|
--- |
|
license: apache-2.0 |
|
language: |
|
- en |
|
--- |
|
|
|
Pre-trained model fine-tuned using Reinforcement Learning on [DIALOCONAN](https://github.com/marcoguerini/CONAN#dialoconan) dataset using [facebook/roberta-hate-speech-dynabench-r4-target](https://huggingface.co/facebook/roberta-hate-speech-dynabench-r4-target) as reward model. |
|
|
|
Toxicity results on [allenai/real-toxicity-prompts](https://huggingface.co/datasets/allenai/real-toxicity-prompts) dataset using custom prompts (see 🥞[RewardLM](https://github.com/DanielSc4/RewardLM) for details). |
|
|
|
| Toxicity Level | RedPajama-INCITE-Chat-3B | |
|
|:--------------:|:------------------------:| |
|
| Pre-Trained | 0.217 | |
|
| **Fine-Tuned** | **0.129** | |
|
| [RL](https://huggingface.co/DanielSc4/RedPajama-INCITE-Chat-3B-v1-RL-LoRA-8bit-test1) | 0.160 | |
|
|
|
# [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_DanielSc4__RedPajama-INCITE-Chat-3B-v1-FT-LoRA-8bit-test1) |
|
|
|
| Metric | Value | |
|
|-----------------------|---------------------------| |
|
| Avg. | 33.13 | |
|
| ARC (25-shot) | 38.65 | |
|
| HellaSwag (10-shot) | 63.53 | |
|
| MMLU (5-shot) | 25.16 | |
|
| TruthfulQA (0-shot) | 36.07 | |
|
| Winogrande (5-shot) | 60.14 | |
|
| GSM8K (5-shot) | 0.08 | |
|
| DROP (3-shot) | 8.24 | |
|
|