Skip to content
Robotics

Jevol-2B

Autonomous Drone Navigation Powered by Qwen

2 min readupdated

Jevol-2B from Racine.ai is an AI model built for autonomous drone control using camera feeds and text mission instructions. Every 0.4 seconds, it processes a 672x504 frame alongside an English task prompt to output movement, altitude, and offset commands in a single pass. Based on the Qwen/Qwen3.5-2B architecture with 24 decoder layers, the model takes up 4.47 GB on disk. It keeps no history or terrain map, making decisions strictly from the current frame, which makes it a solid fit for robotics researchers and autonomous system developers.

Key features

  • Built on Qwen3.5-2B with the original language head removed and replaced by a text head.
  • Averages 165 ms per command on an NVIDIA L40S GPU.
  • Features built-in refusal for dangerous commands ("crash into...") without external filters.
  • Achieves a 50.7% collision-free flight pass rate on the simulation benchmark.

What you can do

  • Run the flight simulator locally using the project's GitHub repository and Node.js.
  • Test the model from the command line to generate flight commands from an image and text prompt.
  • Integrate the model checkpoints into a custom UAV control pipeline via CLI.

Example

git clone [https://github.com/RacineAIOS/jevol](https://github.com/RacineAIOS/jevol) && cd jevol
uv sync --extra cuda --extra server
uvx --from huggingface_hub hf download racineai/Jevol-2B --local-dir checkpoints/Jevol-2B
Read moreDetails
Parameters: 2.21B parameters (331M for vision).
Benchmark: 50.7% pass rate (36.3–64.9% confidence interval).
Hardware: Tested on NVIDIA L40S (165 ms per decision).
License: Racine Research License, version 1.0 (commercial use requires separate approval).
Examples and videos
Discuss Telegram 6

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.1 GB · ready files5.0 GB
  • FP8 / INT8recommendedweights 2.1 GB2.8 GB
  • 4-bitweights 1.2 GB1.8 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 · BF16 — from RTX 2060

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.

    # racineai/Jevol-2B
    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 racineai/Jevol-2B --local-dir ~/models/Jevol-2B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Jevol-2B
  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. Qwen3.5-2B-Basefamily root
  2. Qwen3.5-2B
  3. Jevol-2B

New posts about versions of Qwen3.5-2B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 10All on Hugging Face
Merges 2All on Hugging Face

About the model

racineai/Jevol-2B
Author
racineai
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

06

Ask AI about this model

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

Comments

Audio

Qwen3-ASR

Qwen3-ASR