Open pm-mck opened 2 months ago
Hi @pm-mck ,
Thanks for reporting the issue. I'm trying to reproduce it with our latest release. But I hit the issue below
Loading model - /home/ubuntu/RWKV-x060-World-3B-v2.1-20240417-ctx4096
Traceback (most recent call last):
File "gh.py", line 49, in <module>
model_rwkv = RWKV(model=args.MODEL_NAME, strategy=args.strategy)
File "/shared/on-call/env/lib/python3.8/site-packages/torch/jit/_script.py", line 303, in init_then_script
original_init(self, *args, **kwargs)
File "/shared/on-call/env/lib/python3.8/site-packages/rwkv/model.py", line 186, in __init__
raise ValueError("Invalid strategy. Please read https://pypi.org/project/rwkv/")
ValueError: Invalid strategy. Please read https://pypi.org/project/rwkv/
Did I miss something?
Hi @jyang-aws - thank you for your response. Yes, I had to modify the RWKV import itself to allow xla tensors. Nothing major, but it has a regex check to validate strategies and I also forced it to use cpu for indexed tensors. I can send you a patch if that's helpful.
thanks. I'm able to reproduce the issue now. will fix from our end and keep you updated.
@jyang-aws I am facing the exact same issue! Is there a fix for it ?
Hello,
I am working on tracing RWKV using neuronx and I received the following error:
Error
[TEN404] (_dynamic-update-slice.5283) Internal tensorizer error - Please open a support ticket at https://github.com/aws-neuron/aws-neuron-sdk/issues/new
Version Info
The code I'm using to trace is pretty basic. Maybe it's too basic since it's reporting:
Here's the code, it's derived from the RWKV chat example:
Code
Any help is appreciated!