gpt-est-base
This is the base-size GPT2 model, trained from scratch on 2.2 billion words (Estonian National Corpus + News Crawl + Common Crawl) for 3 epochs. Previously named "gpt-4-est-base", renamed to avoid click-baiting.
Format
For training data was prepended with a text domain tag, and it should be added as prefix when using the model: >general<, >web<, >news<, >doaj< and >wiki< (standing for general texts, web crawled texts, news, article abstracts and wikipedia texts). Use the prefixes like this, e.g: ">web< Kas tead, et".
Model details
- num. of layers: 12
- num. of heads: 12
- embedding size: 768
- context size: 1024
- total size: 118.68M params
Further details to be added soon.
Framework versions
- Transformers 4.13.0.dev0
- Pytorch 1.10.0+cu102
- Datasets 1.15.1
- Tokenizers 0.10.3
- Downloads last month
- 79
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.