benchmark-results / README.md
EXDai's picture
Upload README.md with huggingface_hub
c906f03 verified
|
Raw
History Blame Contribute Delete
828 Bytes
metadata
title: EXD Benchmark Results
emoji: 📊
colorFrom: gray
colorTo: red
sdk: static
pinned: false
license: mit

EXD Benchmark Results

Performance data from inference sweeps on the Gigabyte AI TOP (DGX Spark-class, NVIDIA GB10).

Contents

File Source Description
episodes/Ep03/03_bench.txt Episode 3 First benchmark run — Qwen2.5-7B, TTFT, TPOT, throughput
benchmarks/qwen3.6-35b-a3b-llama-benchy.md Serve harness Qwen3.6-35B-A3B with llama-benchy

Hardware

  • Machine: Gigabyte AI TOP (DGX Spark variant)
  • SoC: NVIDIA Grace-Blackwell (GB10)
  • Memory: 128 GB unified (CPU + GPU shared)
  • Runtime: vLLM inside Docker (NGC container)

Related