diff --git a/vllm-qwen36.sh b/vllm-qwen36.sh
index b90e45a..ee73d63 100755
--- a/vllm-qwen36.sh
+++ b/vllm-qwen36.sh
@@ -5,7 +5,6 @@ set -euo pipefail
 # CPU expert offloading to fit MoE model on 24GB RTX 3090
 
 MODEL="Qwen/Qwen3.6-35B-A3B-FP8"
-DRAFT_MODEL="/home/dank/.cache/huggingface/models/Qwen3.6-35B-A3B-DFlash"
 PORT=8001
 HOST="0.0.0.0"
 
@@ -29,5 +28,4 @@ exec docker run --rm --name vllm-qwen36 \
     --cpu-offload-gb 30 \
     --cpu-offload-params "experts" \
     --enable-chunked-prefill \
-    --enable-prefix-caching \
-    --enforce-eager
+    --enable-prefix-caching