Meeting Note #3 14.03.2019

Location: Bogazici University Computer Engineering Building

Date/Time: 14.03.2019 / 12:00

Attendees:

• Suzan Üsküdarlı
• Onur Güngör
• Uğurcan Arıkan

1. Preparation Before Meeting

• 1.1. Search about sentencepiece
• 1.2. Search about pretraining BERT

2. Agenda

• 2.1. Pretraining BERT after the Turkish vocabulary has been created

3. Discussion

• 3.1. Memory issue during BERT's pretraining due to the corpus' size has been discussed
• 3.2. Creating pretraining data and pretraining BERT has been discussed
• 3.3. Current status of the project has been discussed

4. Outcomes

• 4.1. BERT pretraining
  • 4.1.1. Due to its massive size, corpus will be split into 10 smaller chunks before pretraining

5. TO-DO list

Deadline: 21.03 12:00 Assignee: Uğurcan Arıkan

• 5.1. Split corpus into smaller pieces

Deadline: 21.03 12:00 Assignee: Uğurcan Arıkan

• 5.2. Create pretraining data for BERT

Deadline: 21.03 12:00 Assignee: Uğurcan Arıkan

• 5.3. Run pretraining data on BERT