New Inference Server for DGX Spark: large model C4:55-90 tok/s no spec decode

A new inference server for DGX Spark achieves 55-90 tok/s on large models without speculative decoding.

• DGX Spark
• large model
• inference server
• token speed