Medusa GitHub repository (stale) · inference-research

Fields

id

sources/niche-medusa-repo

URL

https://github.com/FasterDecoding/Medusa

Publisher

FasterDecoding

Kind

repo

Accessed

2026-10-03

notes

Medusa README. Multiple decoding heads added on top of a frozen base model so no separate draft model is needed; Medusa-2 adds full-model training and self-distillation via an axolotl fork. README states plainly: 'We currently support single-GPU inference with a batch size of 1'. API metadata 2026-10-03: created 2023-09-10, last push 2024-06-25, only release ever tagged v0.1 (2023-09-11), Apache-2.0, 100 open issues, 2770 stars. A >2-year gap between last push and today is the maintenance evidence.

Notes

Medusa README. Multiple decoding heads added on top of a frozen base model so no separate draft model is needed; Medusa-2 adds full-model training and self-distillation via an axolotl fork. README states plainly: 'We currently support single-GPU inference with a batch size of 1'. API metadata 2026-10-03: created 2023-09-10, last push 2024-06-25, only release ever tagged v0.1 (2023-09-11), Apache-2.0, 100 open issues, 2770 stars. A >2-year gap between last push and today is the maintenance evidence.

Cross-references

This record references no other records.