Reading a Model Card — Comprehension Exercises

Meridian-13B Model Card — Version 2.1 Intended uses: Meridian-13B is designed for developer-facing tasks including code completion, technical documentation drafting, and API reference summarisation. It performs well in structured output tasks and multi-step reasoning when given explicit instructions. Out-of-scope uses: This model is not intended for medical diagnosis, legal advice, or any high-stakes decision-making where errors could cause harm. It should not be deployed in autonomous systems without human oversight. Use in children's products is not recommended. Limitations: Meridian-13B may reproduce biases present in its training data, which consists of publicly available web text, curated code repositories, and licensed book corpora collected through 2024. The model has no knowledge of events after its training cut-off and may state outdated information confidently. Evaluation results: On MMLU (5-shot), the model achieves 74.1% accuracy. On HumanEval (0-shot), it achieves 61.8% pass@1. These results represent the base instruction-tuned checkpoint and may differ from fine-tuned variants. For deployment guidance, bias evaluations, and red-teaming results, see the full technical report linked in the repository.