RusEnQA
QA for Russian and English based on the rugpt3xl model
Fine-tuning format:
"<s>paragraph: "+eng_context+"\nlang: rus\nquestion: "+rus_question+' answer: '+ rus_answer+"</s>"
About ruGPT-3 XL model
Model was trained with 512 sequence length using Deepspeed and Megatron code by SberDevices team, on 80B tokens dataset for 4 epochs. After that model was finetuned 1 epoch with sequence length 2048.
Note! Model has sparse attention blocks.
Total training time was around 10 days on 256 GPUs. Final perplexity on test set is 12.05. Model parameters: 1.3B.