SpermLLM is a compact distilled reasoning model based on Qwen3-0.6B-Instruct, designed to improve performance in math, coding, and structured reasoning while remaining lightweight and efficient.
Training Method
The model was fine-tuned on a mixture of curated instruction datasets and further distilled from larger teacher models (Mix of GPT-OSS-120B and Kimi K2.5)
Training Overview
Base Model: Qwen3 0.6B Instruct
Training Method: SFT (Supervised Finetuning) + Distillation
Notes
SpermLLM is an experimental model, We plan on making this larger and better! Currently no benchmarks but benchmarks will be soon! (Never)
