zhuhanqing/APOLLO

APOLLO: SGD-like Memory, AdamW-level Performance; MLSys'25 Oustanding Paper Honorable Mention

35
/ 100
Emerging

Combines low-rank gradient approximation with channel-wise or tensor-wise scaling factors computed via pure random projections in an auxiliary space, eliminating expensive SVD computations. Integrated into Hugging Face Transformers, LLaMA-Factory, and axolotl, with support for distributed training via FSDP and quantization-aware variants (QAPOLLO with int8 weights).

271 stars.

No Package No Dependents
Maintenance 6 / 25
Adoption 10 / 25
Maturity 9 / 25
Community 10 / 25

How are scores calculated?

Stars

271

Forks

13

Language

Python

License

Last pushed

Nov 29, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/llm-tools/zhuhanqing/APOLLO"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.