|
--- |
|
license: other |
|
language: |
|
- en |
|
inference: false |
|
widget: |
|
- text: "How do I download this model?" |
|
example_title: "Text Gen Example" |
|
--- |
|
|
|
# OPT-19M-ChatSalad |
|
|
|
This is an experimental OPT-based model with 19 million parameters trained entirely **from scratch** as a datasetting practice. |
|
Thus, it should not be subject to the usual OPT license. You are free to use this model for any purpose. |
|
The model is small enough (under 40mb) that it should run at very fast speeds even entirely on CPU. |
|
|
|
It is recommend to use this model with the KoboldAI software, with the following parameters: |
|
|
|
- Temperature: 0.9 |
|
- Repetition Penalty: 1.1 |
|
- Repetition Penalty Slope: 0.7 |
|
- Repetition Penalty Range: 1024 |
|
- Top-P Sampling: 0.9 |
|
- Disable all other samplers |
|
|
|
All feedback and comments can be directed to Concedo on the KoboldAI discord. |
|
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) |
|
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_concedo__OPT-19M-ChatSalad) |
|
|
|
| Metric | Value | |
|
|-----------------------|---------------------------| |
|
| Avg. | 24.86 | |
|
| ARC (25-shot) | 24.4 | |
|
| HellaSwag (10-shot) | 25.15 | |
|
| MMLU (5-shot) | 23.12 | |
|
| TruthfulQA (0-shot) | 51.36 | |
|
| Winogrande (5-shot) | 49.72 | |
|
| GSM8K (5-shot) | 0.0 | |
|
| DROP (3-shot) | 0.25 | |
|
|
