search

salesforce / CodeT5

Home of CodeT5: Open Code LLMs for Code Understanding and Generation
https://arxiv.org/abs/2305.07922
BSD 3-Clause "New" or "Revised" License
2.65k stars 391 forks source link

Finetuning concode task using finetuned checkpoints #161

Open SBhat2615 opened 6 months ago

SBhat2615 commented 6 months ago

I'm trying to finetune the concode task using 'code' as both input & output, instead of 'nl' & 'code'. I wanted to know if we can directly use the concode finetuned checkpoints of concode task and some more information about using tokenizers and embeddings?

Also, where are all changes need to be done here to load concode model instead of the base codet5 model ??

parser.add_argument("--model_tag", type=str, default='codet5_base', choices=['roberta', 'codebert', 'bart_base', 'codet5_small', 'codet5_base', 'codet5_large']).

Thanks!!

IP1102 commented 5 months ago

Hey @SBhat2615, did you figure it out?