Spaces:
Sleeping
Sleeping
Sean MacAvaney
commited on
Commit
β’
fcfbc2c
1
Parent(s):
9fc16dc
update
Browse files- README.md +21 -6
- app.py +37 -0
- packages.txt +5 -0
- requirements.txt +5 -0
- wrapup.md +4 -0
README.md
CHANGED
@@ -1,12 +1,27 @@
|
|
1 |
---
|
2 |
-
title:
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
-
sdk_version: 3.
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
---
|
11 |
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
title: PyTerrier MonoT5
|
3 |
+
emoji: π
|
4 |
+
colorFrom: green
|
5 |
+
colorTo: green
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 3.7
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
---
|
11 |
|
12 |
+
# π PyTerrier: MonoT5
|
13 |
+
|
14 |
+
This is a demonstration of [PyTerrier's T5 package](https://github.com/terrierteam/pyterrier_t5).
|
15 |
+
|
16 |
+
MonoT5 functions as a `RβR` (reranking, result-to-result) transformer and can be used in pipelines accordingly. For example, you will
|
17 |
+
often pipe the output of a first-stage retrieval function into MonoT5:
|
18 |
+
|
19 |
+
<div class="pipeline">
|
20 |
+
<div class="df" title="Query Frame">Q</div>
|
21 |
+
<div class="transformer" title="PisaRetrieve Transformer">TerrierRetrieve</div>
|
22 |
+
<div class="df" title="Result Frame">R</div>
|
23 |
+
<div class="transformer" title="get_text Transformer">get_text</div>
|
24 |
+
<div class="df" title="Result Frame">R</div>
|
25 |
+
<div class="transformer attn" title="MonoT5 Transformer">MonoT5</div>
|
26 |
+
<div class="df" title="Result Frame">R</div>
|
27 |
+
</div>
|
app.py
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import gradio as gr
|
3 |
+
import pyterrier as pt
|
4 |
+
pt.init()
|
5 |
+
from pyterrier_gradio import Demo, MarkdownFile, interface, df2code, code2md, EX_R
|
6 |
+
from pyterrier_t5 import MonoT5ReRanker
|
7 |
+
|
8 |
+
model = MonoT5ReRanker()
|
9 |
+
|
10 |
+
COLAB_NAME = 'pyterrier_t5.ipynb'
|
11 |
+
COLAB_INSTALL = '''
|
12 |
+
!pip install -q git+https://github.com/terrier-org/pyterrier_t5
|
13 |
+
'''.strip()
|
14 |
+
|
15 |
+
def predict(input):
|
16 |
+
code = f'''import pandas as pd
|
17 |
+
import pyterrier as pt ; pt.init()
|
18 |
+
from pyterrier_t5 import MonoT5ReRanker
|
19 |
+
|
20 |
+
model = MonoT5ReRanker()
|
21 |
+
|
22 |
+
model({df2code(input)})
|
23 |
+
'''
|
24 |
+
res = model(input)
|
25 |
+
res['score'] = res['score'].map(lambda x: round(x, 4))
|
26 |
+
res = res.sort_values(['qid', 'rank'])
|
27 |
+
return (res, code2md(code, COLAB_INSTALL, COLAB_NAME, colab=False))
|
28 |
+
|
29 |
+
interface(
|
30 |
+
MarkdownFile('README.md'),
|
31 |
+
Demo(
|
32 |
+
predict,
|
33 |
+
EX_R,
|
34 |
+
[]
|
35 |
+
),
|
36 |
+
MarkdownFile('wrapup.md'),
|
37 |
+
).launch(share=False)
|
packages.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
openjdk-11-jdk
|
2 |
+
openjdk-11-jre-headless
|
3 |
+
openjdk-11-jre
|
4 |
+
openjdk-11-jre-headless
|
5 |
+
debianutils
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
git+https://github.com/seanmacavaney/pyterrier_gradio@v0.0.5
|
2 |
+
git+https://github.com/terrier-org/pyterrier
|
3 |
+
git+https://github.com/terrier-org/pyterrier_t5
|
4 |
+
ir_datasets
|
5 |
+
ir_measures
|
wrapup.md
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
### References & Credits
|
2 |
+
|
3 |
+
- Ronak Pradeep, Rodrigo Nogueira, and Jimmy Lin. [The Expando-Mono-Duo Design Pattern for Text Ranking withPretrained Sequence-to-Sequence Models.](https://arxiv.org/pdf/2101.05667.pdf)
|
4 |
+
- Craig Macdonald, Nicola Tonellotto, Sean MacAvaney, Iadh Ounis. [PyTerrier: Declarative Experimentation in Python from BM25 to Dense Retrieval](https://dl.acm.org/doi/abs/10.1145/3459637.3482013). CIKM 2021.
|