# buoyancy99/diffusion-forcing

code for "Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion"

Repository: https://github.com/buoyancy99/diffusion-forcing
Canonical: https://ross.abutalabs.com/products/diffusion-forcing
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-07-06T02:31:32+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 91, release rhythm 35, longevity 57
- inputs: {"age_days": 798, "days_push": 59, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1288, forks 71 (observed 2026-08-28T04:04:15.186743+00:00)

## What it is
Official research code for the NeurIPS paper 'Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion', implementing a method that combines next-token prediction with full-sequence diffusion models. It supports video generation, planning, and robotics experiments with both RNN (paper branch) and temporal attention (main branch) implementations.

## Use cases
- reproduce results from the Diffusion Forcing paper
- generate videos with diffusion models
- train sequence models that combine autoregressive prediction with diffusion
- run planning experiments in maze environments
- apply diffusion forcing to robotics control tasks
- experiment with per-token diffusion noise levels

## When to choose
- you want to implement or extend the Diffusion Forcing method from the paper
- you need a research codebase for hybrid autoregressive-diffusion sequence modeling
- you are working on video generation or planning/robotics research based on this technique

## When to avoid
- you only want state-of-the-art video generation - use Diffusion Forcing v2 (diffusion-forcing-transformer) instead
- you need a production-ready library with stable APIs and support
- you want a plug-and-play tool rather than research code requiring conda setup and wandb configuration

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-training, video-processing, simulation
- domain: machine-learning, deep-learning, artificial-intelligence, robotics
- platform: python
- tags: diffusion-models, sequence-modeling, research-code, video-generation, planning, neurips-paper, pytorch, video, linux, gpu

## Member repositories
- buoyancy99/diffusion-forcing (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:15.186743+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:55:48.289759+00:00, confidence not recorded.
  - readme: https://github.com/buoyancy99/diffusion-forcing (fetched 2026-08-28T04:04:15.186743+00:00, sha ee78d9f448bd)
- Data as of 2026-08-30T08:39:29.467469+00:00.