NotImplementedError for GPT-J-6b

kmeng01 / memit

Mass-editing thousands of facts into a transformer memory (ICLR 2023)

https://memit.baulab.info

MIT License

438 stars 53 forks source link

NotImplementedError for GPT-J-6b #3

Closed daniel-furman closed 1 year ago

daniel-furman commented 1 year ago

I was trying to run "memit.ipynb" on the premium GPU in Colab with the GPT-J-6b model. I believe the error message I got is not related to OOM errors, rather, a potential problem in the code? For context, the gpt2-xl does work as expected.

daniel-furman commented 1 year ago

@kmeng01

daniel-furman commented 1 year ago

Closing this issue.

Fix: Change the "AutoModelForCausalLM" to "GPTJForCausalLM" and stripping a couple of the original notebook's variables.

The "NotImplementedError" in the original notebook remains unsolved I believe.

^ with this edit, my experience suggests the memit.ipynb runs on one (Premium) GPU in Google Colab.