Hacker News
- Talking to myself: how I trained GPT2-1.5b for rubber ducking using my chat data http://www.svilentodorov.xyz/blog/gpt-15b-chat-finetune/ 67 comments
- How is search so bad? A case study https://svilentodorov.xyz/blog/bad-search/ 396 comments
- Adding layers to the middle of trained network without invalidating the weights https://svilentodorov.xyz/blog/add-layers 8 comments