calpt commited on
Commit
ef9a6a2
1 Parent(s): 3e9ba5c

Add adapter roberta-base-cosmosqa_pfeiffer version AdapterFusion

Browse files
Files changed (4) hide show
  1. ._adapter_config.json +0 -0
  2. README.md +63 -0
  3. adapter_config.json +41 -0
  4. pytorch_adapter.bin +3 -0
._adapter_config.json ADDED
Binary file (220 Bytes). View file
 
README.md ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - adapter-transformers
4
+ - adapterhub:comsense/cosmosqa
5
+ - roberta
6
+ datasets:
7
+ - cosmos_qa
8
+ license: "apache-2.0"
9
+ ---
10
+
11
+ # Adapter `roberta-base-cosmosqa_pfeiffer` for roberta-base
12
+
13
+ Pfeiffer Adapter trained on Cosmos QA.
14
+
15
+
16
+ **This adapter was created for usage with the [Adapters](https://github.com/Adapter-Hub/adapters) library.**
17
+
18
+ ## Usage
19
+
20
+ First, install `adapters`:
21
+
22
+ ```
23
+ pip install -U adapters
24
+ ```
25
+
26
+ Now, the adapter can be loaded and activated like this:
27
+
28
+ ```python
29
+ from adapters import AutoAdapterModel
30
+
31
+ model = AutoAdapterModel.from_pretrained("roberta-base")
32
+ adapter_name = model.load_adapter("AdapterHub/roberta-base-cosmosqa_pfeiffer")
33
+ model.set_active_adapters(adapter_name)
34
+ ```
35
+
36
+ ## Architecture & Training
37
+
38
+ - Adapter architecture: pfeiffer
39
+ - Prediction head: None
40
+ - Dataset: [Cosmos QA](https://www.tau-nlp.org/commonsenseqa)
41
+
42
+ ## Author Information
43
+
44
+ - Author name(s): Jonas Pfeiffer
45
+ - Author email: [email protected]
46
+ - Author links: [Website](https://pfeiffer.ai), [GitHub](https://github.com/JoPfeiff), [Twitter](https://twitter.com/@PfeiffJo)
47
+
48
+
49
+
50
+ ## Citation
51
+
52
+ ```bibtex
53
+ @article{Pfeiffer2020AdapterFusion,
54
+ author = {Pfeiffer, Jonas and Kamath, Aishwarya and R{\"{u}}ckl{\'{e}}, Andreas and Cho, Kyunghyun and Gurevych, Iryna},
55
+ journal = {arXiv preprint},
56
+ title = {{AdapterFusion}: Non-Destructive Task Composition for Transfer Learning},
57
+ url = {https://arxiv.org/pdf/2005.00247.pdf},
58
+ year = {2020}
59
+ }
60
+
61
+ ```
62
+
63
+ *This adapter has been auto-imported from https://github.com/Adapter-Hub/Hub/blob/master/adapters/ukp/roberta-base-cosmosqa_pfeiffer.yaml*.
adapter_config.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config": {
3
+ "adapter_residual_before_ln": false,
4
+ "cross_adapter": false,
5
+ "dropout": 0.0,
6
+ "factorized_phm_W": true,
7
+ "factorized_phm_rule": false,
8
+ "hypercomplex_nonlinearity": "glorot-uniform",
9
+ "init_weights": "bert",
10
+ "inv_adapter": null,
11
+ "inv_adapter_reduction_factor": null,
12
+ "is_parallel": false,
13
+ "learn_phm": true,
14
+ "leave_out": [],
15
+ "ln_after": false,
16
+ "ln_before": false,
17
+ "mh_adapter": false,
18
+ "non_linearity": "relu",
19
+ "original_ln_after": true,
20
+ "original_ln_before": true,
21
+ "output_adapter": true,
22
+ "phm_bias": true,
23
+ "phm_c_init": "normal",
24
+ "phm_dim": 4,
25
+ "phm_init_range": 0.0001,
26
+ "phm_layer": false,
27
+ "phm_rank": 1,
28
+ "reduction_factor": 16,
29
+ "residual_before_ln": true,
30
+ "scaling": 1.0,
31
+ "shared_W_phm": false,
32
+ "shared_phm_rule": true,
33
+ "use_gating": false
34
+ },
35
+ "hidden_size": 768,
36
+ "model_class": "RobertaAdapterModel",
37
+ "model_name": "roberta-base",
38
+ "model_type": "roberta",
39
+ "name": "cosmosqa",
40
+ "version": "0.2.0"
41
+ }
pytorch_adapter.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39bc7d39165ecae4579c1531e556e5c997a0a114e19b637e0be969dd29680858
3
+ size 3595046