Issues

• Heads-up: llama-server's n-gram table persists across requests, inflating repeat ngram results

  alexziskind1/draftbench#7 · Filter by author zerogeeks opened on Nov 3, 2026

• benchmark for vLLM

  alexziskind1/draftbench#6 · Filter by author vanheemst opened on Feb 26, 2026

• Getting 400 Client Error: Bad Request when running sweep.py

  alexziskind1/draftbench#5 · Filter by author Filip-Anicic opened on Feb 23, 2026
  2

• TPS intermittently reported as 0.0 when benchmarking llama-server (Tesla P40, Qwen3 14B + Draft models)

  alexziskind1/draftbench#3 · Filter by author Mabe75 opened on Feb 19, 2026
  3

• Single Benchmark Directions Don't Work

  alexziskind1/draftbench#2 · Filter by author ForbiddenEra opened on Feb 18, 2026

• Do the GGUF results correlate with MLX?

  alexziskind1/draftbench#1 · Filter by author larsony99 opened on Feb 18, 2026