Tencent
Tencent’s DSpine Cuts LLM Inference Time, Showing Draft Models Still Have Room
Tencent’s DSpine injects prior-position information across model layers, improving parallel draft coherence and inference throughput on Qwen3.
1 article tagged with this topic
Tencent’s DSpine injects prior-position information across model layers, improving parallel draft coherence and inference throughput on Qwen3.