daegon Yu

icon location Seoul icon location I am an NLP researcher with a focus on semantic retriever, retrieval-augmented generation (RAG), and domain adaptation for large language models (LLMs).

Results 14 comments of daegon Yu

Are there any guidelines for loading a CPT (continued pre-training) model and retraining it on a different data set?

Additionally, I have a question. When learning a decoder model, I understand that when the Instruction part is input to the model, only the Response part is learned by calculating...

Are there any guidelines for loading a CPT (continued pre-training) model and retraining it on a different data set?

Oh this is what I was looking for. thank you!

Are there any guidelines for loading a CPT (continued pre-training) model and retraining it on a different data set?

One thing I'm wondering about while researching this is, is it okay to assume that using DataCollatorForCompletionOnlyLM(response_template, tokenizer=tokenizer) and using DataCollatorForSeq2Seq(tokenizer = tokenizer) with train_on_responses_only( trainer, #instruction_part = "user\n\n", response_part...

GRPO merging Adapters+ re-training not working

Is this issue resolved? I am also training LoRA using the unsloth/gemma2-9b-it model, and when I resume training, I find that the log is not continuing the checkpoint reward. ![Image](https://github.com/user-attachments/assets/4c44d6be-e77f-48dc-98c8-bda175aaae06)