gemma_2b_social_reasoning_reward_model

This model is a fine-tuned version of google/gemma-2b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6131
• Accuracy: 0.6713

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.7962	0.24	10	0.6976	0.6077
0.6657	0.48	20	0.6464	0.6217
0.6334	0.72	30	0.6330	0.6392
0.6434	0.96	40	0.6218	0.6427
0.5745	1.2	50	0.6119	0.6567
0.5813	1.44	60	0.6282	0.6445
0.5822	1.68	70	0.6350	0.6585
0.5914	1.92	80	0.6294	0.6550
0.5386	2.16	90	0.6341	0.6550
0.5142	2.4	100	0.6330	0.6567
0.4906	2.63	110	0.6313	0.6620
0.5079	2.87	120	0.6309	0.6637

Framework versions

• PEFT 0.9.0
• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2