Open jt4n opened 1 day ago
Thank you for your interest in our work. You should invoke the AutoModelForCausalLMWithValueHead
in trl
for training to avoid this error.
You can refer to Section Model Loader
in our implementation_details.md for more details. I have updated this file, and thanks for your good question.
Additionally, you should use bf16
in your training script instead of fp16
to avoid any potential errors.
Thanks a lot! I can start the training process referring the Model Loader
section.
Hi, I'm reproducing your job.
When I use the
round3_training_data.json
data to sft thedeepseek-math-7b-base-value_model
(after added value head), I got below error:I added the modified compute_loss function you offered in this page, to the llama_factory
CustomSeq2SeqTrainer
class to override the original compute_loss of transformersTrainer
class.The training code is like:
Did I make any mistake? How to solve it?