Skip to content
Other

Relay-Forcing

checkpoints for real-time streaming video generation

2 min readupdated

This repository contains the weights and checkpoints for Relay Forcing, a method built for real-time streaming video generation at 1.58 NFE. Inside, you will find core models—including variants for Wan2.2-TI2V-5B and LTX-2.3-22B—alongside an auxiliary VAE decoder and a suite of models for ablation studies. The models build on the Wan2.2 and LTX-2.3 architectures, augmented with custom trajectory heads and EMA (exponential moving average) weights. This release is useful for researchers and developers working on accelerating video generation and streaming architectures.

Key features

  • Supports real-time streaming video generation at 1.58 NFE.
  • Includes core checkpoints for Wan2.2-TI2V-5B and LTX-2.3-22B (15-second and 60-second versions).
  • Ships with full training configurations and distributed PyTorch checkpoints.
  • Provides a large set of ablation study models (DMD, GAN step, and head scaling variants).
  • Uses EMA (exponential moving average) weights across all reported results.

What you can do

  • Run and benchmark real-time video generation powered by Wan2.2-TI2V-5B or LTX-2.3-22B.
  • Use the included latent causal group VAE decoder when working with LTX-2.3-22B.
  • Inspect the training configurations and ablation experiment results from the paper.
  • Write custom inference code by loading the weights with the provided Python script.

Example

import json
from pathlib import Path

from huggingface_hub import snapshot_download
from safetensors.torch import load_file

folder = "wan2.2-ti2v-5b/relay-forcing"
root = Path(snapshot_download("oliveryanzuolu/Relay-Forcing", allow_patterns=[f"{folder}/model*", f"{folder}/config.yaml"])) / folder
index = json.loads((root / "model.safetensors.index.json").read_text())
state_dict = {}
for shard in sorted(set(index["weight_map"].values())):
    state_dict.update(load_file(root / shard))
Read moreDetails
Total repository size is 1.79 TB.
Base models: Wan2.2-TI2V-5B and LTX-2.3-22B using Gemma 3 12B encoders.
Inference and training code will be released separately.
Discuss Telegram 0

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 63 GB · ready files70 GB
  • FP8 / INT8recommendedweights 32 GB35 GB
  • 4-bitweights 18 GB20 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.

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.

    # oliveryanzuolu/Relay-Forcing
    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 oliveryanzuolu/Relay-Forcing --local-dir ~/models/Relay-Forcing
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Relay-Forcing
  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 Wan2.2-TI2V-5B (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

oliveryanzuolu/Relay-Forcing
Author
oliveryanzuolu
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