Small models, post-trained deliberately.
A Go reproduction of the VibeThinker pipeline: the Spectrum-to-Signal recipe that takes small dense reasoning models through diversity-exploring SFT, expert fusion, and MaxEnt-guided reinforcement learning.
Research reproduction. Every stage of the published method is implemented on the mlx-go stack with property tests over the design's correctness invariants, and the full recipe runs end to end on toy configurations. Real-model evaluation requires weights you provide (none are redistributed), and the canonical real-model result so far is parity with the base model, not an improvement. The page says so because the repo does.
What it does- Spectrum SFT — diversity-exploring supervised fine-tuning that probes checkpoints per subdomain with an unbiased Pass@K estimator.
- Expert fusion — merges subdomain-specialist checkpoints into one SFT model via weighted parameter averaging over safetensors.
- MaxEnt-guided RL — modulates GRPO advantages with a max-entropy-deviation weight, focusing optimization on prompts with high exploratory value.
- Offline self-distillation — filters correct reasoning traces by length-normalized likelihood to build the next round of student data.
- Evaluation harness — commands for method comparison, toy end-to-end runs, and real-model evaluation against user-provided weights.
mlx-go-vibethinker exists to answer a question the family cares about: can the published recipes that make small models reason be reproduced, audited, and run on hardware you own? It builds on mlx-go and the model registry in mlx-go-lm, keeps external requirements behind pluggable seams, and reports what it measures, including when the measurement is parity. It sits alongside mlx-go-sia and the federated-distillation work in mlx-go-fedistill as the research shelf of the stack.