GR model fine-tuning on the AXOLOTL-24 training sets

deniskokosss / deepchange_at_axolotl

Apache License 2.0

0 stars 0 forks source link

GR model fine-tuning on the AXOLOTL-24 training sets #5

Open nvanva opened 3 months ago

nvanva commented 3 months ago

Scripts to fine-tune GR FiEnRu, GR Fi SG from the original GR. Check that all results are reproduced with the freshly fine-tuned models (if some randomness is involved, it would be nice to estimate std of the results or at least to avoid it by fixing all possible seeds).

deniskokosss commented 2 months ago

Unable to reproduce the finetuning exactly. Published scripts, which reproduce within 2% ARI