Skip to content
Other

AutoDrive-OPD-E057O-student

Qwen3-VL Based Autopilot Model

2 min readupdated

This model is a specialized fine-tune of the Qwen3-VL-4B-Instruct vision-language model, trained for autonomous driving tasks in the NAVSIM simulator. It takes an image from the front camera and a text prompt with driving parameters (navigation command, current speed, acceleration, angular velocity, and past trajectory) as input, and outputs a block containing the predicted trajectory for the next 4 seconds. The training was performed using on-policy distillation based on two expert reinforcement learning (RL) models. This development will be useful for researchers in autonomous transport and computer vision.

Key features

  • Qwen3-VL Architecture: Full-parameter training of a 4-billion-parameter model with support for image and text processing.
  • Policy Distillation: Leveraging data from two RL-driven experts to improve stability.
  • Precise Trajectory Prediction: Generation of 8 coordinate points [x, y, yaw] at 0.5-second intervals in the vehicle's current coordinate system.
  • NAVSIM Evaluation: Tested on 12,146 benchmark test scenes with zero parsing errors.

What you can do

  • Testing in NAVSIM: Evaluate autonomous driving performance on simulation datasets derived from nuPlan.
  • Autonomous Systems Research: Use as a baseline for experiments with end-to-end trajectory planning.
  • Pipeline Integration: Apply generation and evaluation scripts to analyze multimodal agent performance on road scenes.

Example

The model takes a frame from the CAM_F0 front camera and a string prompt containing current speed, acceleration, and past agent poses as input, and then generates a tag with the coordinates of the motion trajectory points.

Read moreDetails
Benchmarks: navtest PDMS — 0.9227, navtest EPDMS (guarded) — 0.9125, navhard EPDMS — 0.3022.
Inference parameters: greedy decoding, max_new_tokens=512, min_pixels=262144, max_pixels=1048576.
Base model: Qwen3-VL-4B-Instruct (Apache-2.0 license), NAVSIM / OpenScene (nuPlan) dataset.
Discuss Telegram 0 42

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 9.0 GB · ready files10 GB
  • FP8 / INT8recommendedweights 4.1 GB5.0 GB
  • 4-bitweights 2.3 GB3.1 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 2080 Ti

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.

    # wulin222/AutoDrive-OPD-E057O-student
    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 wulin222/AutoDrive-OPD-E057O-student --local-dir ~/models/AutoDrive-OPD-E057O-student
    # 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

New posts about versions of Qwen3-VL-4B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 5All on Hugging Face

About the model

wulin222/AutoDrive-OPD-E057O-student
Author
wulin222
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

420

Ask AI about this model

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

Comments