pbihao / slm Goto Github PK

Thanks for your work. I have encountered one bug when runing the code in Line 134 of SLM/SLM-llama/retriever.py:

weight_offset = torch.take_along_dim(self.weight_offset, idx_vote[:, None,None], dim=1)weight_offset = torch.take_along_dim(self.weight_offset, idx_vote[:, None,None], dim=1)

The error is:
RuntimeErrorRuntimeError: : The size of tensor a (12) must match the size of tensor b (2) at non-singleton dimension 0

Could you help figure it out? Thanks.

Hello. Thank you for sharing your source code.

While reviewing it, I noticed the variable "self.pool_size" being used. From my understanding, it seems to represent the number of core vectors for each task. Is my understanding correct? If not, could you please clarify its purpose?

The reason I'm confused about the variable is the bash script file in your repository.

In vectordb/script.sh, you write like following in the line 54

# ------------------------- pool: 24, groups: 1 ------------------------------------------

How we can define the 24 tasks from AGNews (4 classes), Yelp (5 classes), DBPedia (14 classes), Amazon (5 classes), and Yahoo (10 classes) datasets ?

Based on your paper and source code, it seems that the pool_size variable represents the number of tasks excluding the current task. Therefore, the pool_size should be 4 for all continual learning sequences.

Am I correct...?

ps. If you have some time, could you describe the exact experimental steps needed to reproduce your report?

Thank you in advance.