tessera
by zengxiao-he · github.com/zengxiao-he/tessera
From teacher to tiles — a from-scratch LLM distillation & serving engine: custom Triton/CUDA kernels, FSDP distillation, paged-KV continuous batching, speculative decoding, a Rust gateway, a JAX oracle, and interpretability tooling.
Stars
★ 588
Forks
9
Language
Python
Category
LLM Tools
License
NOASSERTION
cudaflash-attentionfsdpinference-enginejaxknowledge-distillationkv-cachellmmechanistic-interpretabilityml-systemspaged-attentionpytorchpython
Featured in
- tessera @ 14:02 · Jun 14, 2026