ -24,12 +24,11 @@ exec docker run --rm --name vllm-awq-qwen36-dflash-ampere \
     --served-model-name "frank-default-small" \
     --speculative-config '{"method": "dflash", "model": "z-lab/Qwen3.6-35B-A3B-DFlash", "num_speculative_tokens": 4}' \
     --max-num-seqs 2 \
-    --max-model-len 262144 \
+    --max-model-len 32768 \
     --tensor-parallel-size 1 \
     --enable-chunked-prefill \
     --max-num-batched-tokens 1536 \
     --performance-mode throughput \
-    # --enable-prefix-caching  # Disabled: Mamba cache align mode conflicts with prefix caching \\
     --enable-auto-tool-choice --tool-call-parser qwen3_xml \
     --reasoning-parser qwen3 \
     --trust-remote-code \