componavt/xlm-roberta-base-topic-classification-2025

XLM-RoBERTa Base for Topic Classification

Model fine-tuned for Russian social media topic classification.

• Trained on: 512_posts_24_topics.csv
• Epochs: 17
• Learning rate: 2e-05

Epoch	Training Loss	Validation Loss	Accuracy	F1 Macro	F1 Weighted
1	No log	3.026107	0.107527	0.018696	0.033548
2	No log	2.870733	0.182796	0.052066	0.116982
3	No log	2.696235	0.225806	0.062794	0.135901
4	No log	2.468899	0.301075	0.117249	0.195119
5	No log	2.275375	0.365591	0.164985	0.264671
6	No log	2.103708	0.430108	0.246894	0.348294
7	No log	1.953564	0.494624	0.297129	0.416058
8	No log	1.905395	0.537634	0.356591	0.465515
9	No log	1.783440	0.516129	0.349648	0.457348
10	No log	1.733219	0.559140	0.385719	0.496481
11	No log	1.668128	0.537634	0.361045	0.470044
12	No log	1.633103	0.559140	0.387321	0.498477
13	No log	1.588825	0.569892	0.392370	0.508810
14	No log	1.581068	0.591398	0.410717	0.533581
15	No log	1.560641	0.591398	0.413310	0.533612
16	No log	1.545080	0.580645	0.403768	0.525466
17	No log	1.550952	0.591398	0.410717	0.533581