---
title: README
emoji: ⚡
colorFrom: green
colorTo: gray
sdk: static
pinned: false
---

# Gittensor Model Hub

Fast, practical checkpoints for single-GPU Blackwell — **RTX 5090** and **RTX PRO 6000** — quantized, benchmarked, and served by our own engine, [**SparkInfer**](https://sparkinfer.com).

**[Qwen3.8-27B-NVFP4-RTX5090](https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090)** — the flagship. Full **262K context** on 32 GB, 81.6 tok/s alone — **155.8 tok/s** with the matched **[DSpark drafter](https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-DSpark-NVFP4)** (1.41 GB, outputs byte-identical).

More: [No-MTP variant](https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090-No-MTP) · [DSpark BF16 source](https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090-DSpark) · [Spark-Hermes-3.8-27B](https://huggingface.co/gittensor-model-hub/Spark-Hermes-3.8-27B) (weights soon)