Sol-H3: Recursive Self-Improvement for MiniMax-H3 Inference Acceleration on Sol-Engine across Cloud and Edge

2026-09-28

AI Summary

This study addresses the high deployment latency and memory constraints of the MiniMax-H3 video diffusion model, which stem from its massive parameter count and multi-step denoising process, by constructing a full-stack inference acceleration pipeline. Methodologically, it proposes a cross-resolution two-stage scheduling strategy coupled with latent-space mapping to eliminate redundant VAE encoding and decoding operations. Furthermore, a recursive self-improvement (RSI) mechanism is introduced to automatically search for optimal kernel fusion and memory layout configurations. Experimental results demonstrate that this approach achieves up to 30× end-to-end speedup and a 20% reduction in GPU memory consumption. Notably, a single DGX Spark card can complete video generation within one minute, effectively overcoming the computational bottlenecks associated with cloud-based deployment.

Abstract

Video diffusion models are rapidly scaling and exhibiting enhanced generation capabilities. Among these recent advancements, MiniMax-H3 stands out as a highly capable, production-level open-source model. However, its 33-billion parameters and multi-step iterative denoising process introduce substantial computational overhead. Consequently, their practical production is hindered by generation latency in the cloud deployment like NVIDIA-GB200, alongside strict memory limits that pose further challenges at the edge device like DGX-Spark. To address these diverse hardware bottlenecks from cloud to edge device, we present a full-stack inference pipeline that integrates efficient algorithmic design with optimized operator implementations. Algorithmically, we introduce a cross-resolution two-stage generation scheduler that exploits the step-wise nature of diffusion: early low-resolution steps rapidly establish the global layout, while later high-resolution steps focus refinements of local and perceptual details. These stages are connected by a learned latent-to-latent mapping module, completely eliminating the computationally expensive VAE decode-reencode cycle for resolution transferring cross different resolutions. For operator implementation, we deploy a Recursive Self-Improvement (RSI) loop that searches kernel fusions and memory layouts, evaluating latency together with numerical agreement. Together, these optimizations deliver up to 30x end-to-end speedup and 20% lower memory: a 5-second 1344x768 video with audio is generated 3.5x faster than real time on an 8xGB200 node, and in under a minute fully memory-resident on a single DGX Spark.

Problem

• Video Diffusion Models
• Inference Acceleration
• Computational Overhead
• Generation Latency
• Memory Constraints

Innovation

• Video Diffusion Models
• Cross-Resolution Scheduler
• Latent-to-Latent Mapping
• Recursive Self-Improvement
• Inference Acceleration