Upload korscideberta-abstractcls.ipynb
Browse files
korscideberta-abstractcls.ipynb
CHANGED
@@ -684,15 +684,6 @@
|
|
684 |
"#### νμΈνλ λ° λͺ¨λΈ μ
λ‘λ μλ£"
|
685 |
]
|
686 |
},
|
687 |
-
{
|
688 |
-
"cell_type": "markdown",
|
689 |
-
"metadata": {
|
690 |
-
"id": "K1SWtSpJ1WpZ"
|
691 |
-
},
|
692 |
-
"source": [
|
693 |
-
"Putting it all together, we can finally instantiate the Trainer by passing all required components. We'll use the `\"validation\"` split as the held-out dataset during training."
|
694 |
-
]
|
695 |
-
},
|
696 |
{
|
697 |
"cell_type": "code",
|
698 |
"execution_count": null,
|
@@ -787,17 +778,6 @@
|
|
787 |
"The Trainer is ready to go π You can start training by calling `trainer.train()`."
|
788 |
]
|
789 |
},
|
790 |
-
{
|
791 |
-
"cell_type": "markdown",
|
792 |
-
"metadata": {
|
793 |
-
"id": "CaLVY2nfmcgS"
|
794 |
-
},
|
795 |
-
"source": [
|
796 |
-
"Cool, we see that the model seems to learn something! Training loss and validation loss is going down and the accuracy also ends up being well over random chance (20%). Interestingly, we see accuracy of around **58.6 %** already after 5000 steps which doesn't improve that much anymore afterward. Choosing a bigger model or training for longer would have probably given better results here, but that's good enough for our hypothetical use case!\n",
|
797 |
-
"\n",
|
798 |
-
"Alright, finally let's upload the model checkpoint to the Hub."
|
799 |
-
]
|
800 |
-
},
|
801 |
{
|
802 |
"cell_type": "markdown",
|
803 |
"metadata": {
|
@@ -812,15 +792,6 @@
|
|
812 |
"Let's dive into evaluating the model π€Ώ."
|
813 |
]
|
814 |
},
|
815 |
-
{
|
816 |
-
"cell_type": "markdown",
|
817 |
-
"metadata": {
|
818 |
-
"id": "wvefwWDdwkl4"
|
819 |
-
},
|
820 |
-
"source": [
|
821 |
-
"The model has been uploaded to the Hub under [`deberta_v3_amazon_reviews`](https://huggingface.co/patrickvonplaten/deberta_v3_amazon_reviews) after training, so in a first step, let's download it from there again. If this notebook is run all at once the following cell will simply load the model from the cache."
|
822 |
-
]
|
823 |
-
},
|
824 |
{
|
825 |
"cell_type": "code",
|
826 |
"execution_count": 21,
|
|
|
684 |
"#### νμΈνλ λ° λͺ¨λΈ μ
λ‘λ μλ£"
|
685 |
]
|
686 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
687 |
{
|
688 |
"cell_type": "code",
|
689 |
"execution_count": null,
|
|
|
778 |
"The Trainer is ready to go π You can start training by calling `trainer.train()`."
|
779 |
]
|
780 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
781 |
{
|
782 |
"cell_type": "markdown",
|
783 |
"metadata": {
|
|
|
792 |
"Let's dive into evaluating the model π€Ώ."
|
793 |
]
|
794 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
795 |
{
|
796 |
"cell_type": "code",
|
797 |
"execution_count": 21,
|