estimation-prometheus-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: 2.4192
• Accuracy: 0.4235
• Spearmanr: 0.4236
• Kendalltau: 0.3285
• Pearsonr: 0.4779
• Rmse: 1.0679
• Mae: 0.8129

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: 16
• 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: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Spearmanr	Kendalltau	Pearsonr	Rmse	Mae
4.38	0.2094	500	4.5787	0.3141	0.0241	0.0177	0.0263	1.3354	1.0377
3.4137	0.4188	1000	3.0643	0.3410	0.1413	0.1072	0.1555	1.2135	0.9624
2.8427	0.6281	1500	2.7700	0.3728	0.2953	0.2259	0.3155	1.1565	0.8857
2.7978	0.8375	2000	2.6360	0.3887	0.3524	0.2708	0.3844	1.1258	0.8747
2.5765	1.0469	2500	2.5750	0.4095	0.3797	0.2938	0.4191	1.1107	0.8387
2.5941	1.2563	3000	2.5392	0.4175	0.3839	0.2963	0.4286	1.1012	0.8465
2.3148	1.4657	3500	2.4901	0.4105	0.4069	0.3154	0.4536	1.0851	0.8245
2.6814	1.6750	4000	2.4642	0.4135	0.4100	0.3173	0.4635	1.0794	0.8221
2.5861	1.8844	4500	2.4569	0.4115	0.4152	0.3213	0.4668	1.0782	0.8185
2.5241	2.0938	5000	2.4320	0.4115	0.4196	0.3263	0.4733	1.0712	0.8132
2.5838	2.3032	5500	2.4252	0.4185	0.4197	0.3260	0.4755	1.0693	0.8125
2.4464	2.5126	6000	2.4232	0.4185	0.4217	0.3270	0.4768	1.0695	0.8110
2.4398	2.7219	6500	2.4218	0.4185	0.4221	0.3276	0.4764	1.0686	0.8122
2.3042	2.9313	7000	2.4192	0.4235	0.4236	0.3285	0.4779	1.0679	0.8129

Framework versions

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