deberta-v3-large-survey-related_passage_old_facts-rater-gpt4

This model is a fine-tuned version of microsoft/deberta-v3-large on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4590
• Krippendorff: 0.9496
• Spearman: 0.9595
• Absolute Agreement: 0.8618
• Agreement Within One: 0.9422

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: 6e-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: 1000
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Krippendorff	Spearman	Absolute Agreement	Agreement Within One
No log	1.0	50	1.9214	-0.1968	-0.1743	0.25	0.6389
No log	2.0	100	1.9230	-0.2449	-0.2660	0.2222	0.625
No log	3.0	150	1.9305	-0.3132	-0.3197	0.2639	0.5972
No log	4.0	200	1.9688	0.0758	0.1922	0.1944	0.875
No log	5.0	250	1.9912	-0.4897	0.1914	0.125	1.0
No log	6.0	300	1.9696	-0.3208	0.0287	0.125	0.9444
No log	7.0	350	1.9609	-0.4675	0.1913	0.125	1.0
No log	8.0	400	1.9717	-0.1927	-0.0683	0.125	0.875
No log	9.0	450	2.0918	-0.3136	-0.1631	0.0972	0.875
1.7449	10.0	500	2.0623	0.0154	0.0896	0.1944	0.8333
1.7449	11.0	550	2.0821	0.0052	0.0822	0.1944	0.8333
1.7449	12.0	600	2.0046	0.2459	0.2394	0.2917	0.7778
1.7449	13.0	650	2.1779	0.1727	0.1624	0.25	0.8333
1.7449	14.0	700	2.2878	0.2452	0.1854	0.2917	0.7917
1.7449	15.0	750	2.3372	0.2134	0.1616	0.2639	0.7917
1.7449	16.0	800	2.5162	0.2219	0.1882	0.2361	0.8194
1.7449	17.0	850	2.7111	0.2041	0.1544	0.2222	0.8194
1.7449	18.0	900	2.7977	0.2679	0.2161	0.2639	0.7361
1.7449	19.0	950	2.7704	0.2809	0.2031	0.2222	0.7917
0.9017	20.0	1000	3.2462	0.2204	0.1785	0.2361	0.7361

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1
• Datasets 2.10.1
• Tokenizers 0.12.1