system-k/gemma-2-2b-estimator

estimation-reward-gemma-2-2b

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

• Loss: 93.8438
• Spearmanr: 0.6688
• Kendalltau: 0.4828
• Pearsonr: 0.0
• Rmse: 9.6873
• Mae: 7.4340

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine_with_min_lr
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Spearmanr	Kendalltau	Pearsonr	Rmse	Mae
142.0495	0.2094	500	172.4941	0.1755	0.1193	0.0	13.1337	10.1489
114.1114	0.4188	1000	145.3719	0.4083	0.2799	0.0	12.0570	9.3769
126.2303	0.6281	1500	123.7460	0.5220	0.3660	0.0	11.1241	8.5576
104.0802	0.8375	2000	110.0878	0.5964	0.4219	0.0	10.4923	8.0832
92.9514	1.0469	2500	101.4019	0.6340	0.4530	0.0	10.0698	7.7131
89.5989	1.2563	3000	98.1783	0.6485	0.4649	0.0	9.9085	7.6020
76.1914	1.4657	3500	96.0021	0.6582	0.4736	0.0	9.7981	7.5160
81.2849	1.6750	4000	95.4644	0.6645	0.4783	0.0	9.7706	7.5254
75.9316	1.8844	4500	93.8438	0.6688	0.4828	0.0	9.6873	7.4340

Framework versions

• PEFT 0.15.1
• Transformers 4.50.2
• Pytorch 2.5.1+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1