Neelectric/Llama-3.1-8B-Instruct_SDFT_sciencev00.01
Neelectric/Llama-3.1-8B-Instruct_SDFT_sciencev00.01 is an 8 billion parameter instruction-tuned language model developed by Neelectric, fine-tuned from Meta's Llama-3.1-8B-Instruct. This model specializes in scientific reasoning and knowledge, having been trained on the Neelectric/MoT_science_Llama3_4096toks_SDFT dataset. It utilizes the Self-Training with On-Policy Self-Distillation (SDFT) method to enhance its alignment and performance in scientific domains, making it suitable for science-related question answering and text generation tasks.
Loading preview...
Overview
Neelectric/Llama-3.1-8B-Instruct_SDFT_sciencev00.01 is an 8 billion parameter instruction-tuned language model, building upon Meta's Llama-3.1-8B-Instruct. Developed by Neelectric, this model is specifically fine-tuned for scientific applications, leveraging the Neelectric/MoT_science_Llama3_4096toks_SDFT dataset.
Key Capabilities
- Scientific Domain Specialization: Optimized for tasks requiring scientific knowledge and reasoning, derived from its specialized training dataset.
- SDFT Training Method: Incorporates Self-Training with On-Policy Self-Distillation (SDFT), a method designed to improve language model alignment, as detailed in the paper "Self-Training with On-Policy Self-Distillation for Language Model Alignment" (arXiv:2601.19897).
- Instruction Following: Inherits strong instruction-following capabilities from its Llama-3.1-8B-Instruct base.
Training Details
The model was trained using the TRL framework, with specific versions including TRL 1.0.0.dev0, Transformers 4.57.6, PyTorch 2.9.0, Datasets 4.8.4, and Tokenizers 0.22.2. The use of SDFT aims to enhance its performance in targeted scientific contexts.
Use Cases
This model is particularly well-suited for:
- Scientific question answering.
- Generating text related to scientific topics.
- Applications requiring a nuanced understanding of scientific concepts.