[Docs] Add PRISM test-time scaling reference for Dream-7B-Instruct in README

[viiika]

Hi authors,

We recently released a preprint on efficient test-time scaling for discrete diffusion language models (dLLMs), titled:

PRISM: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models

PRISM is an test-time scaling framework that integrates:

• Hierarchical Trajectory Search (HTS): dynamically prunes/reallocates compute in an early-to-mid denoising window
• Local branching via partial remasking: explores diverse realizations while preserving high-confidence tokens
• Self-Verified Feedback (SVF): uses the same dLLM as a lightweight Yes/No verifier on intermediate completions (no external RM required)

We evaluate PRISM on three dLLMs, including Dream-7B-Instruct. Would you consider adding a short note/link in the README (e.g., under “Decoding / Inference / Test-time scaling”) pointing to PRISM as a community decoding recipe tested on Dream-7B-Instruct?

Suggested README snippet (feel free to edit):

Test-time scaling (PRISM)

• PRISM proposes HTS + partial-remasking branching + self-verification (SVF) for efficient width scaling in dLLMs.
• Evaluated on Dream-7B-Instruct and shows strong accuracy–compute trade-offs on GSM8K/MATH500/HumanEval/MBPP.
• Code: https://github.com/viiika/Prism
• Paper: https://arxiv.org/abs/2602.01842

Thanks for considering!