Instructions to use jzhoubu/dpr-nq with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use jzhoubu/dpr-nq with Transformers:
# Load model directly from transformers import Retriever model = Retriever.from_pretrained("jzhoubu/dpr-nq", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Upload config
Browse files- config.json +4 -3
config.json
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
{
|
|
|
|
| 2 |
"alias": "dpr",
|
| 3 |
"architectures": [
|
| 4 |
"Retriever"
|
|
@@ -6,16 +7,16 @@
|
|
| 6 |
"device": null,
|
| 7 |
"dropout": 0.1,
|
| 8 |
"encoder_p": {
|
|
|
|
| 9 |
"hidden_size": 768,
|
| 10 |
"max_seq_len": 256,
|
| 11 |
-
"model_id": "bert-base-uncased",
|
| 12 |
"norm": false,
|
| 13 |
"pretrained": true
|
| 14 |
},
|
| 15 |
"encoder_q": {
|
|
|
|
| 16 |
"hidden_size": 768,
|
| 17 |
-
"max_seq_len":
|
| 18 |
-
"model_id": "bert-base-uncased",
|
| 19 |
"norm": false,
|
| 20 |
"pretrained": true
|
| 21 |
},
|
|
|
|
| 1 |
{
|
| 2 |
+
"_name_or_path": "/export/data/jzhoubu/workspace/VDR-dense/experiments/1130-dpr-nq-e40-bs32x8-neg1-lccpu29/train/vdr_40",
|
| 3 |
"alias": "dpr",
|
| 4 |
"architectures": [
|
| 5 |
"Retriever"
|
|
|
|
| 7 |
"device": null,
|
| 8 |
"dropout": 0.1,
|
| 9 |
"encoder_p": {
|
| 10 |
+
"base_model_id": "bert-base-uncased",
|
| 11 |
"hidden_size": 768,
|
| 12 |
"max_seq_len": 256,
|
|
|
|
| 13 |
"norm": false,
|
| 14 |
"pretrained": true
|
| 15 |
},
|
| 16 |
"encoder_q": {
|
| 17 |
+
"base_model_id": "bert-base-uncased",
|
| 18 |
"hidden_size": 768,
|
| 19 |
+
"max_seq_len": 128,
|
|
|
|
| 20 |
"norm": false,
|
| 21 |
"pretrained": true
|
| 22 |
},
|