maicomputer/alpaca-native
The maicomputer/alpaca-native is a 7 billion parameter language model, a replica of the Stanford Alpaca model, trained using the original instructions with minor FSDP modifications. This model is designed for general language understanding and generation tasks, offering a foundational capability for various NLP applications. It provides a accessible option for developers seeking an open-source Alpaca-based model.
Loading preview...
maicomputer/alpaca-native: A Stanford Alpaca Replica
This model is a 7 billion parameter replica of the original Stanford Alpaca, developed by maicomputer. It was trained using the original instruction-following dataset, with a minor modification in its FSDP (Fully Sharded Data Parallel) training mode. The model aims to provide an accessible implementation of the Alpaca architecture.
Key Capabilities & Performance
Evaluated on the Open LLM Leaderboard, maicomputer/alpaca-native demonstrates general language understanding capabilities. Its performance metrics include:
- Avg. Score: 41.96
- ARC (25-shot): 52.3
- HellaSwag (10-shot): 77.09
- MMLU (5-shot): 41.6
- TruthfulQA (0-shot): 37.58
- Winogrande (5-shot): 69.46
- GSM8K (5-shot): 1.44
- DROP (3-shot): 14.23
These scores indicate its proficiency across various reasoning, common sense, and knowledge-based tasks, though with specific areas like mathematical reasoning (GSM8K) showing lower performance.
Use Cases
This model is suitable for developers looking for an open-source, Alpaca-based model for:
- General text generation and completion.
- Instruction-following tasks.
- Experimentation with smaller, efficient language models.
- Applications requiring a balance of performance and computational resources.