Kimi K3
Kimi K3 Replication Reveals the New Battleground in LLM Inference
A CPU-based Kimi K3 replication breaks KDA inference into five layers, showing LLM competition is shifting toward reuse, latency, and deployment cost.
Back to home
LLM inference
2 articles tagged with this topic
Qwen
Qwen's SpecOPD Combines Branch-Parallel Training with Speculative Decoding
Qwen's SpecOPD accelerates inference by combining speculative decoding with branch-parallel training, achieving gains of up to 3.56x in generation spe