AI & ML interests

Local LLM benchmarking and model tuning on NVIDIA DGX Spark. Quantisation, speculative decoding, abliteration.

Recent Activity

scottleimroth  updated a model about 3 hours ago
leimroth-lab/Unbound-LR7
scottleimroth  published a model about 4 hours ago
leimroth-lab/Unbound-LR7
View all activity

Organization Card

Leimroth Lab

Independent AI research on local hardware. Benchmarking and model tuning on NVIDIA DGX Spark (GB10).

Models here are modified builds of open base models, published with their method and evaluation shown so the results can be reproduced.

Benchmarks: local models compared on DGX Spark hardware, each served at its own best configuration, with the measurement method shown alongside every score. https://scottleimroth.com/ai-technology/spark-benchmarks

Naming: builds are numbered in sequence. Earlier builds are named after their base model plus a Leimroth number (Qwen3.8-27B-Leimroth3); newer builds have their own name plus an LR number (Dream Shimmer LR6).

Scott Leimroth · @ScottLeimroth · github.com/scottleimroth

datasets 0

None public yet