Submission ruRoberta-large finetune

Aug. 18, 2021, 11:49 a.m.

Team: SberDevices

Model url: https://huggingface.co/sberbank-ai/ruRoberta-large


Total score: 0.684

Dataset Score Metric
LiDiRus 0.343 Matthew`s Corr
RCB 0.357 / 0.518 F1/Acc
PARus 0.722 Accuracy
MuSeRC 0.861 / 0.63 F1a/Em
TERRa 0.801 Accuracy
RUSSE 0.748 Accuracy
RWSD 0.669 Accuracy
DaNetQA 0.82 Accuracy
RuCoS 0.87 / 0.867 F1/EM
Model description:

Russian RoBERTa large model. 355ml parameters, trained in 250 GB Russian texts.


Parameter description:

Diagnostic (Matthew`s Correlation): 0.343

Category Score
LOGIC 0.1820012297621496
KNOWLEDGE 0.2698822742714764
PREDICATE-ARGUMENT STRUCTURE 0.34132316093258036
LEXICAL SEMANTICS 0.489878955484335
Lexical Semantics - Lexical Entailment 0.4238498361791447
Lexical Semantics - Morphological Negation 0.6366624736381951
Lexical Semantics - Factivity 0.289805474447035
Lexical Semantics - Symmetry/Collectivity 0.5477225575051661
Lexical Semantics - Redundancy 0.7985494095046906
Lexical Semantics - Named Entities 0.5427204202399745
Lexical Semantics - Quantifiers 0.35043832202523123
Predicate-Argument Structure Core Args 0.5111111111111111
Predicate-Argument Structure Prepositional Phrases 0.3972493441698867
Predicate-Argument Structure Ellipsis/Implicits 0.24433144650348906
Predicate-Argument Structure Anaphora/Coreference 0.07955908541183225
Predicate-Argument Structure Active/Passive 0.44353708548308607
Predicate-Argument Structure Nominalization 0.6024640760767093
Predicate-Argument Structure Genitives/Partitives 0.6875
Predicate-Argument Structure Datives 0.7637626158259734
Predicate-Argument Structure Relative Clauses 0.29277002188455997
Predicate-Argument Structure Coordination Scopes 0.2516299559794226
Predicate-Argument Structure Intersectivity 0.20679318146353196
Predicate-Argument Structure Restrictivity 0.09335200560186732
Logic Negation 0.11166573401651324
Logic Double Negation 0.4271410714851651
Logic Interval/Numbers -0.12121248708414586
Logic Conjuction 0.24575816200113812
Logic Disjunction -0.09647574015265818
Logic Conditionals 0.02486823656509969
Logic Universal 0.5606119105813882
Logic Existential 0.42365927286816174
Logic Temporal -0.030303030303030304
Logic Upward Monotone 0.3612343522752406
Logic Downward Monotone 0.06726727939963124
Logic Non-Monotonic 0.2163856337073152
Knowledge Common Sense 0.21361974637528325
Knowledge World Knowledge 0.3084094312371165

Performance:

Dataset Speed RAM
LiDiRus - -
RCB - -
PARus - -
MuSeRC - -
TERRa - -
RUSSE - -
RWSD - -
DaNetQA - -
RuCoS - -