Elron/deberta-v3-large-hate

deberta-v3-large-sentiment

This model is a fine-tuned version of microsoft/deberta-v3-large on an tweet_eval dataset.

Model description

Test set results:

Model	Emotion	Hate	Irony	Offensive	Sentiment
deberta-v3-large	86.3	61.3	87.1	86.4	73.9
BERTweet	79.3	-	82.1	79.5	73.4
RoB-RT	79.5	52.3	61.7	80.5	69.3

source:papers_with_code

Intended uses & limitations

Classifying attributes of interest on tweeter like data.

Training and evaluation data

tweet_eval dataset.

Training procedure

Fine tuned and evaluated with run_glue.py

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 7e-06
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.6362	0.18	100	0.5481	0.7197
0.4264	0.36	200	0.4550	0.8008
0.4174	0.53	300	0.4524	0.7868
0.4197	0.71	400	0.4586	0.7918
0.3819	0.89	500	0.4368	0.8078
0.3558	1.07	600	0.4525	0.8068
0.2982	1.24	700	0.4999	0.7928
0.2885	1.42	800	0.5129	0.8108
0.253	1.6	900	0.5873	0.8208
0.3354	1.78	1000	0.4244	0.8178
0.3083	1.95	1100	0.4853	0.8058
0.2301	2.13	1200	0.7209	0.8018
0.2167	2.31	1300	0.8090	0.7778
0.1863	2.49	1400	0.6812	0.8038
0.2181	2.66	1500	0.6958	0.8138
0.2159	2.84	1600	0.6315	0.8118
0.1828	3.02	1700	0.7173	0.8138
0.1287	3.2	1800	0.9081	0.8018
0.1711	3.37	1900	0.8858	0.8068
0.1598	3.55	2000	0.7878	0.8028
0.1467	3.73	2100	0.9003	0.7948
0.127	3.91	2200	0.9066	0.8048
0.1134	4.09	2300	0.9646	0.8118
0.1017	4.26	2400	0.9778	0.8048
0.085	4.44	2500	1.0529	0.8088
0.0996	4.62	2600	1.0082	0.8058
0.1054	4.8	2700	0.9698	0.8108
0.1375	4.97	2800	0.9334	0.8048
0.0487	5.15	2900	1.1273	0.8108
0.0611	5.33	3000	1.1528	0.8058
0.0668	5.51	3100	1.0148	0.8118
0.0582	5.68	3200	1.1333	0.8108
0.0869	5.86	3300	1.0607	0.8088
0.0623	6.04	3400	1.1880	0.8068
0.0317	6.22	3500	1.2836	0.8008
0.0546	6.39	3600	1.2148	0.8058
0.0486	6.57	3700	1.3348	0.8008
0.0332	6.75	3800	1.3734	0.8018
0.051	6.93	3900	1.2966	0.7978
0.0217	7.1	4000	1.3853	0.8048
0.0109	7.28	4100	1.4803	0.8068
0.0345	7.46	4200	1.4906	0.7998
0.0365	7.64	4300	1.4347	0.8028
0.0265	7.82	4400	1.3977	0.8128
0.0257	7.99	4500	1.3705	0.8108
0.0036	8.17	4600	1.4353	0.8168
0.0269	8.35	4700	1.4826	0.8068
0.0231	8.53	4800	1.4811	0.8118
0.0204	8.7	4900	1.5245	0.8028
0.0263	8.88	5000	1.5123	0.8018
0.0138	9.06	5100	1.5113	0.8028
0.0089	9.24	5200	1.5846	0.7978
0.029	9.41	5300	1.5362	0.8008
0.0058	9.59	5400	1.5759	0.8018
0.0084	9.77	5500	1.5679	0.8018
0.0065	9.95	5600	1.5683	0.8028

Framework versions

• Transformers 4.20.0.dev0
• Pytorch 1.9.0
• Datasets 2.2.2
• Tokenizers 0.11.6