psyche/kollama2-7b-v2
The psyche/kollama2-7b-v2 is a 7 billion parameter language model with a 4096 token context length. This model is currently under construction, with the developer actively working to refine its decoding patterns that originated from the fine-tuning dataset. It is intended for general language tasks, with current evaluations indicating average performance across various benchmarks.
Loading preview...
psyche/kollama2-7b-v2: An Under-Construction 7B Model
This model, psyche/kollama2-7b-v2, is a 7 billion parameter language model with a 4096 token context length. It is currently in an active development phase, with the creator focusing on addressing and refining decoding patterns that emerged from its fine-tuning dataset. The goal is to improve its overall performance and consistency.
Current Evaluation Snapshot
Based on preliminary evaluations on the Open LLM Leaderboard, the model exhibits the following performance metrics:
- Avg. Score: 44.91
- ARC (25-shot): 53.33
- HellaSwag (10-shot): 78.5
- MMLU (5-shot): 43.61
- TruthfulQA (0-shot): 46.37
- Winogrande (5-shot): 75.61
- GSM8K (5-shot): 6.52
- DROP (3-shot): 10.4
Development Focus
The primary development effort is centered on mitigating and removing specific decoding patterns to enhance the model's reliability and output quality. Users should be aware that the model is not yet in its final, optimized state.