Inductive ML

An independent lab studying how machines learn to learn.

We study how learned systems represent, adapt, and generalize, with experiments in representation geometry, efficient adaptation, and the systems that make models practical to run. Founded and run by Aditya Veer Parmar in Bangalore, India.

Website · Experiments · GitHub · About the lab

Explore our work

• LFM2.5 WebGPU Space · MONARCH — Run LFM2.5-230M in your browser with our WGSL kernels, generate text locally, and measure your GPU's tokens per second.
• Qwen3.6-35B-A3B · Evolved Mixed-Bit — A 12.63 GB MLX quantization with per-module precision selected by evolutionary search. Runs on Apple Silicon with stock mlx-lm.
• Ternary Bonsai kernels · Ternel — Custom Metal kernels run Bonsai 27B and Bonsai 2 27B directly from losslessly packed 1.75-bit weights on Apple Silicon. Bonsai 2 preserves the original FP16 vision tower and uses 6.92 GB of loaded model memory, 19.5% less than the official MLX checkpoint. Bonsai 2 with vision · Original Bonsai.

How we work

We publish experiments with their methods, measurements, and limitations. Correctness checks come before performance claims; recorded lab results are distinguished from measurements on your own hardware.

MONARCH's code is Apache-2.0. Its model and tokenizer data retain Liquid AI's LFM Open License v1.0; see each repository for its license and provenance.

Questions, replications, and technical feedback are welcome in our Space discussions.