The conversational AI chatbot app using pre-trained OpenAI GPT-2 model for beginners who want to learn English. I mainly used Hugging Face's training code which used transfer Learning from an OpenAI GPT and GPT-2 Transformer language model.
You can see the result of Similarity and Correct. Similarity means whether you speak well according to the situation. Correct means how much you talk with grammar.
AFL can check Similarity and Correct.
You can easily review the textbook by solving it.
- Open AI GPT
- Open AI GPT2
AFL stands for Assessment For Learning. This word to refer to a way of evaluating users on an achievement basis, away from traditional learning evaluation methods.
Therefore, the project aimed to score user evaluations for continuous learning and motivation using MRPC, CoLA dataset, and Spell Check API.
- MRPC (Microsoft Research Paraphrase Corpus)
- CoLA (Corpus of Linguistic Acceptability)
- Bing Spell Check API
- AI Hub Korean-English translation corpus was used for fine tuning. [ link ]
- Plus, We add the situation data made by English text book.
| Argument | Default value | Modified Value | Description |
|---|---|---|---|
| Model | Open AI GPT | GPT2 | Open AI GPT, GPT2 |
| Num_candidates | 2 | 6 | candidate group for Next Utterance |
| Max_history | 4 | 2 | Number of previous utterances to keep in history |
| Gradient_accumulation_steps | 8 | 4 | Used to troubleshoot memory problems on GPU during Optimization |
| Epochs | 1 | 30 | Number of Epochs |
| Train_batch_size | 4 | 2 | Batch size for training |
| Valid_batch_size | 4 | 2 | Batch size for validation |
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent