-
Notifications
You must be signed in to change notification settings - Fork 38
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
feat: add quantize_model() fn & a MidnightRose70B #79
Conversation
logger.info(f"Volume now contains {quantized_model_path}") | ||
|
||
|
||
def load_open_instruct(tokenizer, n_samples=128): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
this calibration dataset loading is mostly forked from the examples here:
https://github.com/AutoGPTQ/AutoGPTQ/blob/main/examples/quantization/quant_with_alpaca.py
where instead of reading from a json file, we sample from this dataset instead:
https://huggingface.co/datasets/VMware/open-instruct
only adjustments to the code were to limit number of samples & change column labels for the dataset
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM 👍
Details
without a calibration dataset:
https://huggingface.co/sambarnes/Midnight-Rose-70B-v2.0.3-GPTQ-naive
with a calibration dataset (VMWare/open-instruct):
https://huggingface.co/sambarnes/Midnight-Rose-70B-v2.0.3-GPTQ
upon further testing: i couldnt actually notice a difference between the two, so gonna just use the not naive one
Code of Conduct