Sol Video Inference Engine: Agent-Native Full-Stack Acceleration Framework for Efficient Video Generation
Quick Answer
The Sol Video Inference Engine is a training-free acceleration framework for video diffusion models, achieving over 2x end-to-end acceleration with near-lossless VBench quality across models like Cosmos3-Super and LTX-2.3.
Quick Take
The Sol Video Inference Engine is a training-free acceleration framework for video diffusion models, achieving over 2x end-to-end acceleration with near-lossless VBench quality across models like Cosmos3-Super and LTX-2.3. By utilizing techniques such as cache, sparse attention, and token pruning, it optimizes performance with minimal human intervention.
Key Points
- Achieves over 2x acceleration while maintaining near-lossless quality.
- Utilizes five techniques: cache, sparse attention, token pruning, quantization, and kernel fusion.
- Optimizes performance for specific model and hardware configurations.
- Demonstrated on models including 64B Cosmos3-Super and 22B LTX-2.3.
- Minimizes manual tuning efforts through an agentic optimization framework.
Paper Resources
Article Content
From source RSS / original summaryarXiv:2606. 23743v1 Announce Type: new Abstract: Modern video diffusion models achieve higher generation quality through scaling, but this also increases inference cost. Although many acceleration methods have been proposed, a central challenge is that the most effective acceleration strategy is highly instance-specific: a recipe that works well for one combination of model, hardware, and inference configuration often does not transfer to another.
Different models vary in architecture, numerical sensitivity, and attention concentration patterns. Inference settings differ in spatial and temporal resolution and video duration, while hardware platforms differ in memory hierarchy, supported numerical formats, and kernel throughput. These factors create a large tuning space, making manual performance engineering costly. We present Sol Video Inference Engine, an agentic, native, training-free acceleration framework for video diffusion models.
It organizes five broadly applicable techniques, cache, sparse attention, token pruning, quantization, and kernel fusion, into an agentic acceleration stack for instance-specific optimization. For a concrete deployment target defined by a model, hardware platform, and serving configuration, parallel skill agents optimize the implementation of each technique, an agent integrator composes them into a global acceleration stack, and a human validator provides feedback on generation quality.
We instantiate this workflow on three video models with different sizes and architectures: 64B Cosmos3-Super, 22B LTX-2. 3, and 2B SANA-Video. With little human effort, the full stack achieves more than 2x end-to-end acceleration while maintaining near-lossless VBench quality, demonstrating the effectiveness of the agent framework for video diffusion acceleration.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →LLM-Guided ANN Index Optimization for Human-Object Interaction Retrieval
A phase-aware LLM agent optimizes human-object interaction retrieval, outperforming Optuna TPE by 33.3% and VDTuner by 34.2% on the HICO-DET benchmark. This method enhances throughput by 15.3x over UniIR and demonstrates strong transferability across vector database management systems.