Hi, I am working on a text regression task using the AutoTrain API.
My request to the /projects/create endpoint looks like this
{
“username”: “loganamcnichols”,
“proj_name”: “test”,
“task”: 10,
“config”: {
“language”: “unk”,
“advanced”: false,
“autotrain”: true,
“max_models”: 1,
“hub_model”: “microsoft/deberta-base”,
“instance”: {
“provider”: “ovh”,
“instance_type”: “p3”,
“max_runtime_seconds”: 172800,
“num_instances”: 1,
“disk_size_gb”: 150
},
“evaluation”: {
“models”: ,
“metrics”:
},
“benchmark”: {
“model”: “”,
“dataset”: “”,
“submission_dataset”: “”,
“submission_id”: “”,
“create_prediction_repo”: true
},
“params”: [{“task”: “text_single_column_regression”, “learning_rate”: X}]
}
}
I’ve experimented with extreme values for the learning rate, e.g. 0.0000001, but training still seems to produce similar results regardless of what value I put in. So it seems like these parameters aren’t being used. Does anyone know why?