metadata
license: mit
tags:
- alignment-handbook
- dpo
- trl
- selm
base_model: ZhangShenao/SELM-Llama-3-8B-Instruct-iter-2
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: SELM-Llama-3-8B-Instruct-iter-3
results: []
Self-Exploring Language Models: Active Preference Elicitation for Online Alignment.
SELM-Llama-3-8B-Instruct-iter-3
This model is a fine-tuned version of ZhangShenao/SELM-Llama-3-8B-Instruct-iter-2 using synthetic data based on on the HuggingFaceH4/ultrafeedback_binarized dataset.
Model description
- Model type: A 8B parameter Llama3-instruct-based Self-Exploring Language Models (SELM).
- License: MIT
Results
| AlpacaEval 2.0 (LC WR) | MT-Bench (Average) | |
|---|---|---|
| SELM-Llama-3-8B-Instruct-iter-3 | 33.47 | 8.29 |
| SELM-Llama-3-8B-Instruct-iter-2 | 35.65 | 8.09 |
| SELM-Llama-3-8B-Instruct-iter-1 | 32.02 | 7.92 |
| Meta-Llama-3-8B-Instruct | 24.31 | 7.93 |
Our model also ranks highly on WildBench! 🔥
Training hyperparameters
The following hyperparameters were used during training:
- alpha: 0.0001
- beta: 0.01
- train_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- num_epochs: 1
Framework versions
- Transformers 4.40.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.19.1
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 23.56 |
| IFEval (0-Shot) | 69.03 |
| BBH (3-Shot) | 29.08 |
| MATH Lvl 5 (4-Shot) | 5.74 |
| GPQA (0-shot) | 1.12 |
| MuSR (0-shot) | 5.50 |
| MMLU-PRO (5-shot) | 30.92 |