BramVanroy/robbert-v2-dutch-base-hebban-reviews

robbert-v2-dutch-base-hebban-reviews

Dataset

• —dataset_name: BramVanroy/hebban-reviews
• —datasetconfig: filteredsentiment
• —dataset_revision: 2.0.0
• —labelcolumn: review_sentiment
• —textcolumn: reviewtextwithout_quotes

Training

• —optim: adamw_hf
• —learning_rate: 5e-05
• —perdevicetrainbatchsize: 64
• —perdeviceevalbatchsize: 64
• —gradientaccumulationsteps: 1
• —max_steps: 5001
• —save_steps: 500
• —metricforbest_model: qwk

Best checkedpoint based on validation

• —best_metric: 0.7412639349881154
• —bestmodelcheckpoint: trained/hebban-reviews/robbert-v2-dutch-base/checkpoint-3500

Test results of best checkpoint

• —accuracy: 0.8070512820512821
• —f1: 0.8144966061997005
• —precision: 0.8275999429062602
• —qwk: 0.7336245557372719
• —recall: 0.8070512820512821

Confusion matrix

[image]

Normalized confusion matrix

[image]

Environment

• —cuda_capabilities: 8.0; 8.0
• —cudadevicecount: 2
• —cuda_devices: NVIDIA A100-SXM4-80GB; NVIDIA A100-SXM4-80GB
• —finetuner_commit: 66294c815326c93682003119534cb72009f558c2
• —platform: Linux-4.18.0-305.49.1.el84.x8664-x86_64-with-glibc2.28
• —python_version: 3.9.5
• —toch_version: 1.10.0
• —transformers_version: 4.21.0