Spec decoding: Models in our catalog come assigned with an assigned drafter model for speculative decoding based on the best known method and model available for that target model (support DFlash, DSpark, and DFlash2).
Using too much memory for KV cache: We use a TurboQuant-inspired quantization of KV cache to 8-bit keys and 4-bit values. This drops KV memory usage by over half and also speeds up decode. Based on long context quality benchmarking we've done it does not seem to negatively impact retrieval or coherence over long context.
Large context sizes: our KV quantization helps a lot for this, and we focus our optimizations on specifically longer-context requests since that's what most agent inference actually looks like.
Quantization is another thing. There are so many engines launched with claims about speed, but in many cases it's optimizing specific lower-quality quants. When you have enough resources, you usually want something like W8A16 + full precision KV cache working as fast as possible, not yet another W8A8 or W4A16.
In general, it seems new models are released so fast now - engines don't always have time to really polish the implementation before the next model is released
All our benchmarks are open source so you can check it out here if you'd like: https://github.com/magnitudedev/magnitude/blob/main/inferenc...