prism-diffusion-scaling

name	prism-diffusion-scaling
title	Prism: Efficient Test-Time Scaling via Hierarchical Search for Diffusion Language Models
version	0.0.2
engine	skillxiv-v0.0.2-claude-opus-4.6
license	MIT
url	https://arxiv.org/abs/2602.01842
keywords	["Test-Time Scaling","Discrete Diffusion","Inference Optimization","Adaptive Search","Self-Verification"]
description	Scale inference efficiency for discrete diffusion language models through hierarchical trajectory search with adaptive pruning and self-verified feedback. Achieve 3-4× speedup versus best-of-N with equal quality.

Prism: Efficient Test-Time Scaling for Discrete Diffusion Language Models

Most inference-scaling methods optimize for autoregressive decoding, but discrete diffusion models generate sequences through parallel iterative denoising—fundamentally different. Naive best-of-N approaches require O(NT) function evaluations (N trajectories, T denoising steps) which is computationally prohibitive. Prism introduces three innovations: hierarchical trajectory search that progressively prunes low-quality trajectories during mid-denoising, self-verified feedback reusing the model as verifier, and partial remasking preserving high-confidence tokens while exploring alternatives.

The key insight is that diffusion's bidirectional context enables effective mid-generation pruning unavailable to autoregressive models.

Core Concept

Prism uses three complementary mechanisms:

1. Hierarchical Trajectory Search (HTS): Divide inference into stages with geometric decay of active trajectories from N → K during early denoising when "logic skeletons" stabilize, then final refinement with K survivors

2. Self-Verified Feedback (SVF): Reuse the dLLM itself as verifier through dedicated Yes/No prompts on intermediate completions, eliminating overhead of separate models

3. Local Branching via Partial Remasking: Preserve high-confidence tokens as fixed "logic skeleton" while selectively re-masking low-confidence positions, enabling diverse exploration within fixed budget

These reduce complexity to approximately O(N + KT), achieving significant speedups.

Architecture Overview

• Trajectory Initializer: Launch N parallel trajectories
• Denoising Step Executor: Standard diffusion denoising step on all active trajectories
• Confidence Scorer: Estimate token confidence from model logits
• Pruning Engine: Geometric decay schedule reducing trajectories from N → K
• Self-Verification Module: Query dLLM for Yes/No feedback on intermediates
• Partial Remasking: Selectively remask low-confidence positions
• Skeleton Preservation: Keep high-confidence tokens fixed across variants

Implementation

The method involves trajectory management, confidence scoring, and adaptive pruning.

Initialize and manage parallel diffusion trajectories:

import torch
from typing import , 

 :
    

     ():
        .trajectories = [initial_noise.clone()  _  (num_trajectories)]
        .active_mask = torch.ones(num_trajectories, dtype=torch.)
        .confidence_scores = []
        .step_count = 

     ():
        
        .active_mask = torch.zeros((.trajectories), dtype=torch.)
        .active_mask[new_active_indices] = 

        
        .trajectories = [.trajectories[i]  i  new_active_indices]

     ():
        
         [traj  traj, active  (.trajectories, .active_mask)  active]

     ():
        
        .step_count = step_idx

batch = TrajectoryBatch(initial_noise, num_trajectories=)