twitter-roberta-base-sentiment-latest_v3_scratch

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base-sentiment-latest on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9484
• Accuracy: 0.8522
• Precision Macro: 0.7150
• Recall Macro: 0.6816
• F1 Macro: 0.6946
• F1 Weighted: 0.8484

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: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 40
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision Macro	Recall Macro	F1 Macro	F1 Weighted
No log	1.0	45	0.7207	0.6728	0.4697	0.4605	0.4446	0.6413
0.8468	2.0	90	0.7087	0.6721	0.5134	0.4813	0.4506	0.6414
0.7104	3.0	135	0.5202	0.7979	0.5365	0.5604	0.5445	0.7796
0.6128	4.0	180	0.5060	0.8105	0.7116	0.6006	0.6166	0.7984
0.505	5.0	225	0.4700	0.8212	0.7406	0.6029	0.6075	0.8079
0.446	6.0	270	0.5124	0.8143	0.5551	0.5735	0.5559	0.7957
0.4095	7.0	315	0.4568	0.8408	0.6926	0.6578	0.6700	0.8360
0.413	8.0	360	0.4471	0.8484	0.5658	0.5917	0.5783	0.8282
0.3451	9.0	405	0.4247	0.8534	0.7131	0.6305	0.6403	0.8419
0.3124	10.0	450	0.4419	0.8591	0.7675	0.6622	0.6852	0.8508
0.3124	11.0	495	0.4351	0.8566	0.7533	0.6746	0.6953	0.8502
0.2778	12.0	540	0.4418	0.8579	0.7673	0.6778	0.7026	0.8514
0.2572	13.0	585	0.5176	0.8497	0.7386	0.6759	0.6962	0.8443
0.2252	14.0	630	0.4637	0.8503	0.7174	0.6935	0.7034	0.8479
0.2127	15.0	675	0.5122	0.8509	0.7916	0.6596	0.6877	0.8424
0.1798	16.0	720	0.5424	0.8433	0.7153	0.6577	0.6757	0.8369
0.1483	17.0	765	0.5590	0.8522	0.7237	0.6779	0.6940	0.8475
0.1414	18.0	810	0.6393	0.8610	0.7674	0.6679	0.6913	0.8532
0.1173	19.0	855	0.6078	0.8465	0.7128	0.6905	0.7000	0.8442
0.1019	20.0	900	0.6100	0.8547	0.7817	0.6473	0.6684	0.8444
0.1019	21.0	945	0.7196	0.8515	0.7398	0.6640	0.6854	0.8445
0.0913	22.0	990	0.7781	0.8332	0.6793	0.6502	0.6608	0.8285
0.0764	23.0	1035	0.7152	0.8522	0.7098	0.6866	0.6960	0.8494
0.0656	24.0	1080	0.7987	0.8446	0.6926	0.6917	0.6918	0.8444
0.0613	25.0	1125	0.7605	0.8572	0.7165	0.6988	0.7063	0.8551
0.0527	26.0	1170	0.7406	0.8572	0.7322	0.6900	0.7056	0.8532
0.0432	27.0	1215	0.8191	0.8572	0.7354	0.6893	0.7062	0.8530
0.038	28.0	1260	0.7626	0.8598	0.7403	0.6714	0.6908	0.8531
0.0354	29.0	1305	0.8406	0.8553	0.7084	0.6848	0.6942	0.8523
0.0298	30.0	1350	0.8857	0.8490	0.7100	0.6873	0.6969	0.8464
0.0298	31.0	1395	0.8856	0.8553	0.7516	0.6753	0.6981	0.8491
0.0243	32.0	1440	0.8640	0.8623	0.7457	0.7018	0.7184	0.8586
0.0256	33.0	1485	0.8976	0.8515	0.7116	0.6892	0.6987	0.8490
0.0189	34.0	1530	0.8932	0.8579	0.7306	0.6822	0.6988	0.8530
0.0172	35.0	1575	0.9222	0.8497	0.6978	0.6842	0.6902	0.8478
0.0146	36.0	1620	0.9238	0.8579	0.7263	0.6860	0.7009	0.8537
0.0148	37.0	1665	0.9424	0.8528	0.7068	0.6862	0.6950	0.8502
0.0129	38.0	1710	0.9515	0.8522	0.7128	0.6815	0.6938	0.8485
0.0106	39.0	1755	0.9490	0.8528	0.7155	0.6820	0.6950	0.8490
0.0136	40.0	1800	0.9484	0.8522	0.7150	0.6816	0.6946	0.8484

Framework versions

• Transformers 4.55.0
• Pytorch 2.7.0+cu126
• Datasets 4.0.0
• Tokenizers 0.21.4