Downloads · 30 days
385
6% of all-time downloads
Sindhu/rembert-squad2
rembert-squad2 is a question answering model from Sindhu. Use it when the input is a question plus a passage. It is set up for transformers.
This model is finetuned for QA task on Squad2 from Rembert checkpoint.
Downloads · 30 days
385
6% of all-time downloads
All-time downloads
6.3K
Public
Repo size
4.6 GB
Likes
3
Public
Click a slice to open those files.
.bin2.3 GB · 99%
From the Hugging Face model README
This model is finetuned for QA task on Squad2 from Rembert checkpoint.
Batch Size: 4
Grad Accumulation Steps = 8
Total epochs = 3
MLM Checkpoint = "rembert"
max_seq_len = 256
learning_rate = 1e-5
lr_schedule = LinearWarmup
warmup_ratio = 0.1
doc_stride = 128
Metrics generated from the official Squad2 evaluation script
{
"exact": 84.51107554956624,
"f1": 87.46644042781853,
"total": 11873,
"HasAns_exact": 80.97165991902834,
"HasAns_f1": 86.89086491219469,
"HasAns_total": 5928,
"NoAns_exact": 88.04037005887301,
"NoAns_f1": 88.04037005887301,
"NoAns_total": 5945
}
For any questions, you can reach out to me on Twitter