lr0.0001_bs16_0620_0942

This model is a fine-tuned version of nvidia/mit-b0 on the greenkwd/upwellingdetection_SST dataset. It achieves the following results on the evaluation set:

• Loss: 0.1335
• Mean Iou: 0.8871
• Mean Accuracy: 0.9459
• Overall Accuracy: 0.9536
• Accuracy Land: 0.9552
• Accuracy Upwelling: 0.9692
• Accuracy Not Upwelling: 0.9133
• Iou Land: 0.9542
• Iou Upwelling: 0.9274
• Iou Not Upwelling: 0.7796
• Dice Macro: 0.9383
• Dice Micro: 0.9536

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: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Land	Accuracy Upwelling	Accuracy Not Upwelling	Iou Land	Iou Upwelling	Iou Not Upwelling	Dice Macro	Dice Micro
1.0882	0.4	20	1.0699	0.1883	0.4295	0.3022	0.0162	0.3817	0.8905	0.0149	0.3664	0.1835	0.2919	0.3022
0.9223	0.8	40	0.8895	0.5561	0.7282	0.7212	0.6576	0.7873	0.7397	0.6574	0.7021	0.3088	0.6967	0.7212
0.7699	1.2	60	0.6288	0.6063	0.7519	0.7768	0.7339	0.8893	0.6323	0.7339	0.7609	0.3240	0.7334	0.7768
0.69	1.6	80	0.4913	0.6720	0.8138	0.8249	0.7968	0.8865	0.7580	0.7968	0.7955	0.4238	0.7894	0.8249
0.6536	2.0	100	0.4191	0.6957	0.8285	0.8440	0.7989	0.9377	0.7489	0.7989	0.8361	0.4519	0.8072	0.8440
0.5298	2.4	120	0.3944	0.6962	0.8132	0.8531	0.8292	0.9750	0.6354	0.8292	0.8257	0.4337	0.8054	0.8531
0.4779	2.8	140	0.3525	0.7445	0.8604	0.8775	0.8585	0.9409	0.7818	0.8585	0.8477	0.5273	0.8440	0.8775
0.4727	3.2	160	0.3321	0.7514	0.8651	0.8818	0.8577	0.9509	0.7868	0.8577	0.8596	0.5370	0.8489	0.8818
0.5746	3.6	180	0.3068	0.7629	0.8791	0.8865	0.8587	0.9392	0.8395	0.8587	0.8685	0.5616	0.8576	0.8865
0.5181	4.0	200	0.2654	0.8091	0.8977	0.9163	0.9140	0.9619	0.8172	0.9138	0.8833	0.6302	0.8887	0.9163
0.4094	4.4	220	0.2525	0.8288	0.9177	0.9246	0.9247	0.9402	0.8882	0.9241	0.8895	0.6729	0.9022	0.9246
0.5539	4.8	240	0.2300	0.8317	0.9224	0.9254	0.9214	0.9374	0.9085	0.9209	0.8944	0.6799	0.9042	0.9254
0.4994	5.2	260	0.2150	0.8199	0.9171	0.9186	0.9011	0.9446	0.9055	0.9010	0.8998	0.6588	0.8965	0.9186
0.3206	5.6	280	0.2043	0.8570	0.9325	0.9391	0.9449	0.9469	0.9056	0.9435	0.9035	0.7240	0.9200	0.9391
0.3138	6.0	300	0.1909	0.8538	0.9301	0.9377	0.9408	0.9510	0.8986	0.9398	0.9041	0.7176	0.9181	0.9377
0.3412	6.4	320	0.1935	0.8630	0.9280	0.9435	0.9517	0.9680	0.8644	0.9498	0.9082	0.7311	0.9236	0.9435
0.3777	6.8	340	0.1728	0.8422	0.9188	0.9328	0.9245	0.9758	0.8560	0.9243	0.9106	0.6917	0.9105	0.9328
0.4217	7.2	360	0.1847	0.8545	0.9357	0.9370	0.9393	0.9373	0.9304	0.9386	0.9028	0.7221	0.9186	0.9370
0.33	7.6	380	0.1690	0.8596	0.9250	0.9420	0.9460	0.9758	0.8532	0.9450	0.9102	0.7234	0.9214	0.9420
0.4913	8.0	400	0.1574	0.8682	0.9323	0.9456	0.9511	0.9689	0.8770	0.9500	0.9133	0.7413	0.9268	0.9456
0.3707	8.4	420	0.1526	0.8627	0.9253	0.9437	0.9484	0.9798	0.8476	0.9474	0.9114	0.7295	0.9234	0.9437
0.4486	8.8	440	0.1451	0.8643	0.9323	0.9433	0.9415	0.9707	0.8847	0.9407	0.9169	0.7352	0.9245	0.9433
0.2992	9.2	460	0.1411	0.8752	0.9440	0.9475	0.9520	0.9497	0.9304	0.9508	0.9151	0.7597	0.9313	0.9475
0.3912	9.6	480	0.1465	0.8637	0.9308	0.9432	0.9388	0.9774	0.8763	0.9384	0.9201	0.7325	0.9241	0.9432
0.3323	10.0	500	0.1501	0.8854	0.9351	0.9544	0.9686	0.9803	0.8564	0.9652	0.9182	0.7729	0.9372	0.9544
0.3496	10.4	520	0.1311	0.8917	0.9470	0.9559	0.9621	0.9683	0.9105	0.9600	0.9263	0.7888	0.9411	0.9559
0.256	10.8	540	0.1320	0.8841	0.9463	0.9520	0.9521	0.9647	0.9221	0.9511	0.9263	0.7747	0.9366	0.9520
0.3223	11.2	560	0.1451	0.8734	0.9436	0.9465	0.9405	0.9608	0.9296	0.9401	0.9247	0.7554	0.9302	0.9465
0.4234	11.6	580	0.1335	0.8871	0.9459	0.9536	0.9552	0.9692	0.9133	0.9542	0.9274	0.7796	0.9383	0.9536

Framework versions

• Transformers 4.41.2
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1