Yili Hong

Access denied

1

I have already applied for access to meta llama3. And I also receive the email about how can I use download scripts to download it, but I still cannot have...

> In our model, we share the same weight matrix between the two embedding layers and the pre-softmax linear transformation. Hello! I have the paper recently and find that this...

I want to test the ability of llama2-7b-hf instead of llama2-7b-chat. But llama2-7b-hf is a base model not having a chat template, so how can I use llama2-7b-hf and transformers...

How to finetune mistral-7b with fastchat?

1

When I use fastchat to finetune llama2, everything is ok. But when I want to finetune mistral, it shows that "transformer layer not found". I know the main reason is...

Could not find the transformer layer class to wrap in the model.

2

When I use training code to finetune mistral-7b,I have met the exception like "Could not find the transformer layer class to wrap in the model.". How can I fix it?

How to set customized reward function?

1

How to set customized reward function in RL algorithms like ppo, grpo etc.? I found that the `compute_score` parameter in the file `main_ppo.py` is always set to None and cannot...

[Question] Does verl support muilti-round conversation RL training?

1

Does verl support muilti-round conversation RL training? if it does, which format should I set the dataset parquet files?