VORTEX's picture

Building on HF

VORTEX

Abhaykoul

·

OEvortex

AI & ML interests

None yet

Recent Activity

reacted to KingNish's post with 🔥 4 days ago

Muon vs MuonClip vs Muon+Adamw Muon has gone from an experiment to a mainstream optimizer, but does it hold up for fine‑tuning? We ran head‑to‑head tests on Qwen3‑4B (10k+ high‑quality instruction rows) to find out. Short story: Pure Muon converged fastest at the start, but its gradient‑norm spikes made training unstable. MuonClip (Kimi K2’s clipping) stabilizes long pretraining runs, yet in our small‑scale fine‑tune it underperformed, lower token accuracy and slower convergence. The winner was the hybrid: Muon for 2D layers + AdamW for 1D layers. It delivered the best balance of stability and final performance and even beat vanilla AdamW. Takeaway: for small-scale fine-tuning, hybrid = practical and reliable. Next Step: scale to larger models/datasets to see if Muon’s spikes become catastrophic or if clipping wins out. Full Blog Link: https://huggingface.co/blog/KingNish/optimizer-part1

View all activity

Organizations

Abhaykoul 's datasets 11

Abhaykoul/priya-v1

Viewer • Updated Oct 17 • 7.64k • 23 • 1

Abhaykoul/title-gen

Viewer • Updated Aug 27 • 6.36k • 18

Abhaykoul/hai3-textmoderation

Viewer • Updated Aug 14 • 6.98k • 24

Abhaykoul/tools

Viewer • Updated Jul 18 • 1k • 7

Abhaykoul/HAI-SER

Viewer • Updated Jan 29 • 7.63k • 30

Abhaykoul/jenny-tts-text-tags-6h-v1

Viewer • Updated Dec 8, 2024 • 4k • 19

Abhaykoul/test

Viewer • Updated Nov 3, 2024 • 28.2k • 41

Abhaykoul/me_dpo

Viewer • Updated Jun 11, 2024 • 1.46k • 13

Abhaykoul/abhay

Viewer • Updated May 23, 2024 • 1.37k • 79 • 1

Abhaykoul/JARVIS

Viewer • Updated May 23, 2024 • 1.5k • 24 • 2

Abhaykoul/Ancient-Indian-Wisdom

Viewer • Updated May 1, 2024 • 616 • 180 • 4