Fine tuning instructions
#2
by
hoper2
- opened
Hi, I'd like to fine-tune tinyllava. Could you please point me to any instructions for how to do this? I tried running the LLaVa finetune scripts using tinyllava-1.1b-v0.1 folder for the model folder argument, but I keep running into issues, like
WARNING: tokenization mismatch: 140 vs. 139. (ignored)
...
RuntimeError: expected scalar type Half but found Float