akhisreelibra/bert-malayalam-pos-tagger

This model is a fine-tuned version of bert-base-multilingual-uncased on a set of tagged Malayalam sentence dataset It achieves the following results on the evaluation set:

• Loss: 0.4383
• Precision: 0.7380
• Recall: 0.7767
• F1: 0.7569
• Accuracy: 0.8552

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Tag abbreviation	Tag name
'CC_CCD'	Co-ordinator
'CC_CCS'	Subordinator
'CC_CCS_UT'	Quotative
'DM_DMD'	Deictic demonstrative
'DM_DMQ'	Wh-word
'DM_DMR'	Relative demonstrative
'JJ'	Adjective
'N_NN'	Common noun
'N_NNP'	Proper noun
'N_NST'	Locative noun
'PR_PRC'	Reciprocal pronoun
'PR_PRF'	Reflexive pronoun
'PR_PRL'	Relative pronoun
'PR_PRP'	Personal pronoun
'PR_PRQ'	Wh-word
'PSP'	Postposition
'QT_QTC'	Cardinals
'QT_QTF'	General quantifier
'QT_QTO'	Ordinals
'RB'	Adverb
'RD_ECH'	Echo words
'RD_RDF'	Foreign words
'RD_SYM'	Symbol
'RD_UNK'	Unknown
'RP_CL'	Classifier particle
'RP_INJ'	Interjection particle
'RP_INTF'	Intensifier particle
'RP_NEG'	Negation particle
'RP_RPD'	Default particle
'V_VAUX'	Auxiliary verb
'V_VM'	Main verb
'V_VM_VF'	Finite verb
'V_VM_VINF'	Infinite verb
'V_VM_VNF'	Non-finite verb
'V_VN'	Verbal noun

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.518	1.0	2692	0.4987	0.7124	0.7415	0.7267	0.8374
0.4415	2.0	5384	0.4515	0.7221	0.7679	0.7443	0.8481
0.3645	3.0	8076	0.4383	0.7380	0.7767	0.7569	0.8552

Framework versions

• Transformers 4.18.0
• Pytorch 1.11.0
• Datasets 2.1.0
• Tokenizers 0.12.1