---
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, 128 GB unified memory).

## 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 |

## Related

- [EXD org](https://huggingface.co/EXD-AI) | [GitHub](https://github.com/Ramshreyas/EXD)