altomek/YiSM-34B-0rn
altomek/YiSM-34B-0rn is a 34 billion parameter language model based on the Yi-1.5 architecture, created by altomek through a self-merge process. It is designed to follow instructions effectively while retaining the core characteristics of its base models. This neutral version balances accessibility with openness, making it suitable for exploration and intellectual exchange, and achieves an average score of 75.65 on the Open LLM Leaderboard.
Loading preview...
YiSM-34B-0rn: A Self-Merged Instruction-Following Model
YiSM-34B-0rn is a 34 billion parameter language model developed by altomek, created by merging the Yi-1.5-34B-Chat and Yi-1.5-34B models. The primary goal of this self-merge was to produce a model that excels at following instructions while preserving the inherent nature of its foundational Yi-1.5 architecture.
Key Capabilities and Features
- Instruction Following: Engineered to respond accurately to a wide range of instructions.
- Neutral Stance: This specific version is categorized as "Neutral," aiming to balance accessibility with openness, encouraging broad exploration and intellectual exchange without explicit censorship or unfiltered access.
- Performance: Achieves a competitive average score of 75.65 on the Open LLM Leaderboard (as of 2024-06-10), placing it among the top models in its size class. Notable scores include 78.51 on MMLU and 86.67 on HellaSwag.
- Context Length: Utilizes a
max_seq_lenof 8K with analpha_valueof 2.65, indicating support for extended context processing. - Quantization Options: Available in various quantization formats, including GGUF and multiple EXL2
bpwoptions, to accommodate different hardware constraints.
Ideal Use Cases
- Research and Development: Suitable for researchers and developers seeking a capable 34B model for experimentation and intellectual inquiry.
- General Purpose Applications: Can be employed in scenarios requiring robust instruction following and balanced content generation.
- Educational Environments: Its neutral characteristic makes it appropriate for academic settings where balanced information and exploration are valued.