2121-8/TinySlime-1.1B-Chat-v1.0
TinySlime-1.1B-Chat-v1.0 by 2121-8 is a 1.1 billion parameter, 2048-token context length chat model specifically fine-tuned for Japanese language tasks. It is based on TinySlime-1.1B-v1.0 and uses synthetic data generated by Mixtral-8x7B-Instruct-v0.1. This model is optimized for deployment on embedded environments such as smartphones and NVIDIA Jetson, demonstrating strong performance in Japanese language benchmarks.
Loading preview...
TinySlime-1.1B-Chat-v1.0 Overview
TinySlime-1.1B-Chat-v1.0 is a compact, 1.1 billion parameter language model developed by 2121-8, specifically designed for chat applications in Japanese. It is built upon the TinySlime-1.1B-v1.0 base model and has been fine-tuned using synthetic data generated by Mixtral-8x7B-Instruct-v0.1. A key focus during its development was optimization for resource-constrained embedded environments, including smartphones and NVIDIA Jetson devices.
Key Capabilities & Performance
This model demonstrates competitive performance in Japanese language benchmarks, as evaluated by the JP Language Model Evaluation Harness. Notably, TinySlime-Chat achieves:
- 66.49 on JCommonsenseQA (3-shot)
- 64.36 on JNLI (3-shot)
- 92.68 on MARC-ja (0-shot)
- 84.60 on JSQuAD (2-shot)
These scores highlight its proficiency in various Japanese understanding and reasoning tasks, often outperforming other models in its size class across several metrics.
Intended Use Cases
TinySlime-1.1B-Chat-v1.0 is particularly well-suited for:
- Japanese-centric chat applications: Its fine-tuning with chat data makes it effective for conversational AI in Japanese.
- Edge device deployment: Designed for low-resource environments, it's ideal for running on mobile devices or embedded systems where larger models are impractical.
- Applications requiring efficient Japanese language processing: Its small size combined with strong Japanese performance makes it a good choice for various localized tasks.