xtremedistil-l6-h384-uncased-v2.0

This model is a fine-tuned version of microsoft/xtremedistil-l6-h384-uncased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5278
• F1 Macro: 0.6999
• F1 Micro: 0.7000
• Accuracy Balanced: 0.7017
• Accuracy: 0.7000
• Precision Macro: 0.7009
• Recall Macro: 0.7017
• Precision Micro: 0.7000
• Recall Micro: 0.7000

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 128
• seed: 40
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.06
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Micro	Accuracy Balanced	Accuracy	Precision Macro	Recall Macro	Precision Micro	Recall Micro
0.6275	0.17	200	0.6177	0.3647	0.5463	0.5039	0.5463	0.6163	0.5039	0.5463	0.5463
0.5811	0.34	400	0.5808	0.5807	0.6194	0.5976	0.6194	0.6331	0.5976	0.6194	0.6194
0.5769	0.51	600	0.5680	0.6564	0.6585	0.6703	0.6585	0.6796	0.6703	0.6585	0.6585
0.5647	0.68	800	0.5634	0.6703	0.6728	0.6855	0.6728	0.6976	0.6855	0.6728	0.6728
0.5607	0.85	1000	0.5720	0.6176	0.6448	0.6264	0.6448	0.6569	0.6264	0.6448	0.6448
0.5645	1.02	1200	0.5617	0.6523	0.6601	0.6521	0.6601	0.6581	0.6521	0.6601	0.6601
0.5665	1.19	1400	0.5479	0.6802	0.6840	0.6986	0.6840	0.7172	0.6986	0.6840	0.6840
0.5432	1.35	1600	0.5540	0.6642	0.6665	0.6644	0.6665	0.6641	0.6644	0.6665	0.6665
0.5427	1.52	1800	0.5520	0.6533	0.6617	0.6532	0.6617	0.6601	0.6532	0.6617	0.6617
0.5453	1.69	2000	0.5487	0.6756	0.6781	0.6755	0.6781	0.6757	0.6755	0.6781	0.6781
0.5528	1.86	2200	0.5492	0.6720	0.6771	0.6713	0.6771	0.6747	0.6713	0.6771	0.6771
0.531	2.03	2400	0.5476	0.6799	0.6803	0.6882	0.6803	0.6911	0.6882	0.6803	0.6803
0.5199	2.2	2600	0.5454	0.6823	0.6824	0.6863	0.6824	0.6856	0.6863	0.6824	0.6824
0.535	2.37	2800	0.5441	0.6797	0.6803	0.6817	0.6803	0.6804	0.6817	0.6803	0.6803
0.5246	2.54	3000	0.5453	0.6746	0.6750	0.6771	0.6750	0.6759	0.6771	0.6750	0.6750
0.5405	2.71	3200	0.5408	0.6824	0.6861	0.6819	0.6861	0.6836	0.6819	0.6861	0.6861
0.5414	2.88	3400	0.5404	0.6826	0.6834	0.6841	0.6834	0.6828	0.6841	0.6834	0.6834

eval result

Datasets	asadfgglie/nli-zh-tw-all/test	asadfgglie/BanBan_2024-10-17-facial_expressions-nli/test	eval_dataset	test_dataset
eval_loss	0.511	0.701	0.539	0.528
eval_f1_macro	0.717	0.488	0.684	0.7
eval_f1_micro	0.718	0.506	0.684	0.7
eval_accuracy_balanced	0.719	0.501	0.687	0.702
eval_accuracy	0.718	0.506	0.684	0.7
eval_precision_macro	0.718	0.501	0.686	0.701
eval_recall_macro	0.719	0.501	0.687	0.702
eval_precision_micro	0.718	0.506	0.684	0.7
eval_recall_micro	0.718	0.506	0.684	0.7
eval_runtime	8.868	0.186	1.876	7.42
eval_samples_per_second	958.472	5096.766	1006.985	1018.427
eval_steps_per_second	7.555	43.102	7.996	8.086
Size of dataset	8500	946	1889	7557

Framework versions

• Transformers 4.33.3
• Pytorch 2.5.1+cu121
• Datasets 2.14.7
• Tokenizers 0.13.3