results_1011

This model is a fine-tuned version of beomi/gemma-ko-2b on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5227

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: 0.0002
• train_batch_size: 3
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.0356	0.08	100	1.8434
1.7068	0.16	200	1.6624
1.6553	0.24	300	1.6477
1.6456	0.32	400	1.6441
1.6403	0.4	500	1.6378
1.6347	0.48	600	1.6315
1.6218	0.56	700	1.6256
1.6259	0.64	800	1.6192
1.6156	0.72	900	1.6116
1.6202	0.8	1000	1.6075
1.6031	0.88	1100	1.6058
1.6018	0.96	1200	1.6031
1.5965	1.04	1300	1.6022
1.5988	1.12	1400	1.6002
1.6043	1.2	1500	1.5978
1.5933	1.28	1600	1.5962
1.5909	1.3600	1700	1.5953
1.6014	1.44	1800	1.5932
1.584	1.52	1900	1.5912
1.5865	1.6	2000	1.5897
1.5871	1.6800	2100	1.5880
1.5838	1.76	2200	1.5865
1.5878	1.8400	2300	1.5850
1.58	1.92	2400	1.5835
1.5819	2.0	2500	1.5812
1.5652	2.08	2600	1.5806
1.573	2.16	2700	1.5796
1.5677	2.24	2800	1.5779
1.572	2.32	2900	1.5764
1.5688	2.4	3000	1.5748
1.5663	2.48	3100	1.5730
1.5669	2.56	3200	1.5719
1.5613	2.64	3300	1.5704
1.564	2.7200	3400	1.5690
1.5619	2.8	3500	1.5681
1.5622	2.88	3600	1.5667
1.5628	2.96	3700	1.5651
1.5514	3.04	3800	1.5645
1.5597	3.12	3900	1.5628
1.5499	3.2	4000	1.5622
1.5436	3.2800	4100	1.5610
1.5521	3.36	4200	1.5598
1.5389	3.44	4300	1.5585
1.5518	3.52	4400	1.5577
1.545	3.6	4500	1.5559
1.5383	3.68	4600	1.5552
1.5338	3.76	4700	1.5538
1.5452	3.84	4800	1.5522
1.5269	3.92	4900	1.5516
1.5342	4.0	5000	1.5507
1.5243	4.08	5100	1.5503
1.5209	4.16	5200	1.5498
1.5337	4.24	5300	1.5487
1.5261	4.32	5400	1.5477
1.5255	4.4	5500	1.5463
1.5342	4.48	5600	1.5459
1.5211	4.5600	5700	1.5447
1.5293	4.64	5800	1.5441
1.5203	4.72	5900	1.5425
1.5171	4.8	6000	1.5421
1.5239	4.88	6100	1.5412
1.5184	4.96	6200	1.5404
1.508	5.04	6300	1.5405
1.5113	5.12	6400	1.5396
1.5035	5.2	6500	1.5385
1.5082	5.28	6600	1.5380
1.5144	5.36	6700	1.5376
1.5052	5.44	6800	1.5367
1.5096	5.52	6900	1.5358
1.5139	5.6	7000	1.5348
1.5026	5.68	7100	1.5344
1.5061	5.76	7200	1.5339
1.5073	5.84	7300	1.5332
1.5082	5.92	7400	1.5323
1.5118	6.0	7500	1.5320
1.4939	6.08	7600	1.5323
1.4986	6.16	7700	1.5322
1.492	6.24	7800	1.5324
1.4889	6.32	7900	1.5309
1.4986	6.4	8000	1.5301
1.5003	6.48	8100	1.5297
1.5059	6.5600	8200	1.5295
1.4961	6.64	8300	1.5291
1.4938	6.72	8400	1.5279
1.5039	6.8	8500	1.5276
1.4892	6.88	8600	1.5272
1.5	6.96	8700	1.5268
1.4944	7.04	8800	1.5270
1.4941	7.12	8900	1.5265
1.4849	7.2	9000	1.5270
1.4924	7.28	9100	1.5261
1.4903	7.36	9200	1.5256
1.4909	7.44	9300	1.5254
1.4884	7.52	9400	1.5253
1.4874	7.6	9500	1.5253
1.4973	7.68	9600	1.5251
1.4835	7.76	9700	1.5247
1.4844	7.84	9800	1.5245
1.4845	7.92	9900	1.5242
1.4857	8.0	10000	1.5239
1.483	8.08	10100	1.5241
1.4875	8.16	10200	1.5238
1.488	8.24	10300	1.5238
1.4816	8.32	10400	1.5236
1.4887	8.4	10500	1.5233
1.4785	8.48	10600	1.5236
1.4802	8.56	10700	1.5232
1.4846	8.64	10800	1.5231
1.4832	8.72	10900	1.5231
1.4821	8.8	11000	1.5229
1.4837	8.88	11100	1.5230
1.4865	8.96	11200	1.5229
1.4855	9.04	11300	1.5228
1.4841	9.12	11400	1.5229
1.4765	9.2	11500	1.5230
1.4795	9.28	11600	1.5228
1.4848	9.36	11700	1.5228
1.4827	9.44	11800	1.5229
1.4883	9.52	11900	1.5228
1.4796	9.6	12000	1.5228
1.4899	9.68	12100	1.5228
1.4852	9.76	12200	1.5227
1.48	9.84	12300	1.5227
1.4823	9.92	12400	1.5227
1.4796	10.0	12500	1.5227

Framework versions

• PEFT 0.12.0
• Transformers 4.45.0
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.20.1