Abt Finetuning

#4
by CS437help - opened

Great work! Thanx a lot I really liked the model. Do you have any notebook to further fine-tuning it?
Thnx in advance

Thank you for your interest!

We plan to set up a GitHub repo to include all training and evaluating scripts.

I am looking forward to your work. I want to try it myself. If I use sft, is the data format like this:
{
"instruction": "Translate this from Chinese to Korean:\nChinese: ",
"input": "{value}\n",
"output": "Korean:"
},

The SFT data format is:
{
"instruction": "Translate this from Chinese to Korean:",
"input": "Chinese: {Chinese_sentence}\nKorean:",
"output": "{Korean_sentence}"
},

Thanks for sharing the model, the translation quality is incredible! I'm eager to finetune and test on my custom data. Is there an expected date when you release the Git repo and the training script? Thanks!

Thanks for sharing the model, the translation quality is incredible! I'm eager to finetune and test on my custom data. Is there an expected date when you release the Git repo and the training script? Thanks!

Thanks for your interest! The GitHub repo should be ready next week.

Thanks for your reply, looking forward to your work!

The SFT data format is:
{
"instruction": "Translate this from Chinese to Korean:",
"input": "Chinese: {Chinese_sentence}\nKorean:",
"output": "{Korean_sentence}"
},

Thanks for your reply, looking forward to your GitHub repo!!!!!!!!!!!

Thanks for the model I tried it too and it is really good. As others said it would be really good a github repo. I was trying to finetune the model with some instructions but in some cases it translate the instruction itself, i tried unsloth for finetuning but looks like there is an incompatiblity with the framework bcs of chat template ( the current chat template doesnt have add_generation_prompt). I m excited too abt the repo :).

Thanks for sharing the model, the translation quality is incredible! I'm eager to finetune and test on my custom data. Is there an expected date when you release the Git repo and the training script? Thanks!

https://github.com/xiaomi-research/gemmax

Thanks for the model I tried it too and it is really good. As others said it would be really good a github repo. I was trying to finetune the model with some instructions but in some cases it translate the instruction itself, i tried unsloth for finetuning but looks like there is an incompatiblity with the framework bcs of chat template ( the current chat template doesnt have add_generation_prompt). I m excited too abt the repo :).

https://github.com/xiaomi-research/gemmax

Sign up or log in to comment