gpt2-multilingual-20-zh-repair_3epochs_lr1e-4

This model is a fine-tuned version of CausalNLP/gpt2-hf_multilingual-20 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.5282

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.95) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
3.6036	0.0626	500	3.6583
3.6222	0.1252	1000	3.6372
3.5958	0.1879	1500	3.6347
3.6297	0.2505	2000	3.6452
3.6016	0.3131	2500	3.6591
3.6284	0.3757	3000	3.6530
3.5823	0.4384	3500	3.6493
3.646	0.5010	4000	3.6412
3.6329	0.5636	4500	3.6358
3.5469	0.6262	5000	3.6302
3.5987	0.6889	5500	3.6242
3.4773	0.7515	6000	3.6181
3.561	0.8141	6500	3.6142
3.5243	0.8767	7000	3.6092
3.5998	0.9393	7500	3.6058
3.6267	1.0019	8000	3.6007
3.5314	1.0645	8500	3.5958
3.5579	1.1271	9000	3.5915
3.541	1.1897	9500	3.5861
3.5823	1.2524	10000	3.5820
3.534	1.3150	10500	3.5767
3.4906	1.3776	11000	3.5718
3.5538	1.4402	11500	3.5673
3.527	1.5029	12000	3.5631
3.5119	1.5655	12500	3.5584
3.4633	1.6281	13000	3.5537
3.5098	1.6907	13500	3.5503
3.4336	1.7534	14000	3.5474
3.5241	1.8160	14500	3.5444
3.4846	1.8786	15000	3.5409
3.4802	1.9412	15500	3.5385
3.5026	2.0038	16000	3.5368
3.494	2.0664	16500	3.5351
3.4524	2.1290	17000	3.5341
3.4478	2.1916	17500	3.5329
3.4458	2.2543	18000	3.5317
3.4925	2.3169	18500	3.5305
3.4913	2.3795	19000	3.5298
3.4331	2.4421	19500	3.5293
3.5182	2.5047	20000	3.5290
3.446	2.5674	20500	3.5286
3.5018	2.6300	21000	3.5285
3.475	2.6926	21500	3.5283
3.4299	2.7552	22000	3.5282
3.4538	2.8179	22500	3.5282
3.4471	2.8805	23000	3.5282
3.4295	2.9431	23500	3.5282

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.0
• Datasets 4.4.1
• Tokenizers 0.22.1