METASTONE推出多模态推理服务
Story overview
是石科技(METASTONE)发布「多模态模型服务」,面向生产工作流,整合视频生成加速、缓存与显存管理、token 压缩及 Context Engine(把参考资料转成结构化生成指令)。 公司自测称,其基于历史缓存残差的自适应加速技术在展示的测试条件下相比 Dense 基线实现端到端平均 2.48 倍加速并保持生成质量;该结果未经独立验证。服务另结合 SGLang 优化缓存与显存管理,并以模态自适应免训练 Token 压缩在 WorldSense 评测中实现 1.33 倍加速。
AI-generated from reports · Updated 41m ago
Report timeline
Follow the reports to explore different perspectives.
- PandailyMETASTONE launches multimodal inference service, reports 2.48x video speedup in self-tests
Chinese computing infrastructure provider METASTONE has launched a multimodal model service for production workflows, combining video-generation acceleration, cache and memory management, token compression and a Context Engine that turns reference material into structured generation instructions. The company reports an average end-to-end speedup of 2.48 times over a dense-computation baseline with quality maintained, but these are vendor self-tests, not independently verified.
- 量子位 QbitAIShishi Technology launches multimodal model service with 2.48x video generation speedup
Chinese AI company Shishi Technology released a multimodal model service that it says speeds up video generation inference by an average of 2.48x end-to-end versus a Dense baseline while maintaining quality, using adaptive acceleration based on historical cache residuals. The service also pairs cache and memory management with SGLang and adds training-free modality-adaptive token compression, which it reports as 1.33x faster on the WorldSense benchmark.
Story activity trend
Not enough continuous observations to show a trend yet.