# Ideas in Inference-time Scaling can Benefit Generative Pre-training Algorithms

arXiv: https://arxiv.org/abs/2503.07154 · HTML: https://mlsanity.com/paper/2503.07154/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion reasoning and test-time scaling, one of 17 directions in LLM reasoning.
- Builds on 36 of its 81 references that are in the map.
- First posted on arXiv 10 Mar 2025.

## Placed in

- LLM reasoning / [Diffusion reasoning and test-time scaling](https://mlsanity.com/lr/diffusion-reasoning-and-test-time-scaling/)
- Diffusion models / [Discrete and text diffusion](https://mlsanity.com/diffusion/discrete-and-text-diffusion/)

## Builds on (most cited first, 30 of 36)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [Denoising Diffusion Implicit Models](https://mlsanity.com/paper/2010.02502/) (2020)
- [Score-Based Generative Modeling through Stochastic Differential Equations](https://mlsanity.com/paper/2011.13456/) (2020)
- [Flow Matching for Generative Modeling](https://mlsanity.com/paper/2210.02747/) (2022)
- [Auto-Encoding Variational Bayes](https://mlsanity.com/paper/1312.6114/) (2013)
- [Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow](https://mlsanity.com/paper/2209.03003/) (2022)
- [Categorical Reparameterization with Gumbel-Softmax](https://mlsanity.com/paper/1611.01144/) (2016)
- [Variational Inference with Normalizing Flows](https://mlsanity.com/paper/1505.05770/) (2015)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [Consistency Models](https://mlsanity.com/paper/2303.01469/) (2023)
- [Large Language Monkeys: Scaling Inference Compute with Repeated Sampling](https://mlsanity.com/paper/2407.21787/) (2024)
- [Large Language Diffusion Models](https://mlsanity.com/paper/2502.09992/) (2025)
- [Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference](https://mlsanity.com/paper/2310.04378/) (2023)
- [Pixel Recurrent Neural Networks](https://mlsanity.com/paper/1601.06759/) (2016)
- [DeepSeek-V3 Technical Report](https://mlsanity.com/paper/2412.19437/) (2024)
- [Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution](https://mlsanity.com/paper/2310.16834/) (2023)
- [Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion](https://mlsanity.com/paper/2407.01392/) (2024)
- [Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion](https://mlsanity.com/paper/2310.02279/) (2023)
- [One Step Diffusion via Shortcut Models](https://mlsanity.com/paper/2410.12557/) (2024)
- [Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data](https://mlsanity.com/paper/2406.03736/) (2024)
- [Better & Faster Large Language Models via Multi-token Prediction](https://mlsanity.com/paper/2404.19737/) (2024)
- [Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps](https://mlsanity.com/paper/2501.09732/) (2025)
- [Continuous diffusion for categorical data](https://mlsanity.com/paper/2211.15089/) (2022)
- [Diffusion Adversarial Post-Training for One-Step Video Generation](https://mlsanity.com/paper/2501.08316/) (2025)
- [Flow map matching with stochastic interpolants: A mathematical framework for consistency models](https://mlsanity.com/paper/2406.07507/) (2024)
- [Backpropagation through the Void: Optimizing control variates for black-box gradient estimation](https://mlsanity.com/paper/1711.00123/) (2017)
- [Rolling Diffusion Models](https://mlsanity.com/paper/2402.09470/) (2024)
- [Normalizing Flows are Capable Generative Models](https://mlsanity.com/paper/2412.06329/) (2024)
- [Progressive Distillation for Fast Sampling of Diffusion Models](https://mlsanity.com/paper/2202.00512/) (2022)

## Built on by (most cited first, 12 of 12)

- [Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding](https://mlsanity.com/paper/2505.22618/) (2025)
- [Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing](https://mlsanity.com/paper/2508.09192/) (2025)
- [Simulating the Visual World with Artificial Intelligence: A Roadmap](https://mlsanity.com/paper/2511.08585/) (2025)
- [DAPD: Dependency-Aware Parallel Decoding via Attention for Diffusion LLMs](https://mlsanity.com/paper/2603.12996/) (2026)
- [DAWN: Dependency-Aware Fast Inference for Diffusion LLMs](https://mlsanity.com/paper/2602.06953/) (2026)
- [Dimathtt[M]O: Distilling Masked Diffusion Models into One-step Generator](https://mlsanity.com/paper/2503.15457/) (2025)
- [Variational Masked Diffusion Models](https://mlsanity.com/paper/2510.23606/) (2025)
- [Rejection Mixing: Fast Semantic Propagation of Mask Tokens for Efficient DLLM Inference](https://mlsanity.com/paper/2602.22868/) (2026)
- [Gumbel Distillation for Parallel Text Generation](https://mlsanity.com/paper/2603.22216/) (2026)
- [K-Forcing: Joint Next-K-Token Decoding via Push-Forward Language Modeling](https://mlsanity.com/paper/2606.10820/) (2026)
- [FluxLite: Inference-Time Proposal Control for Discrete Diffusion Models](https://mlsanity.com/paper/2609.35947/) (2026)
- [Dependency-Aware Revocable Decoding for Efficient Diffusion Large Language Model Inference](https://mlsanity.com/paper/2608.26574/) (2026)

## Abstract (arXiv)

Generative pre-training is often framed through a false dichotomy between autoregressive models for discrete signals and diffusion models for continuous signals. We argue that the dichotomy is false because it conflates model family, data representation, training objective, and inference procedure. Autoregression is an inference procedure that expands a sequence through normalized conditional draws, while diffusion is a refinement procedure that repeatedly revises an existing state. The more useful contrast is therefore not autoregressive versus diffusion, but discrete tokens learned with cross-entropy versus continuous tokens learned with diffusion-style objectives, together with the inference algorithms used to sample from them. From this perspective, algorithmic progress should prioritize inference-time efficiency along two axes: sequence expansion and state refinement. We advocate designing the inference procedure before the training objective, because a training method cannot compensate for an inference map that omits necessary arguments or imposes an incorrect factorization. We illustrate this principle through a target-time limitation of DDIM-style samplers, a joint-distribution limitation of multi-token prediction, and recent flow-map and few-step distillation methods that directly parameterize long-range inference moves.