What do you like about the model? And what should we improve?
Thank you all for your interest in our model 🤗. While it's still developing, and there are many ways to improve the model, I would like to ask what the most relevant use case for you is. What do you like, and what do we need to improve considering your request? Your feedback is really helpful for us because it allows us to focus on the really important stuff.
Thanks for the good work ! Unfortunately my use case is in italian, it would be great if it was available in other languages too
@Batchounet , thank you for your feedback! Btw we have the multilingual model. You can find it here.
It is spectacular, however I work with big data and I need to process thousands of records as fast as possible. Is there any way to apply some method of compressing the model as quantization?
I have used that library several times but only for classification models, but when using it in those that are zero-shot it is impossible to use the pipeline. Maybe you know how to solve it? I leave you a screenshot running from colab.
The parameter of the zero-shot classification pipeline that takes labels is called candidate_labels. Also, for someone who is struggling with using a pipeline with ORTModel, you can limit the tokenizer key using the following command:
tokenizer.model_input_names = ['input_ids', 'attention_mask']
Hi,
The quality of the model is the best I've seen so far, in particular in the zero-shot classifiers! Really stunning. Though where I feel it struggles is with performance on CPU, in particular when you have a semi-large list of labels (I've tried with 30). Is there any way to improve inference speed on CPU while retaining its accuracy?
Thanks loads!