Hugging Face's logo

aubmindlab
/
aragpt2-base

Text Generation
Transformers
PyTorch
TensorFlow
JAX
TensorBoard
Safetensors
Arabic
gpt2
text-generation-inference
Model card Files
xet
 Metrics Community
2

PEFT LoRA adapter compatibility and redistribution note?

#2
by ISLAM-PO - opened
Discussion
ISLAM-PO

Hello,

I published a small experimental LoRA adapter on top of aragpt2-base:
https://huggingface.co/ISLAM-PO/MasryGPT-Flash-Adapter

adapter_config.json records the base model, and loading works with:
PeftModel.from_pretrained(AutoModelForCausalLM.from_pretrained("aubmindlab/aragpt2-base"), adapter_id).

Two questions:

  1. Is there a recommended tokenizer revision to pin for adapter compatibility?
  2. The adapter card reproduces the AraGPT2 license notice. Is there anything else you require for redistribution of adapter-only weights?

I kept the adapter experimental and documented it as non-standalone.
Thanks.
wissamantoun
 AUB MIND LAB org
  1. just pin the latest
  2. feel free to publish nothing else required

Sign up or log in to comment