Question Answering
Transformers
English
information-retrieval
reranking
query-expansion
bm25
deberta-v3
Instructions to use voidism/EAR with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use voidism/EAR with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("question-answering", model="voidism/EAR")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("voidism/EAR", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Add files using upload-large-folder tool
Browse files- README.md +86 -0
- SOURCE_MANIFEST.sha256 +13 -0
- best_nq/vanilla/scorer-dev-answer-best.bin +3 -0
- best_nq/vanilla/scorer-dev-sentence-best.bin +3 -0
- best_nq/vanilla/scorer-dev-title-best.bin +3 -0
- best_nq/wtop1/scorer-dev-answer-best.bin +3 -0
- best_nq/wtop1/scorer-dev-sentence-best.bin +3 -0
- best_nq/wtop1/scorer-dev-title-best.bin +3 -0
- best_trivia/vanilla/scorer-dev-answer-best.bin +3 -0
- best_trivia/vanilla/scorer-dev-sentence-best.bin +3 -0
- best_trivia/vanilla/scorer-dev-title-best.bin +3 -0
- best_trivia/wtop1/scorer-answer-best.bin +3 -0
- best_trivia/wtop1/scorer-dev-answer-best.bin +3 -0
- best_trivia/wtop1/scorer-dev-sentence-best.bin +3 -0
- best_trivia/wtop1/scorer-dev-title-best.bin +3 -0
README.md
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
library_name: transformers
|
| 6 |
+
base_model: microsoft/deberta-v3-base
|
| 7 |
+
tags:
|
| 8 |
+
- question-answering
|
| 9 |
+
- information-retrieval
|
| 10 |
+
- reranking
|
| 11 |
+
- query-expansion
|
| 12 |
+
- bm25
|
| 13 |
+
- deberta-v3
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# EAR query reranker checkpoints
|
| 17 |
+
|
| 18 |
+
Official scorer checkpoints for **Expand, Rerank, and Retrieve: Query
|
| 19 |
+
Reranking for Open-Domain Question Answering** (Findings of ACL 2023).
|
| 20 |
+
|
| 21 |
+
EAR is a query Expansion And Reranking method. It samples diverse query
|
| 22 |
+
expansions and trains a reranker to select expansions that improve passage
|
| 23 |
+
retrieval.
|
| 24 |
+
|
| 25 |
+
## Contents
|
| 26 |
+
|
| 27 |
+
The repository preserves the original experiment directory names:
|
| 28 |
+
|
| 29 |
+
| Path | Dataset | Variant | Files |
|
| 30 |
+
| --- | --- | --- | ---: |
|
| 31 |
+
| best_nq/vanilla | Natural Questions | EAR-RI | 3 |
|
| 32 |
+
| best_nq/wtop1 | Natural Questions | EAR-RD | 3 |
|
| 33 |
+
| best_trivia/vanilla | TriviaQA | EAR-RI | 3 |
|
| 34 |
+
| best_trivia/wtop1 | TriviaQA | EAR-RD | 4 |
|
| 35 |
+
|
| 36 |
+
Each directory contains answer, sentence, and title scorer checkpoints.
|
| 37 |
+
best_trivia/wtop1 additionally preserves the historical
|
| 38 |
+
scorer-answer-best.bin file from the original release.
|
| 39 |
+
|
| 40 |
+
These are raw PyTorch checkpoint/state-dict files used by the EAR codebase.
|
| 41 |
+
They are not standalone AutoModel.from_pretrained repositories.
|
| 42 |
+
|
| 43 |
+
## Usage
|
| 44 |
+
|
| 45 |
+
Use these checkpoints with the official code:
|
| 46 |
+
|
| 47 |
+
git clone https://github.com/voidism/EAR
|
| 48 |
+
cd EAR
|
| 49 |
+
|
| 50 |
+
For EAR-RI:
|
| 51 |
+
|
| 52 |
+
bash one_pass_eval_ri.sh nq /path/to/best_nq/vanilla
|
| 53 |
+
bash one_pass_eval_ri.sh trivia /path/to/best_trivia/vanilla
|
| 54 |
+
|
| 55 |
+
For EAR-RD:
|
| 56 |
+
|
| 57 |
+
bash one_pass_eval_rd.sh nq /path/to/best_nq/wtop1
|
| 58 |
+
bash one_pass_eval_rd.sh trivia /path/to/best_trivia/wtop1
|
| 59 |
+
|
| 60 |
+
The original release used Python 3.7.13, PyTorch 1.10.1,
|
| 61 |
+
Transformers 4.24.0, Tokenizers 0.11.1, Pyserini, and Weights & Biases.
|
| 62 |
+
The rerankers were trained from microsoft/deberta-v3-base.
|
| 63 |
+
|
| 64 |
+
SOURCE_MANIFEST.sha256 contains a SHA-256 checksum for every checkpoint.
|
| 65 |
+
The source archive was models_ear.tar.gz with SHA-256
|
| 66 |
+
d756d111404fe8859fd094e313f1e2b95c489691228dfb05044c6471fe31c819.
|
| 67 |
+
|
| 68 |
+
No training or evaluation dataset is included in this model repository.
|
| 69 |
+
|
| 70 |
+
## Links
|
| 71 |
+
|
| 72 |
+
- Paper: https://aclanthology.org/2023.findings-acl.768/
|
| 73 |
+
- Code: https://github.com/voidism/EAR
|
| 74 |
+
- arXiv: https://arxiv.org/abs/2305.17080
|
| 75 |
+
|
| 76 |
+
## Citation
|
| 77 |
+
|
| 78 |
+
@inproceedings{chuang-etal-2023-expand,
|
| 79 |
+
title = {Expand, Rerank, and Retrieve: Query Reranking for Open-Domain Question Answering},
|
| 80 |
+
author = {Chuang, Yung-Sung and Fang, Wei and Li, Shang-Wen and Yih, Wen-tau and Glass, James},
|
| 81 |
+
booktitle = {Findings of the Association for Computational Linguistics: ACL 2023},
|
| 82 |
+
year = {2023},
|
| 83 |
+
pages = {12131--12147},
|
| 84 |
+
doi = {10.18653/v1/2023.findings-acl.768},
|
| 85 |
+
url = {https://aclanthology.org/2023.findings-acl.768/}
|
| 86 |
+
}
|
SOURCE_MANIFEST.sha256
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
5bd9aa0ef82d5784837ffd6fed1b438379b9752f96cafd5d8d44be679b0052e2 ./best_nq/vanilla/scorer-dev-answer-best.bin
|
| 2 |
+
e041c7c0424ceaecb55133303e884447581429cfda01273866a55b7856dce4d0 ./best_nq/vanilla/scorer-dev-sentence-best.bin
|
| 3 |
+
4f96e4f9d8efc7eaa8f260e01cac6aecbbe129759d2f9cf1e2b9c856e68714cf ./best_nq/vanilla/scorer-dev-title-best.bin
|
| 4 |
+
715546d5ce835c03b0ae3b792b507ce7d6f1b07697534cf53e323ded99845c80 ./best_nq/wtop1/scorer-dev-answer-best.bin
|
| 5 |
+
2b3327e633db4eb33ad38aa20ca11dc85f4b39d84d7ba246015a37ff6c632c5c ./best_nq/wtop1/scorer-dev-sentence-best.bin
|
| 6 |
+
c51f02480ce687cf365d45e21e4a7e7aba43e87c70b7deec863774de809240f4 ./best_nq/wtop1/scorer-dev-title-best.bin
|
| 7 |
+
7fd105ea12d2de413fe245f6c58ee41af0a2e142d6f7d39e75fd8a5271fa78d0 ./best_trivia/vanilla/scorer-dev-answer-best.bin
|
| 8 |
+
42ba53412795d5e747d614489aac2351ea28dfbf2bebbfd9bfab801b26b85efa ./best_trivia/vanilla/scorer-dev-sentence-best.bin
|
| 9 |
+
56d7672f2713e64363b843977b79b66bac8d229f13bd975950aae74c7f79ccba ./best_trivia/vanilla/scorer-dev-title-best.bin
|
| 10 |
+
034a0e2933f582876088c58d1646a7e69be497a2f9e3127ce3281ebe19914dcf ./best_trivia/wtop1/scorer-answer-best.bin
|
| 11 |
+
d06c092e15253b597d846ec76e6708ae9b12ad9fa9c4a084b284d35184062bf7 ./best_trivia/wtop1/scorer-dev-answer-best.bin
|
| 12 |
+
1e9265e0bd051336dce54068a7dfc622f9c14c13397e0695301c06e760119dce ./best_trivia/wtop1/scorer-dev-sentence-best.bin
|
| 13 |
+
441c04cece5f5cff39b74738c719dfdfbf43a803ee0584448c2adc9b81598cc7 ./best_trivia/wtop1/scorer-dev-title-best.bin
|
best_nq/vanilla/scorer-dev-answer-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5bd9aa0ef82d5784837ffd6fed1b438379b9752f96cafd5d8d44be679b0052e2
|
| 3 |
+
size 438018349
|
best_nq/vanilla/scorer-dev-sentence-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e041c7c0424ceaecb55133303e884447581429cfda01273866a55b7856dce4d0
|
| 3 |
+
size 438018349
|
best_nq/vanilla/scorer-dev-title-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4f96e4f9d8efc7eaa8f260e01cac6aecbbe129759d2f9cf1e2b9c856e68714cf
|
| 3 |
+
size 438018349
|
best_nq/wtop1/scorer-dev-answer-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:715546d5ce835c03b0ae3b792b507ce7d6f1b07697534cf53e323ded99845c80
|
| 3 |
+
size 438018349
|
best_nq/wtop1/scorer-dev-sentence-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2b3327e633db4eb33ad38aa20ca11dc85f4b39d84d7ba246015a37ff6c632c5c
|
| 3 |
+
size 438018349
|
best_nq/wtop1/scorer-dev-title-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c51f02480ce687cf365d45e21e4a7e7aba43e87c70b7deec863774de809240f4
|
| 3 |
+
size 438018349
|
best_trivia/vanilla/scorer-dev-answer-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7fd105ea12d2de413fe245f6c58ee41af0a2e142d6f7d39e75fd8a5271fa78d0
|
| 3 |
+
size 438018349
|
best_trivia/vanilla/scorer-dev-sentence-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:42ba53412795d5e747d614489aac2351ea28dfbf2bebbfd9bfab801b26b85efa
|
| 3 |
+
size 438018349
|
best_trivia/vanilla/scorer-dev-title-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:56d7672f2713e64363b843977b79b66bac8d229f13bd975950aae74c7f79ccba
|
| 3 |
+
size 438018349
|
best_trivia/wtop1/scorer-answer-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:034a0e2933f582876088c58d1646a7e69be497a2f9e3127ce3281ebe19914dcf
|
| 3 |
+
size 438018349
|
best_trivia/wtop1/scorer-dev-answer-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d06c092e15253b597d846ec76e6708ae9b12ad9fa9c4a084b284d35184062bf7
|
| 3 |
+
size 438018349
|
best_trivia/wtop1/scorer-dev-sentence-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1e9265e0bd051336dce54068a7dfc622f9c14c13397e0695301c06e760119dce
|
| 3 |
+
size 438018349
|
best_trivia/wtop1/scorer-dev-title-best.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:441c04cece5f5cff39b74738c719dfdfbf43a803ee0584448c2adc9b81598cc7
|
| 3 |
+
size 438018349
|