Skip to content
Other

Arete-27B-Gosset: running a 27B model on a single 16 GB VRAM GPU

2 min read

Arete-27B-Gosset is a specially quantized version of Nous Research's Arete-27B hybrid language model, compressed down to 3.79 bits per weight (bpw) using Gosset Packing. This deep compression brings the file size on disk down to about 13 GB, letting the model fit entirely inside the VRAM of 16 GB cards (like the RTX 5080) without offloading to system RAM. It retains the base architecture's multimodal capabilities and a default context window of up to 32,768 tokens, delivering fast generation speeds alongside OpenAI and Ollama API compatibility.

Key features

  • Full VRAM residency: Uses only 12.07 GiB of static memory and peaks at 13.48 GiB at the default 32K context.
  • Fast generation speeds: Hits 53.92 tokens per second during decoding on tested hardware.
  • Hybrid architecture: Uses DeltaNet layers with O(1) linear attention, cutting down KV-cache memory usage.
  • Ready-to-use runtime: Ships with an interactive chat CLI and a local server compatible with the OpenAI and Ollama APIs.

What you can do

  • Local chat: Run a full-featured interactive assistant right on a home PC with a single GPU.
  • Local API server: Spin up a local endpoint compatible with OpenAI and Ollama to plug into third-party apps.
  • Problem-solving and coding: Handle programming tasks (backed by strong HumanEval scores) and reasoning locally without relying on cloud APIs.

Example

Launch an interactive chat session with the model in your terminal after installing dependencies:

python run_e8_chat.py
Read moreDetails
Benchmarks: HumanEval (pass@1) — 91.46%, GSM8K (100-problem sample) — 93%.
Requirements: NVIDIA GPU with at least 16 GB VRAM, Python 3.11+, ~14 GB of free disk space.
Limitations: Non-commercial Gosset Evaluation License; perplexity is roughly 5–7% higher than the BF16 original.
Architecture notes: 48 out of 64 layers use linear attention with zero KV-cache overhead.
Discuss Telegram 1

Will it run?

Runs in I32

Fits without quantisation.

  • I32weights 12 GB · ready files14 GB
  • FP8 / INT8recommendedweights 4.5 GB5.4 GB
  • 4-bitweights 2.5 GB3.3 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · I32 — from RTX 5080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # GossetWorks/Arete-27B-Gosset-3.65BPW
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download GossetWorks/Arete-27B-Gosset-3.65BPW --local-dir ~/models/Arete-27B-Gosset-3.65BPW
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Qwen3.8-27Bfamily root
  2. Arete-27B
  3. Arete-27B-Gosset-3.65BPW

New posts about versions of Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

GossetWorks/Arete-27B-Gosset-3.65BPW
Author
GossetWorks
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Language models

Arete-27B

Arete-27B