Single DGX Spark Tests 124B Model with High-Speed Decoding

Benchmark tests on a single DGX Spark reveal that the 124B Ling-3.0-flash model achieves 38.7 tok/s with almost no decoding speed loss during quantization, thanks to its MoE architecture.

2026-08-12 ~ 2026-08-14 · 2 related posts

• Ling-3.0-flash Quantization Benchmarks: MoE Architecture Preserves Decode Speed — AcanthisittaOk1699 · 2026-08-12
• Single DGX Spark Decodes 15K Tokens from 124B Model with Stable Throughput — AcanthisittaOk1699 · 2026-08-14