How did you create AWQ-quantized weights?

#5
by nightdude - opened

Could you share how you created the AWQ weights? Sorry, I'm new to AWQ (always used BNB). Thanks!

Please have a look at AutoAWQ
https://github.com/casper-hansen/AutoAWQ

@casperhansen Can you share the config that you used to quantize this model, do you just use the default config? What is the data that you used for calibration?

Struggled a bit with quantisation, made some notes on what worked for me here: https://huggingface.co/mattmalcher/Llama-3-70B-Instruct-awq

This comment has been hidden

Sign up or log in to comment