Baseline: ThinkingCap 27B FP8 on dankclaw with enforce-eager and validation timeout

scripts/serve-thinkingcap.sh

@@ -21,6 +21,7 @@ exec "${VENV}/bin/vllm" serve "${MODEL}" \
     --host 0.0.0.0 \
     --kv-cache-dtype fp8 \
     --gpu-memory-utilization 0.80 \
+    --enforce-eager \
     --max-model-len 32768 \
     --max-num-batched-tokens 2048 \
     --max-num-seqs 32 \