STACKQUADRANT

zengxiao-he/tessera

Inference Engines

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.

4.3
GitHub Metrics
Stars
526
Forks
9
Open Issues
Watchers
5
Contributors
1
Weekly Commits
0
Language
Python
License
NOASSERTION
Last Commit
Jun 5, 2026
Created
Jun 5, 2026
Latest Release
Release Date
Synced: Aug 28, 2026
Quality Scores
Documentation Qualityw: 20%
4.2

No dedicated docs site. Description: 232 chars. Stars signal: 526. Contributors: 1. Score: 4.2/10

Community Healthw: 20%
2.9

Stars: 526. Contributors: 1. Watchers: 5. Forks: 9. Issue ratio: 0.0%. Score: 2.9/10

Maintenance Velocityw: 15%
4.6

Last commit: 84d ago. Weekly commits: 0. No releases published. Score: 4.6/10

API Design & DXw: 20%
6.6

Stars/issues ratio: 526. Dynamic language: Python. No dedicated API docs. License: NOASSERTION. Popularity signal: 526 stars. Score: 6.6/10

Production Readinessw: 15%
3.2

Battle-tested: 526 stars. Peer review: 1 contributors. No versioned releases. Licensed: NOASSERTION. Age: 0.2 years. Maintenance: last commit 84d ago. Score: 3.2/10

Ecosystem Integrationw: 10%
4.3

Fork interest: 9. Major ecosystem: Python. License: NOASSERTION. Adoption: 526 stars. Score: 4.3/10

Tags
cudaflash-attentionfsdpinference-enginejaxknowledge-distillationkv-cachellmmechanistic-interpretabilityml-systems
Radar
Documentation Quality
Community Health
Maintenance Velocity
API Design & DX
Production Readiness
Ecosystem Integration