metadata

license: mit
base_model: ai-forever/ruElectra-medium
tags:
  - generated_from_trainer
model-index:
  - name: rubert-electra-srl
    results: []

rubert-electra-srl

This model is a fine-tuned version of ai-forever/ruElectra-medium on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.0448
Addressee Precision: 0.9583
Addressee Recall: 1.0
Addressee F1: 0.9787
Addressee Number: 23
Benefactive Precision: 0.0
Benefactive Recall: 0.0
Benefactive F1: 0.0
Benefactive Number: 2
Causator Precision: 0.9773
Causator Recall: 0.9773
Causator F1: 0.9773
Causator Number: 44
Cause Precision: 0.9259
Cause Recall: 0.7143
Cause F1: 0.8065
Cause Number: 35
Contrsubject Precision: 1.0
Contrsubject Recall: 0.9429
Contrsubject F1: 0.9706
Contrsubject Number: 35
Deliberative Precision: 0.9231
Deliberative Recall: 1.0
Deliberative F1: 0.9600
Deliberative Number: 24
Destinative Precision: 1.0
Destinative Recall: 1.0
Destinative F1: 1.0
Destinative Number: 7
Directivefinal Precision: 1.0
Directivefinal Recall: 1.0
Directivefinal F1: 1.0
Directivefinal Number: 1
Experiencer Precision: 0.9030
Experiencer Recall: 0.9441
Experiencer F1: 0.9231
Experiencer Number: 286
Instrument Precision: 0.9
Instrument Recall: 0.9
Instrument F1: 0.9
Instrument Number: 10
Object Precision: 0.9484
Object Recall: 0.9519
Object F1: 0.9502
Object Number: 541
Overall Precision: 0.9369
Overall Recall: 0.9425
Overall F1: 0.9397
Overall Accuracy: 0.9883
Limitative F1: 0.0
Limitative Number: 0.0
Limitative Precision: 0.0
Limitative Recall: 0.0
Directiveinitial F1: 0.0
Directiveinitial Number: 0.0
Directiveinitial Precision: 0.0
Directiveinitial Recall: 0.0
Mediative F1: 0.0
Mediative Number: 0.0
Mediative Precision: 0.0
Mediative Recall: 0.0

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: 0.00016666401556632117
train_batch_size: 1
eval_batch_size: 1
seed: 708526
gradient_accumulation_steps: 8
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.21
num_epochs: 3
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Addressee Precision	Addressee Recall	Addressee F1	Addressee Number	Benefactive Number	Causator Precision	Causator Recall	Causator F1	Causator Number	Cause Precision	Cause Recall	Cause F1	Cause Number	Contrsubject Precision	Contrsubject Recall	Contrsubject F1	Contrsubject Number	Deliberative Precision	Deliberative Recall	Deliberative F1	Deliberative Number	Destinative Precision	Destinative Recall	Destinative F1	Destinative Number	Directivefinal Precision	Directivefinal Recall	Directivefinal F1	Directivefinal Number	Experiencer Precision	Experiencer Recall	Experiencer F1	Experiencer Number	Instrument Precision	Instrument Recall	Instrument F1	Instrument Number	Object Precision	Object Recall	Object F1	Object Number	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
0.1548	1.0	1471	0.1755	0.6667	0.5217	0.5854	23	2	0.5714	0.8182	0.6729	44	0.5217	0.3429	0.4138	35	0.4103	0.4571	0.4324	35	0.0	0.0	0.0	24	0.0	0.0	0.0	7	0.0	0.0	0.0	1	0.8645	0.8252	0.8444	286	0.0	0.0	0.0	10	0.7711	0.8965	0.8291	541	0.7627	0.7907	0.7764	0.9582
0.1209	2.0	2942	0.0797	0.9130	0.9130	0.9130	23	2	0.9348	0.9773	0.9556	44	0.8462	0.6286	0.7213	35	0.8889	0.9143	0.9014	35	0.75	0.875	0.8077	24	1.0	0.4286	0.6	7	0.0	0.0	0.0	1	0.8993	0.8741	0.8865	286	0.875	0.7	0.7778	10	0.9336	0.9094	0.9213	541	0.9138	0.8839	0.8986	0.9808
0.0559	3.0	4413	0.0448	0.9583	1.0	0.9787	23	2	0.9773	0.9773	0.9773	44	0.9259	0.7143	0.8065	35	1.0	0.9429	0.9706	35	0.9231	1.0	0.9600	24	1.0	1.0	1.0	7	1.0	1.0	1.0	1	0.9030	0.9441	0.9231	286	0.9	0.9	0.9	10	0.9484	0.9519	0.9502	541	0.9369	0.9425	0.9397	0.9883

Framework versions

Transformers 4.42.4
Pytorch 2.3.1+cu121
Datasets 2.20.0
Tokenizers 0.19.1