Skip to content
Robotics

LiberalMind Video VAE

video variational autoencoder

2 min readupdated

This model from the LiberalMind research initiative handles video sequence compression and reconstruction, serving as a core component for robotics systems and world models. The architecture is trained on 256×256 pixel clips of 17 frames using a combination of frame-by-frame reconstruction, LPIPS perceptual similarity, and latent regularization. The checkpoint targets researchers in embodied AI and temporal visual data analysis. To run the model, you need to separately connect the architecture description file vae.py in addition to the weights.

Key features

  • Video representation: Encoding and reconstructing short video sequences using a variational autoencoder (VAE).
  • TPU training: Configuration optimized for Google Cloud TPU using the PyTorch/XLA stack and SPMD parallelism.
  • Perceptual loss: Using the LPIPS (VGG) metric to evaluate visual similarity.
  • Data format: Processing input tensors of dimension [B, 3, 17, 256, 256] in a precise BF16 grid.

What you can do

  • Robotics research: Use the model as a foundation for extracting visual features in perception tasks for autonomous agents.
  • Video sequence analysis: Apply it for temporal visual understanding tasks and creating compact latent representations.
  • Run training on TPU: Deploy the fine-tuning pipeline on custom datasets using the provided instructions and scripts.

Example

Example code for loading the downloaded checkpoint and initializing the model:

import torch
from vae import VAE

checkpoint = torch.load(
    "last.pt",
    map_location="cpu",
    weights_only=False,
)

model = VAE()
model.load_state_dict(checkpoint["model"])
model.eval()

print("Model loaded successfully")
print("Training step:", checkpoint["step"])
Read moreDetails
Input resolution: 256×256 pixels
Sequence length: 17 frames
Optimizer: AdamW (learning rate 1e-4)
Framework: PyTorch / PyTorch/XLA
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 5.5 GB · ready files6.5 GB
  • FP8 / INT8recommendedweights 2.7 GB3.5 GB
  • 4-bitweights 1.5 GB2.2 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 5060 Ti 8GB

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.

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

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

About the model

LiberalMind/vae-model
Author
LiberalMind
Category
Robotics
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