Search papers, labs, and topics across Lattice.
Institue of Foundation Models
2
0
7
Auxiliary draft models are no longer necessary for speculative decoding: distilling lightweight diffusion heads directly into standard LLMs yields lossless 3脳 generation speedups even at peak batch sizes.
Achieve SOTA reasoning performance with a 30B parameter model using up to 95% fewer tokens by explicitly controlling when and how deeply an LLM plans.
