totob-1.5B

Overview

DeepSeek-R1 has garnered attention for matching OpenAI’s O1 reasoning model while being fully open-source, making it an attractive option for users who value local deployment for data privacy, reduced latency, and offline access. Traditionally, running such large models on personal devices involves quantization (e.g., Q4_K_M), which can compromise accuracy by as much as ~22% and diminish the benefits of local inference. With our new totob-1.5B model, we’ve overcome this trade-off by quantizing the DeepSeek-R1 Distilled model to just a quarter of its original size — without any loss in accuracy.

Benchmarks

Coming soon!!