Model Overview
Neelectric/Llama-3.1-8B-Instruct_SFT_Math-220kv00.34 is an 8 billion parameter instruction-tuned model developed by Neelectric. It is a specialized variant of the meta-llama/Llama-3.1-8B-Instruct base model, fine-tuned using Supervised Fine-Tuning (SFT) on a dedicated mathematical dataset.
Key Capabilities
- Enhanced Mathematical Reasoning: The model has undergone specific training on the
Neelectric/Replay_0.03.OpenR1-Math-220k_extended.wildguardmix.Llama3_4096toks dataset, significantly improving its performance on mathematical tasks. - Instruction Following: Retains the strong instruction-following capabilities of the Llama-3.1-8B-Instruct base model.
- Context Length: Supports a substantial context window of 32768 tokens, beneficial for complex multi-step problems or detailed mathematical explanations.
Training Details
This model was fine-tuned using the TRL library, a framework for Transformer Reinforcement Learning, indicating a focus on refining its responses through supervised learning techniques. The training leveraged specific versions of TRL, Transformers, PyTorch, Datasets, and Tokenizers, ensuring a consistent and reproducible training environment.
Good For
- Applications requiring accurate mathematical problem-solving.
- Generating explanations for mathematical concepts.
- Educational tools focused on math and logic.
- Tasks where robust instruction following combined with mathematical proficiency is crucial.