(WIP) Finding a runnable configuration

#4
by coalee - opened

In my case, the value of text_config.no_rope_layers in config.json needs to be updated to [] instead of 4, like the meta-llama/Llama-4-Scout-17B-16E . Otherwise, the below error pops up.

File ".../python3.11/site-packages/transformers/models/llama4/configuration_llama4.py", line 382, in __init__
    self.layer_types = [
                       ^
TypeError: 'int' object is not iterable

Environment:

Hopper
transformers==4.57.0
axolotl==0.12.2

I am still failing to start a training even with the above fix :/ . I may append coming-up fixes.

Sign up or log in to comment