Teknium

Since its fixed by reverting hf transformers, I dont think its cuda related?

Any updates to this? Are all things good now? Can we fix old models by changing the tokenizer config or?

> > @teknium1 You need to retrain on the fixed/updated base HF models. Anything trained using old transformer code on the decapoda models are bound to break. You can hack...

> > For everyone's convenience, I've uploaded **llama models converted with the latest transformer git head** here: > > **7B** - https://huggingface.co/yahma/llama-7b-hf **13B** - https://huggingface.co/yahma/llama-13b-hf > > Unfortunately, unlike the...

I did both of the suggestions (had bnb 0.37.2 and latest git transformers) but still ran into the issue

Please add this because I have alpaca model and it was trained on a bad dataset with many cases of input and output fields having "" text in them which...

> @teknium1 I think that `bad_words_list` as it is would be enough for your example. But if you still feel something like the `logit_bias` parameter is what you need, react...

I just ran into this error training alpaca-lora.. no fix available now?

> Hi @teknium1 [this PR](https://github.com/wandb/wandb/pull/5283) may fix the issue, but it's currently under review. I will keep this thread updated once it's merged to master branch. Its okay, for whatever...

Can you actually just "fine tune more context size"?