2023-02-13 14:35:40 +01:00
|
|
|
# Projekt na przedmiot Uczenie głębokie w przetwarzaniu tekstu
|
|
|
|
|
|
|
|
|
|
## Skład grupy:
|
|
|
|
|
|
|
|
|
|
- Michał Kozłowski
|
|
|
|
|
- Szymon Jadczak
|
|
|
|
|
|
|
|
|
|
## Modele:
|
|
|
|
|
- RobertaForSequenceClassification
|
|
|
|
|
- GPT-2
|
|
|
|
|
- T5
|
|
|
|
|
- FLAN-T5
|
|
|
|
|
|
2023-02-13 14:46:19 +01:00
|
|
|
## Accuracy test split
|
|
|
|
|
- RobertaForSequenceClassification -> 0.9392201834862385
|
|
|
|
|
- GPT-2 ->
|
|
|
|
|
- T5 -> Accuracy: 0.9129464285714286
|
|
|
|
|
- FLAN-T5 ->
|
|
|
|
|
|
2023-02-13 14:35:40 +01:00
|
|
|
## Trenowanie
|
|
|
|
|
- Google Colab
|
|
|
|
|
|
2023-02-13 14:41:40 +01:00
|
|
|
## Link do modeli na google drive
|
2023-02-13 14:42:13 +01:00
|
|
|
https://drive.google.com/drive/folders/1GWNah7-LZI7jrFzUpL9Le1E7b73RutbU?usp=sharing
|
2023-02-13 14:41:40 +01:00
|
|
|
|
2023-02-13 14:47:32 +01:00
|
|
|
### Linki do poszczególnych modeli na huggingface oraz wykresów na tensorboard znajdują się w notatnikach
|
2023-02-13 14:35:40 +01:00
|
|
|
|
|
|
|
|
## Dataset:
|
|
|
|
|
- Orginalny dataset: https://huggingface.co/datasets/sst2
|
|
|
|
|
- Przetworzony dataset: https://huggingface.co/datasets/Zombely/sst2-project-dataset
|
