Cross-lingual Transfer of Reward Models
Collection
This is the collection of synthetic preference data and trained reward models in "Cross-lingual Transfer of Reward Models in Multilingual Alignment". • 5 items • Updated
Qwen2.5-3B-MP-RM
This model is a fine-tuned version of Qwen/Qwen2.5-3B-Instruct on the iqwiki-kor/MP-86k dataset.
RewardBench Evaluation
| Model | Chat | Chat-Hard | Safety | Reasoning | Avg. |
|---|---|---|---|---|---|
| iqwiki-kor/Qwen2.5-3B-MP-RM | 89.1 | 75.2 | 87.3 | 95.4 | 86.8 |
| RLHFlow/ArmoRM-Llama3-8B-v0.1 | 96.9 | 76.8 | 90.5 | 97.3 | 90.4 |
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1
- Downloads last month
- 5