The flag/option should allow to
- Show the expected training cost for a finetuning run for that particular model.
- Validate finetuning JSONL files before starting
Note: 1. is particularly useful for big files and to decide whether to run with say 4.1 or 4.1-mini, etc
while 2. would reduce the number of failed training runs
Safety tooling (https://github.com/safety-research/safety-tooling) has something like that to validate jsonl files and get the training costs. See their readme
The flag/option should allow to
Note: 1. is particularly useful for big files and to decide whether to run with say 4.1 or 4.1-mini, etc
while 2. would reduce the number of failed training runs
Safety tooling (https://github.com/safety-research/safety-tooling) has something like that to validate jsonl files and get the training costs. See their readme