Model Overview

The uukuguy/speechless-hermes-coig-lite-13b is a 13 billion parameter language model built upon the Llama-2 architecture. It is a fine-tuned version of the Nous-Hermes-Llama2-13b model, specifically enhanced with the COIG-PC-LITE dataset to introduce and improve its Chinese language capabilities.

Key Characteristics & Capabilities

• Base Model: Fine-tuned from Nous-Hermes-Llama2-13b, which was developed by Nous Research.
• Instruction Following: The base model was fine-tuned on over 300,000 instructions, primarily using high-quality synthetic GPT-4 outputs from diverse sources like GPTeacher, Wizard LM, and Nous Research Instruct Dataset.
• Context Length: Supports a 4096-token sequence length.
• Response Quality: Noted for generating long responses and a lower hallucination rate compared to some alternatives.
• Multilingual Support: Enhanced with COIG-PC-LITE for improved Chinese language proficiency.
• Censorship: Absence of OpenAI-style censorship mechanisms.

Performance Highlights

Based on the underlying Nous-Hermes-Llama2-13b, the model shows competitive performance on various benchmarks:

• GPT4All Benchmark Average: Achieved 70.0.
• BigBench Average: Scored 0.3657.
• AGIEval Average: Reached 0.372.
• Open LLM Leaderboard: Achieves an average score of 53.31, with specific scores like ARC (25-shot) at 59.47 and HellaSwag (10-shot) at 82.28.

Recommended Use Cases

This model is suitable for a wide range of language tasks, particularly those requiring detailed instruction following and generation of extended responses. Its added Chinese capability makes it valuable for applications needing bilingual English-Chinese interaction. It can be used for creative text generation, complex instruction processing, and chatbot implementations, especially where a lower hallucination rate and uncensored output are desired.