maywell/EXAONE-3.0-7.8B-Instruct-Llamafied
The maywell/EXAONE-3.0-7.8B-Instruct-Llamafied model is a 7.8 billion parameter instruction-tuned causal language model, originally EXAONE 3.0, that has been converted to a Llama-compatible architecture. This conversion allows it to leverage the broader ecosystem and tooling available for Llama models. It maintains the original EXAONE 3.0's capabilities while offering enhanced interoperability for developers. This model is primarily designed for general instruction-following tasks within a Llama-based environment.
Loading preview...
Overview
This model, maywell/EXAONE-3.0-7.8B-Instruct-Llamafied, is a 7.8 billion parameter instruction-tuned language model. It is a Llama-compatible version of the original LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct model, developed by LG AI Research. The "Llamafied" conversion process adapts the EXAONE 3.0 architecture to align with the Llama model structure, enhancing its compatibility with the Llama ecosystem.
Key Characteristics
- Architecture Conversion: The model's original EXAONE 3.0 architecture has been systematically ported to a Llama-compatible format, including embedding, attention, MLP, and normalization layers.
- Instruction-Tuned: It retains the instruction-following capabilities of the base EXAONE 3.0-Instruct model.
- Interoperability: By adopting the Llama format, this model benefits from wider tool support and integration possibilities within the Llama-centric development environment.
- Context Length: Supports a context length of 8192 tokens.
Good For
- Developers seeking to utilize the EXAONE 3.0-Instruct model within a Llama-based workflow.
- General instruction-following tasks where Llama compatibility is a priority.
- Experimentation with a Llama-formatted version of a model from LG AI Research.