Carlos Mocholí

Results 427 comments of Carlos Mocholí

Then you are in luck because since it's LLaMA based, everything should be supported already. You just need to add a config just like https://github.com/Lightning-AI/lit-gpt/blob/72958cbd4b72f79e7403d63c363e3b3da3c72b29/lit_gpt/config.py#L282-L298 but for that model specifically

@ht0rohit FastChat uses the t5 model architecture which we don't plan to support

@codeur-rapide Were you successful? If so, would you like to contribute adding support for it with a PR?

Can you share a branch from a fork or a diff with all the changes that you've made to the repository and the commands to repro your results?

What does `model.max_seq_length` return? I suspect that you have samples in your data that are longer than this. You should be seeing this here: https://github.com/Lightning-AI/lit-gpt/blob/main/finetune/adapter.py#L159-L164

Notice that the error appears during validation. Did you check the validation data?

We haven't tested AMD support. But it should generally work, the code might need minor modifications. Feel free to try and report back

We might want to revive https://github.com/Lightning-AI/litgpt/pull/878 if we are doing this. What do you prefer?

> The reason is that in the main branch, the yaml file contains the old bias notation. This also signals a breaking change. Can you add backwards-compatibility code to `Config`...