docker run --rm --network host -v /data/hzy/sskj/experiments/pro6000/kimi3_pro6000_pd_dflash_validation/../../../datasets/gsm8k/test.jsonl:/dataset/gsm8k.jsonl:ro -v /data/hzy/sskj/experiments/pro6000/kimi3_pro6000_pd_dflash_validation/bench_gsm8k_acceptance.py:/bench.py:ro -v /data/hzy/sskj/experiments/pro6000/kimi3_pro6000_pd_dflash_validation/results/pd-dflash-kvbounds1-20260831-1740:/results --entrypoint python3 local/sglang:kimi-k3-pp-dflash-33863-fi0618-situ4460-kvbounds1 /bench.py --data-path /dataset/gsm8k.jsonl --num-questions 64 --num-shots 5 --max-new-tokens 512 --temperature 0 --top-p 1 --parallel 8 --speculative-algorithm DFLASH --host 174.1.60.1 --port 31000 --backend srt --output-file /results/bench/gsm8k_c8.jsonl --summary-file /results/bench/gsm8k_c8_summary.json 
