When DPO Yi-34B Assertion srcIndex < srcSelectDimSize failed

[victorShawFan]

when i use train_dpo_llama_34b.sh to dpo Yi-34B-Chat, there will be an "Array out of bounds" kind of problem



i use huggingface Yi-34B-Chat ckpt and tokenizer
i didn't modify any code about tokenizers, please help

[hijkzzz]

Thank you for your feedback, we will reproduce and fix it as soon as possible.

[hijkzzz]

Please set use_fast=False in the openrlhf/utils/utils.py tokenizer = AutoTokenizer.from_pretrained(pretrain, trust_remote_code=True, use_fast=False, **sp_tokens)

[hijkzzz]

now you can use the option "--disable_fast_tokenizer" in train_dpo.py

[victorShawFan]

still don't work, same bug, what problem could that be?

[hijkzzz]

still don't work, same bug, what problem could that be?

use "--disable_fast_tokenizer"

[victorShawFan]

problem solved, now it worked, appreciate

btw, is this loss normal ?

[hijkzzz]

It makes more sense to look at acc mean