Skip to content
Other

GyroFlow

generative checkpoints for gyrokinetics

2 min readupdated

A temporary release of weights for the GyroFlow scientific model, built for simulating plasma turbulence and calculating heat flux. It includes checkpoints for an autoencoder (a model that compresses data into a compact representation) and a flow-matching diffusion transformer (DiT — an architecture for generating states along trajectories), trained as part of the neural-gyrokinetics project by ml-jku. The repository lets you reproduce heat flux estimates on pre-computed trajectories without downloading the full training dataset. Ideal for researchers in computational physics and scientific machine learning.

Key features

  • Two-stage pipeline: a trained autoencoder (400 epochs) and a flow-matching DiT (120 epochs) in PyTorch (.pth) format.
  • Includes the original training configs (config.yaml), generation parameters (generate_gyroflow.yaml), and training set normalization statistics.
  • To run quality evaluations, you only need to download the lightweight trajectory metadata from the cbc-gyroswin-256traj dataset.

What you can do

  • Reproduce time-averaged heat flux calculations (Qbar RMSE) across 31 in-distribution (ID) test trajectories as well as out-of-distribution (OOD) scenarios.
  • Use the pre-trained weights and normalization statistics as a starting point for ML experiments in gyrokinetic modeling based on the ml-jku/neural-gyrokinetics repository.

Example

from huggingface_hub import snapshot_download

snapshot_download(
    "gerkone/cbc-gyroswin-256traj",
    repo_type="dataset",
    allow_patterns=["id/*/metadata_light.npz", "ood/*/metadata_light.npz"],
    local_dir="data/preprocessed_kvikio"
)
Read moreDetails
Expected Qbar RMSE on the 31 ID trajectories is 19.0 (Table 1 from the paper), and around 14 on ood_iteration_0-4.
Sampling parameters: 128 samples, 15 Euler steps, latent_scale 0.0478053.
Running the evaluation script requires PR #10 from the ml-jku/neural-gyrokinetics repository; a full release is expected later.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.4 GB · ready files1.0 GB
  • FP8 / INT8recommendedweights 0.2 GB0.7 GB
  • 4-bitweights 0.1 GB0.6 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.

    # gerkone/gyroflow-tmp
    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 gerkone/gyroflow-tmp --local-dir ~/models/gyroflow-tmp
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to gyroflow-tmp
  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 gyroflow-tmp (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

gerkone/gyroflow-tmp
Author
gerkone
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments