Model Overview
NeuralNovel/Mistral-7B-Instruct-v0.2-Neural-Story is a 7 billion parameter language model developed by NeuralNovel, fine-tuned from the Mistral-7B-Instruct-v0.2 base model. Its primary focus is on generating instructive and narrative text, with a specific optimization for storytelling and creative writing.
Key Capabilities
- Narrative Generation: Tailored to produce detailed and creative responses for narrative prompts.
- Storytelling: Optimized for generating short stories and engaging in creative writing tasks.
- Instruction Following: Retains the instruction-following capabilities of its base model, applied to narrative contexts.
Performance Benchmarks
Evaluated on the HuggingFaceH4/open_llm_leaderboard, the model achieved an average score of 64.96. Notable scores include:
- HellaSwag: 66.89
- Winogrande: 75.85
- MMLU: 60.67
Training Details
The model was fine-tuned using the Neural-Story-v1 dataset over 3 epochs, with a batch size of 12 and a learning rate of 1e-5. This training regimen aimed to enhance its narrative generation abilities.
Licensing
Based on Mistral AI's architecture, this model is released under the Apache-2.0 license, making it suitable for both commercial and non-commercial use.
Limitations
This model may not perform optimally in scenarios unrelated to instructive and narrative text generation. Users should be aware of potential biases or limitations inherited from the training data, particularly genre or writing style biases from the Neural-Story-v0.1 dataset.