Early Inkling 2 Mini LoRA tests: same results with under a third of the energy
ziqiao_ma · x · 2026-11-21
A developer training Inkling 2 Mini on a DGX Spark reports that LoRA fine-tuning only six layers matches results from all 24 layers while using less than a third of the energy.
ziqiaoma adds that the model is a good candidate for an OOD autoresearch benchmark variant for PostTrainBench or InferenceBench, noting big optimization headroom in looped transformer inference and post-trainability.
More from Infra
- Optical waveguide crossing lets light pass through while keeping circuits electrically isolated — jwt0625 · 2026-11-21
- Inkling 2 Mini quantized to 13.9GB runs at ~60 t/s decode on a single RTX 4090 — indrasmirror · 2026-11-21
- Sentdex: RTX PRO 6K stock returning and prices falling as loopholes close — good for consumers — Sentdex · 2026-11-21
- Son approached Wall Street to fund a $1T SK Hynix buyout, per Bloomberg scoop — tengyanAI · 2026-11-21
- RTX Spark desktops swap 200G ConnectX-7 for 10GbE, killing two-box clusters for large MoEs — AIFlow_ML · 2026-11-21
- AMD AI DevDay Paris on Dec 12: vLLM Maintainer to Talk Decode Optimization, Free ROCm Cert — AccBalanced · 2026-11-21