LoRA fine-tunes.

Vertical fine-tunes shipped with the research story attached — what the model does, who picks this training lane, how it was evaluated on Spark, and what drifts you should expect bounded by count. 01 published.

LoRA fine-tune trained with nemo BF16

patent-strategist-v3-nemo

Offline patent-prosecution reasoning on Spark-class hardware

Patent prosecution work — claim construction, MPEP-grounded office-action responses, Markush analysis, doctrine-of-equivalents reasoning — happens inside firms that can't ship privileged client text to a hosted frontier API. This release distills DeepSeek-R1's chain-of-thought reasoning onto a 5,000-row synthetic patent-reasoning corpus so a single Spark-class box can run the workflow offline, with full IRAC-shaped reasoning chains.

free apache-2.0 base: DeepSeek-R1-0528-Qwen3-8B
More fine-tunes in preparation End of May 2026