---
title: qwen38-27b-gguf-t4ares-thor01
canonical_url: "https://www.modelscope.cn/models/navyyang/qwen38-27b-gguf-t4ares-thor01"
md_url: "https://www.modelscope.cn/models/navyyang/qwen38-27b-gguf-t4ares-thor01.md"
repository: navyyang/qwen38-27b-gguf-t4ares-thor01
chinese_name: "Qwen3.8-27B GGUF（llama.cpp T4/ARES DRIVE Thor 生产权重）"
last_updated: 2026-10-01
license: apache-2.0
library_name:
  - gguf
downloads: 55
stars: 0
tags:
  - gguf
---

# qwen38-27b-gguf-t4ares-thor01

> qwen38-27b-gguf-t4ares-thor01 - navyyang 在 ModelScope 开源的模型。Qwen3.8-27B GGUF — llama.cpp T4/ARES production weights (DRIVE Thor)

navyyang/qwen38-27b-gguf-t4ares-thor01 是 ModelScope 魔搭社区上的机器学习模型，采用 apache-2.0 许可。

- **Repository**: navyyang/qwen38-27b-gguf-t4ares-thor01
- **License**: apache-2.0
- **Tags**: gguf
- **Downloads**: 55
- **Stars**: 0
- **Last updated**: 2026-10-01

Source: https://www.modelscope.cn/models/navyyang/qwen38-27b-gguf-t4ares-thor01

---

# Qwen3.8-27B GGUF — llama.cpp T4/ARES production weights (DRIVE Thor)

Production GGUF files serving on our DRIVE Thor (Tegra264, sm_101a) board via
llama.cpp (commit 72797e89, T4/ARES optimized build):

| File | Size (bytes) | Role |
|---|---:|---|
| `RadixArk-F8attn-v2.gguf` | 21,016,353,984 | main model — our own conversion (NVFP4 FFN + F8_E4M3 attention projections, qwen35 arch) |
| `Qwen3.8-27B-DFlash2-BF16.gguf` | 3,860,293,216 | DFlash2 speculative-decoding draft |
| `mmproj-Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-BF16.gguf` | 931,146,624 | vision projector (HauhauCS public file, mirrored) |

All Apache 2.0 (base: Qwen/Qwen3.8-27B; draft: z-lab/Qwen3.8-27B-DFlash2; mmproj: HauhauCS).
Provenance details in `NOTICE.md`. Verify after download: `sha256sum -c SHA256SUMS.txt`.

Deploy docs: [thor01-t4ares-deploy branch](https://github.com/isenlink/thor-fp8-llm/tree/thor01-t4ares-deploy) → `docs/09-t4ares-deploy/README.md`.
Binary (llama-server) + source: [ModelScope thor01-qwen38-27b-dflash2-t4ares-deploy](https://www.modelscope.cn/models/navyyang/thor01-qwen38-27b-dflash2-t4ares-deploy).
Companion vLLM-line weights (separate stack): [qwen38-27b-nvfp4-vllm-p3-thor04](https://www.modelscope.cn/models/navyyang/qwen38-27b-nvfp4-vllm-p3-thor04).