Hey! If this is your first finetuning of a model you might want to follow a guide like this https://www.youtube.com/watch?v=dby8y4hkJQU -- examples of the format you need to wrangle your own writing into, in order to make a high-quality set, can be found on HuggingFace or Augmentoolkit
(basically for the "sharegpt" format which is common it's a list of dicts with each dict having a "conversations" key, which itself links to a list of dicts each with the "from" and "value" keys. "from" is either system|user|assistant and value is the contents of that message.)
Get some data in that format together using your own writing, rent some GPUs on runpod (recommend like 6 or 7 A40s for finetuning mistral with a full finetune) and then you should be good to go!
Nice read. Any advice of how to get started fast for fine-tuning a model with my own writing?
Hey! If this is your first finetuning of a model you might want to follow a guide like this https://www.youtube.com/watch?v=dby8y4hkJQU -- examples of the format you need to wrangle your own writing into, in order to make a high-quality set, can be found on HuggingFace or Augmentoolkit
(basically for the "sharegpt" format which is common it's a list of dicts with each dict having a "conversations" key, which itself links to a list of dicts each with the "from" and "value" keys. "from" is either system|user|assistant and value is the contents of that message.)
Get some data in that format together using your own writing, rent some GPUs on runpod (recommend like 6 or 7 A40s for finetuning mistral with a full finetune) and then you should be good to go!