Antonio Cheong
commited on
Commit
•
4d7378e
1
Parent(s):
4f15858
structure
Browse files- CODE_OF_CONDUCT.md +4 -0
- CONTRIBUTING.md +59 -0
- LICENSE +175 -0
- NOTICE +1 -0
- evaluations.py +100 -0
- main.py +383 -0
- model.py +194 -0
- requirements.txt +11 -0
- run_inference.sh +17 -0
- run_training.sh +15 -0
- utils_data.py +228 -0
- utils_evaluate.py +108 -0
- utils_prompt.py +240 -0
- vision_features/mm-cot.png +0 -0
CODE_OF_CONDUCT.md
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
## Code of Conduct
|
2 |
+
This project has adopted the [Amazon Open Source Code of Conduct](https://aws.github.io/code-of-conduct).
|
3 |
+
For more information see the [Code of Conduct FAQ](https://aws.github.io/code-of-conduct-faq) or contact
|
4 |
+
opensource-codeofconduct@amazon.com with any additional questions or comments.
|
CONTRIBUTING.md
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Contributing Guidelines
|
2 |
+
|
3 |
+
Thank you for your interest in contributing to our project. Whether it's a bug report, new feature, correction, or additional
|
4 |
+
documentation, we greatly value feedback and contributions from our community.
|
5 |
+
|
6 |
+
Please read through this document before submitting any issues or pull requests to ensure we have all the necessary
|
7 |
+
information to effectively respond to your bug report or contribution.
|
8 |
+
|
9 |
+
|
10 |
+
## Reporting Bugs/Feature Requests
|
11 |
+
|
12 |
+
We welcome you to use the GitHub issue tracker to report bugs or suggest features.
|
13 |
+
|
14 |
+
When filing an issue, please check existing open, or recently closed, issues to make sure somebody else hasn't already
|
15 |
+
reported the issue. Please try to include as much information as you can. Details like these are incredibly useful:
|
16 |
+
|
17 |
+
* A reproducible test case or series of steps
|
18 |
+
* The version of our code being used
|
19 |
+
* Any modifications you've made relevant to the bug
|
20 |
+
* Anything unusual about your environment or deployment
|
21 |
+
|
22 |
+
|
23 |
+
## Contributing via Pull Requests
|
24 |
+
Contributions via pull requests are much appreciated. Before sending us a pull request, please ensure that:
|
25 |
+
|
26 |
+
1. You are working against the latest source on the *main* branch.
|
27 |
+
2. You check existing open, and recently merged, pull requests to make sure someone else hasn't addressed the problem already.
|
28 |
+
3. You open an issue to discuss any significant work - we would hate for your time to be wasted.
|
29 |
+
|
30 |
+
To send us a pull request, please:
|
31 |
+
|
32 |
+
1. Fork the repository.
|
33 |
+
2. Modify the source; please focus on the specific change you are contributing. If you also reformat all the code, it will be hard for us to focus on your change.
|
34 |
+
3. Ensure local tests pass.
|
35 |
+
4. Commit to your fork using clear commit messages.
|
36 |
+
5. Send us a pull request, answering any default questions in the pull request interface.
|
37 |
+
6. Pay attention to any automated CI failures reported in the pull request, and stay involved in the conversation.
|
38 |
+
|
39 |
+
GitHub provides additional document on [forking a repository](https://help.github.com/articles/fork-a-repo/) and
|
40 |
+
[creating a pull request](https://help.github.com/articles/creating-a-pull-request/).
|
41 |
+
|
42 |
+
|
43 |
+
## Finding contributions to work on
|
44 |
+
Looking at the existing issues is a great way to find something to contribute on. As our projects, by default, use the default GitHub issue labels (enhancement/bug/duplicate/help wanted/invalid/question/wontfix), looking at any 'help wanted' issues is a great place to start.
|
45 |
+
|
46 |
+
|
47 |
+
## Code of Conduct
|
48 |
+
This project has adopted the [Amazon Open Source Code of Conduct](https://aws.github.io/code-of-conduct).
|
49 |
+
For more information see the [Code of Conduct FAQ](https://aws.github.io/code-of-conduct-faq) or contact
|
50 |
+
opensource-codeofconduct@amazon.com with any additional questions or comments.
|
51 |
+
|
52 |
+
|
53 |
+
## Security issue notifications
|
54 |
+
If you discover a potential security issue in this project we ask that you notify AWS/Amazon Security via our [vulnerability reporting page](http://aws.amazon.com/security/vulnerability-reporting/). Please do **not** create a public github issue.
|
55 |
+
|
56 |
+
|
57 |
+
## Licensing
|
58 |
+
|
59 |
+
See the [LICENSE](LICENSE) file for our project's licensing. We will ask you to confirm the licensing of your contribution.
|
LICENSE
ADDED
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
Apache License
|
3 |
+
Version 2.0, January 2004
|
4 |
+
http://www.apache.org/licenses/
|
5 |
+
|
6 |
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
7 |
+
|
8 |
+
1. Definitions.
|
9 |
+
|
10 |
+
"License" shall mean the terms and conditions for use, reproduction,
|
11 |
+
and distribution as defined by Sections 1 through 9 of this document.
|
12 |
+
|
13 |
+
"Licensor" shall mean the copyright owner or entity authorized by
|
14 |
+
the copyright owner that is granting the License.
|
15 |
+
|
16 |
+
"Legal Entity" shall mean the union of the acting entity and all
|
17 |
+
other entities that control, are controlled by, or are under common
|
18 |
+
control with that entity. For the purposes of this definition,
|
19 |
+
"control" means (i) the power, direct or indirect, to cause the
|
20 |
+
direction or management of such entity, whether by contract or
|
21 |
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
22 |
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
23 |
+
|
24 |
+
"You" (or "Your") shall mean an individual or Legal Entity
|
25 |
+
exercising permissions granted by this License.
|
26 |
+
|
27 |
+
"Source" form shall mean the preferred form for making modifications,
|
28 |
+
including but not limited to software source code, documentation
|
29 |
+
source, and configuration files.
|
30 |
+
|
31 |
+
"Object" form shall mean any form resulting from mechanical
|
32 |
+
transformation or translation of a Source form, including but
|
33 |
+
not limited to compiled object code, generated documentation,
|
34 |
+
and conversions to other media types.
|
35 |
+
|
36 |
+
"Work" shall mean the work of authorship, whether in Source or
|
37 |
+
Object form, made available under the License, as indicated by a
|
38 |
+
copyright notice that is included in or attached to the work
|
39 |
+
(an example is provided in the Appendix below).
|
40 |
+
|
41 |
+
"Derivative Works" shall mean any work, whether in Source or Object
|
42 |
+
form, that is based on (or derived from) the Work and for which the
|
43 |
+
editorial revisions, annotations, elaborations, or other modifications
|
44 |
+
represent, as a whole, an original work of authorship. For the purposes
|
45 |
+
of this License, Derivative Works shall not include works that remain
|
46 |
+
separable from, or merely link (or bind by name) to the interfaces of,
|
47 |
+
the Work and Derivative Works thereof.
|
48 |
+
|
49 |
+
"Contribution" shall mean any work of authorship, including
|
50 |
+
the original version of the Work and any modifications or additions
|
51 |
+
to that Work or Derivative Works thereof, that is intentionally
|
52 |
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
53 |
+
or by an individual or Legal Entity authorized to submit on behalf of
|
54 |
+
the copyright owner. For the purposes of this definition, "submitted"
|
55 |
+
means any form of electronic, verbal, or written communication sent
|
56 |
+
to the Licensor or its representatives, including but not limited to
|
57 |
+
communication on electronic mailing lists, source code control systems,
|
58 |
+
and issue tracking systems that are managed by, or on behalf of, the
|
59 |
+
Licensor for the purpose of discussing and improving the Work, but
|
60 |
+
excluding communication that is conspicuously marked or otherwise
|
61 |
+
designated in writing by the copyright owner as "Not a Contribution."
|
62 |
+
|
63 |
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
64 |
+
on behalf of whom a Contribution has been received by Licensor and
|
65 |
+
subsequently incorporated within the Work.
|
66 |
+
|
67 |
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
68 |
+
this License, each Contributor hereby grants to You a perpetual,
|
69 |
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
70 |
+
copyright license to reproduce, prepare Derivative Works of,
|
71 |
+
publicly display, publicly perform, sublicense, and distribute the
|
72 |
+
Work and such Derivative Works in Source or Object form.
|
73 |
+
|
74 |
+
3. Grant of Patent License. Subject to the terms and conditions of
|
75 |
+
this License, each Contributor hereby grants to You a perpetual,
|
76 |
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
77 |
+
(except as stated in this section) patent license to make, have made,
|
78 |
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
79 |
+
where such license applies only to those patent claims licensable
|
80 |
+
by such Contributor that are necessarily infringed by their
|
81 |
+
Contribution(s) alone or by combination of their Contribution(s)
|
82 |
+
with the Work to which such Contribution(s) was submitted. If You
|
83 |
+
institute patent litigation against any entity (including a
|
84 |
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
85 |
+
or a Contribution incorporated within the Work constitutes direct
|
86 |
+
or contributory patent infringement, then any patent licenses
|
87 |
+
granted to You under this License for that Work shall terminate
|
88 |
+
as of the date such litigation is filed.
|
89 |
+
|
90 |
+
4. Redistribution. You may reproduce and distribute copies of the
|
91 |
+
Work or Derivative Works thereof in any medium, with or without
|
92 |
+
modifications, and in Source or Object form, provided that You
|
93 |
+
meet the following conditions:
|
94 |
+
|
95 |
+
(a) You must give any other recipients of the Work or
|
96 |
+
Derivative Works a copy of this License; and
|
97 |
+
|
98 |
+
(b) You must cause any modified files to carry prominent notices
|
99 |
+
stating that You changed the files; and
|
100 |
+
|
101 |
+
(c) You must retain, in the Source form of any Derivative Works
|
102 |
+
that You distribute, all copyright, patent, trademark, and
|
103 |
+
attribution notices from the Source form of the Work,
|
104 |
+
excluding those notices that do not pertain to any part of
|
105 |
+
the Derivative Works; and
|
106 |
+
|
107 |
+
(d) If the Work includes a "NOTICE" text file as part of its
|
108 |
+
distribution, then any Derivative Works that You distribute must
|
109 |
+
include a readable copy of the attribution notices contained
|
110 |
+
within such NOTICE file, excluding those notices that do not
|
111 |
+
pertain to any part of the Derivative Works, in at least one
|
112 |
+
of the following places: within a NOTICE text file distributed
|
113 |
+
as part of the Derivative Works; within the Source form or
|
114 |
+
documentation, if provided along with the Derivative Works; or,
|
115 |
+
within a display generated by the Derivative Works, if and
|
116 |
+
wherever such third-party notices normally appear. The contents
|
117 |
+
of the NOTICE file are for informational purposes only and
|
118 |
+
do not modify the License. You may add Your own attribution
|
119 |
+
notices within Derivative Works that You distribute, alongside
|
120 |
+
or as an addendum to the NOTICE text from the Work, provided
|
121 |
+
that such additional attribution notices cannot be construed
|
122 |
+
as modifying the License.
|
123 |
+
|
124 |
+
You may add Your own copyright statement to Your modifications and
|
125 |
+
may provide additional or different license terms and conditions
|
126 |
+
for use, reproduction, or distribution of Your modifications, or
|
127 |
+
for any such Derivative Works as a whole, provided Your use,
|
128 |
+
reproduction, and distribution of the Work otherwise complies with
|
129 |
+
the conditions stated in this License.
|
130 |
+
|
131 |
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
132 |
+
any Contribution intentionally submitted for inclusion in the Work
|
133 |
+
by You to the Licensor shall be under the terms and conditions of
|
134 |
+
this License, without any additional terms or conditions.
|
135 |
+
Notwithstanding the above, nothing herein shall supersede or modify
|
136 |
+
the terms of any separate license agreement you may have executed
|
137 |
+
with Licensor regarding such Contributions.
|
138 |
+
|
139 |
+
6. Trademarks. This License does not grant permission to use the trade
|
140 |
+
names, trademarks, service marks, or product names of the Licensor,
|
141 |
+
except as required for reasonable and customary use in describing the
|
142 |
+
origin of the Work and reproducing the content of the NOTICE file.
|
143 |
+
|
144 |
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
145 |
+
agreed to in writing, Licensor provides the Work (and each
|
146 |
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
147 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
148 |
+
implied, including, without limitation, any warranties or conditions
|
149 |
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
150 |
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
151 |
+
appropriateness of using or redistributing the Work and assume any
|
152 |
+
risks associated with Your exercise of permissions under this License.
|
153 |
+
|
154 |
+
8. Limitation of Liability. In no event and under no legal theory,
|
155 |
+
whether in tort (including negligence), contract, or otherwise,
|
156 |
+
unless required by applicable law (such as deliberate and grossly
|
157 |
+
negligent acts) or agreed to in writing, shall any Contributor be
|
158 |
+
liable to You for damages, including any direct, indirect, special,
|
159 |
+
incidental, or consequential damages of any character arising as a
|
160 |
+
result of this License or out of the use or inability to use the
|
161 |
+
Work (including but not limited to damages for loss of goodwill,
|
162 |
+
work stoppage, computer failure or malfunction, or any and all
|
163 |
+
other commercial damages or losses), even if such Contributor
|
164 |
+
has been advised of the possibility of such damages.
|
165 |
+
|
166 |
+
9. Accepting Warranty or Additional Liability. While redistributing
|
167 |
+
the Work or Derivative Works thereof, You may choose to offer,
|
168 |
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
169 |
+
or other liability obligations and/or rights consistent with this
|
170 |
+
License. However, in accepting such obligations, You may act only
|
171 |
+
on Your own behalf and on Your sole responsibility, not on behalf
|
172 |
+
of any other Contributor, and only if You agree to indemnify,
|
173 |
+
defend, and hold each Contributor harmless for any liability
|
174 |
+
incurred by, or claims asserted against, such Contributor by reason
|
175 |
+
of your accepting any such warranty or additional liability.
|
NOTICE
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
|
evaluations.py
ADDED
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
'''
|
2 |
+
Adapted from https://github.com/lupantech/ScienceQA
|
3 |
+
'''
|
4 |
+
|
5 |
+
import re
|
6 |
+
from rouge import Rouge
|
7 |
+
from nltk.translate.bleu_score import sentence_bleu
|
8 |
+
from sentence_transformers import util
|
9 |
+
|
10 |
+
########################
|
11 |
+
## BLEU
|
12 |
+
########################
|
13 |
+
def tokenize(text):
|
14 |
+
tokens = re.split(r'\s|\.', text)
|
15 |
+
tokens = [t for t in tokens if len(t) > 0]
|
16 |
+
return tokens
|
17 |
+
|
18 |
+
|
19 |
+
def bleu_score(reference, hypothesis, gram):
|
20 |
+
reference_tokens = tokenize(reference)
|
21 |
+
hypothesis_tokens = tokenize(hypothesis)
|
22 |
+
|
23 |
+
if gram == 1:
|
24 |
+
bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1., )) # BELU-1
|
25 |
+
elif gram == 2:
|
26 |
+
bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1. / 2., 1. / 2.)) # BELU-2
|
27 |
+
elif gram == 3:
|
28 |
+
bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1. / 3., 1. / 3., 1. / 3.)) # BELU-3
|
29 |
+
elif gram == 4:
|
30 |
+
bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1. / 4., 1. / 4., 1. / 4., 1. / 4.)) # BELU-4
|
31 |
+
|
32 |
+
return bleu
|
33 |
+
|
34 |
+
|
35 |
+
def caculate_bleu(results, data, gram):
|
36 |
+
bleus = []
|
37 |
+
for qid, output in results.items():
|
38 |
+
prediction = output
|
39 |
+
target = data[qid]
|
40 |
+
target = target.strip()
|
41 |
+
if target == "":
|
42 |
+
continue
|
43 |
+
bleu = bleu_score(target, prediction, gram)
|
44 |
+
bleus.append(bleu)
|
45 |
+
|
46 |
+
avg_bleu = sum(bleus) / len(bleus)
|
47 |
+
|
48 |
+
return avg_bleu
|
49 |
+
|
50 |
+
|
51 |
+
########################
|
52 |
+
## Rouge-L
|
53 |
+
########################
|
54 |
+
def score_rouge(str1, str2):
|
55 |
+
rouge = Rouge(metrics=["rouge-l"])
|
56 |
+
scores = rouge.get_scores(str1, str2, avg=True)
|
57 |
+
rouge_l = scores['rouge-l']['f']
|
58 |
+
return rouge_l
|
59 |
+
|
60 |
+
|
61 |
+
def caculate_rouge(results, data):
|
62 |
+
rouges = []
|
63 |
+
for qid, output in results.items():
|
64 |
+
prediction = output
|
65 |
+
target = data[qid]
|
66 |
+
target = target.strip()
|
67 |
+
if prediction == "":
|
68 |
+
continue
|
69 |
+
if target == "":
|
70 |
+
continue
|
71 |
+
rouge = score_rouge(target, prediction)
|
72 |
+
rouges.append(rouge)
|
73 |
+
|
74 |
+
avg_rouge = sum(rouges) / len(rouges)
|
75 |
+
return avg_rouge
|
76 |
+
|
77 |
+
|
78 |
+
########################
|
79 |
+
## Sentence Similarity
|
80 |
+
########################
|
81 |
+
def similariry_score(str1, str2, model):
|
82 |
+
# compute embedding for both lists
|
83 |
+
embedding_1 = model.encode(str1, convert_to_tensor=True)
|
84 |
+
embedding_2 = model.encode(str2, convert_to_tensor=True)
|
85 |
+
score = util.pytorch_cos_sim(embedding_1, embedding_2).item()
|
86 |
+
return score
|
87 |
+
|
88 |
+
|
89 |
+
def caculate_similariry(results, data, model):
|
90 |
+
scores = []
|
91 |
+
for qid, output in results.items():
|
92 |
+
prediction = output
|
93 |
+
target = data[qid]
|
94 |
+
target = target.strip()
|
95 |
+
|
96 |
+
score = similariry_score(target, prediction, model)
|
97 |
+
scores.append(score)
|
98 |
+
|
99 |
+
avg_score = sum(scores) / len(scores)
|
100 |
+
return avg_score
|
main.py
ADDED
@@ -0,0 +1,383 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import numpy as np
|
3 |
+
import torch
|
4 |
+
import os
|
5 |
+
import re
|
6 |
+
import json
|
7 |
+
import argparse
|
8 |
+
import random
|
9 |
+
from transformers import T5Tokenizer, DataCollatorForSeq2Seq, Seq2SeqTrainingArguments, Seq2SeqTrainer, T5ForConditionalGeneration
|
10 |
+
from model import T5ForConditionalGeneration, T5ForMultimodalGeneration
|
11 |
+
from utils_data import img_shape, load_data_std, load_data_img, ScienceQADatasetStd, ScienceQADatasetImg
|
12 |
+
from utils_prompt import *
|
13 |
+
from utils_evaluate import get_scores
|
14 |
+
from rich.table import Column, Table
|
15 |
+
from rich import box
|
16 |
+
from rich.console import Console
|
17 |
+
console = Console(record=True)
|
18 |
+
from torch import cuda
|
19 |
+
import nltk
|
20 |
+
import evaluate
|
21 |
+
|
22 |
+
|
23 |
+
def parse_args():
|
24 |
+
parser = argparse.ArgumentParser()
|
25 |
+
parser.add_argument('--data_root', type=str, default='data')
|
26 |
+
parser.add_argument('--output_dir', type=str, default='experiments')
|
27 |
+
parser.add_argument('--model', type=str, default='allenai/unifiedqa-t5-base')
|
28 |
+
parser.add_argument('--options', type=list, default=["A", "B", "C", "D", "E"])
|
29 |
+
parser.add_argument('--epoch', type=int, default=20)
|
30 |
+
parser.add_argument('--lr', type=float, default=5e-5)
|
31 |
+
parser.add_argument('--bs', type=int, default=16)
|
32 |
+
parser.add_argument('--input_len', type=int, default=512)
|
33 |
+
parser.add_argument('--output_len', type=int, default=64)
|
34 |
+
parser.add_argument('--eval_bs', type=int, default=16)
|
35 |
+
parser.add_argument('--eval_acc', type=int, default=None, help='evaluate accumulation step')
|
36 |
+
parser.add_argument('--train_split', type=str, default='train', choices=['train', 'trainval', 'minitrain'])
|
37 |
+
parser.add_argument('--val_split', type=str, default='val', choices=['test', 'val', 'minival'])
|
38 |
+
parser.add_argument('--test_split', type=str, default='test', choices=['test', 'minitest'])
|
39 |
+
|
40 |
+
parser.add_argument('--use_generate', action='store_true', help='only for baseline to improve inference speed')
|
41 |
+
parser.add_argument('--final_eval', action='store_true', help='only evaluate the model at the final epoch')
|
42 |
+
parser.add_argument('--user_msg', type=str, default="baseline", help='experiment type in the save_dir')
|
43 |
+
parser.add_argument('--img_type', type=str, default=None, choices=['detr', 'clip', 'resnet'], help='type of image features')
|
44 |
+
parser.add_argument('--eval_le', type=str, default=None, help='generated rationale for the dev set')
|
45 |
+
parser.add_argument('--test_le', type=str, default=None, help='generated rationale for the test set')
|
46 |
+
parser.add_argument('--evaluate_dir', type=str, default=None, help='the directory of model for evaluation')
|
47 |
+
parser.add_argument('--caption_file', type=str, default='data/captions.json')
|
48 |
+
parser.add_argument('--use_caption', action='store_true', help='use image captions or not')
|
49 |
+
parser.add_argument('--prompt_format', type=str, default='QCM-A', help='prompt format template',
|
50 |
+
choices=['QCM-A', 'QCM-LE', 'QCMG-A', 'QCM-LEA', 'QCM-ALE'])
|
51 |
+
parser.add_argument('--seed', type=int, default=42, help='random seed')
|
52 |
+
|
53 |
+
args = parser.parse_args()
|
54 |
+
return args
|
55 |
+
|
56 |
+
def T5Trainer(
|
57 |
+
dataframe, args,
|
58 |
+
):
|
59 |
+
torch.manual_seed(args.seed) # pytorch random seed
|
60 |
+
np.random.seed(args.seed) # numpy random seed
|
61 |
+
torch.backends.cudnn.deterministic = True
|
62 |
+
|
63 |
+
if args.evaluate_dir is not None:
|
64 |
+
args.model = args.evaluate_dir
|
65 |
+
|
66 |
+
tokenizer = T5Tokenizer.from_pretrained(args.model)
|
67 |
+
|
68 |
+
console.log(f"""[Model]: Loading {args.model}...\n""")
|
69 |
+
console.log(f"[Data]: Reading data...\n")
|
70 |
+
problems = dataframe['problems']
|
71 |
+
qids = dataframe['qids']
|
72 |
+
train_qids = qids['train']
|
73 |
+
test_qids = qids['test']
|
74 |
+
val_qids = qids['val']
|
75 |
+
|
76 |
+
if args.evaluate_dir is not None:
|
77 |
+
save_dir = args.evaluate_dir
|
78 |
+
else:
|
79 |
+
model_name = args.model.replace("/","-")
|
80 |
+
gpu_count = torch.cuda.device_count()
|
81 |
+
save_dir = f"{args.output_dir}/{args.user_msg}_{model_name}_{args.img_type}_{args.prompt_format}_lr{args.lr}_bs{args.bs * gpu_count}_op{args.output_len}_ep{args.epoch}"
|
82 |
+
if not os.path.exists(save_dir):
|
83 |
+
os.mkdir(save_dir)
|
84 |
+
|
85 |
+
padding_idx = tokenizer._convert_token_to_id(tokenizer.pad_token)
|
86 |
+
if args.img_type is not None:
|
87 |
+
patch_size = img_shape[args.img_type]
|
88 |
+
model = T5ForMultimodalGeneration.from_pretrained(args.model, patch_size=patch_size, padding_idx=padding_idx, save_dir=save_dir)
|
89 |
+
name_maps = dataframe['name_maps']
|
90 |
+
image_features = dataframe['image_features']
|
91 |
+
train_set = ScienceQADatasetImg(
|
92 |
+
problems,
|
93 |
+
train_qids,
|
94 |
+
name_maps,
|
95 |
+
tokenizer,
|
96 |
+
args.input_len,
|
97 |
+
args.output_len,
|
98 |
+
args,
|
99 |
+
image_features,
|
100 |
+
)
|
101 |
+
eval_set = ScienceQADatasetImg(
|
102 |
+
problems,
|
103 |
+
val_qids,
|
104 |
+
name_maps,
|
105 |
+
tokenizer,
|
106 |
+
args.input_len,
|
107 |
+
args.output_len,
|
108 |
+
args,
|
109 |
+
image_features,
|
110 |
+
args.eval_le,
|
111 |
+
)
|
112 |
+
test_set = ScienceQADatasetImg(
|
113 |
+
problems,
|
114 |
+
test_qids,
|
115 |
+
name_maps,
|
116 |
+
tokenizer,
|
117 |
+
args.input_len,
|
118 |
+
args.output_len,
|
119 |
+
args,
|
120 |
+
image_features,
|
121 |
+
args.test_le,
|
122 |
+
)
|
123 |
+
else:
|
124 |
+
model = T5ForConditionalGeneration.from_pretrained(args.model)
|
125 |
+
train_set = ScienceQADatasetStd(
|
126 |
+
problems,
|
127 |
+
train_qids,
|
128 |
+
tokenizer,
|
129 |
+
args.input_len,
|
130 |
+
args.output_len,
|
131 |
+
args,
|
132 |
+
)
|
133 |
+
eval_set = ScienceQADatasetStd(
|
134 |
+
problems,
|
135 |
+
val_qids,
|
136 |
+
tokenizer,
|
137 |
+
args.input_len,
|
138 |
+
args.output_len,
|
139 |
+
args,
|
140 |
+
args.eval_le,
|
141 |
+
)
|
142 |
+
|
143 |
+
test_set = ScienceQADatasetStd(
|
144 |
+
problems,
|
145 |
+
test_qids,
|
146 |
+
tokenizer,
|
147 |
+
args.input_len,
|
148 |
+
args.output_len,
|
149 |
+
args,
|
150 |
+
args.test_le,
|
151 |
+
)
|
152 |
+
|
153 |
+
datacollator = DataCollatorForSeq2Seq(tokenizer)
|
154 |
+
print("model parameters: ", model.num_parameters())
|
155 |
+
def extract_ans(ans):
|
156 |
+
pattern = re.compile(r'The answer is \(([A-Z])\)')
|
157 |
+
res = pattern.findall(ans)
|
158 |
+
|
159 |
+
if len(res) == 1:
|
160 |
+
answer = res[0] # 'A', 'B', ...
|
161 |
+
else:
|
162 |
+
answer = "FAILED"
|
163 |
+
return answer
|
164 |
+
|
165 |
+
# accuracy for answer inference
|
166 |
+
def compute_metrics_acc(eval_preds):
|
167 |
+
if args.use_generate:
|
168 |
+
preds, targets = eval_preds
|
169 |
+
if isinstance(preds, tuple):
|
170 |
+
preds = preds[0]
|
171 |
+
else:
|
172 |
+
preds = eval_preds.predictions[0]
|
173 |
+
targets = eval_preds.label_ids
|
174 |
+
preds = preds.argmax(axis=2)
|
175 |
+
preds = tokenizer.batch_decode(preds, skip_special_tokens=True, clean_up_tokenization_spaces=True)
|
176 |
+
targets = tokenizer.batch_decode(targets, skip_special_tokens=True, clean_up_tokenization_spaces=True)
|
177 |
+
correct = 0
|
178 |
+
assert len(preds) == len(targets)
|
179 |
+
for idx, pred in enumerate(preds):
|
180 |
+
reference = targets[idx]
|
181 |
+
reference = extract_ans(reference)
|
182 |
+
extract_pred = extract_ans(pred)
|
183 |
+
best_option = extract_pred
|
184 |
+
if reference == best_option:
|
185 |
+
correct +=1
|
186 |
+
return {'accuracy': 1.0*correct/len(targets)}
|
187 |
+
|
188 |
+
# rougel for rationale generation
|
189 |
+
metric = evaluate.load("rouge")
|
190 |
+
def postprocess_text(preds, labels):
|
191 |
+
preds = [pred.strip() for pred in preds]
|
192 |
+
labels = [label.strip() for label in labels]
|
193 |
+
preds = ["\n".join(nltk.sent_tokenize(pred)) for pred in preds]
|
194 |
+
labels = ["\n".join(nltk.sent_tokenize(label)) for label in labels]
|
195 |
+
return preds, labels
|
196 |
+
|
197 |
+
def compute_metrics_rougel(eval_preds):
|
198 |
+
if args.use_generate:
|
199 |
+
preds, targets = eval_preds
|
200 |
+
if isinstance(preds, tuple):
|
201 |
+
preds = preds[0]
|
202 |
+
else:
|
203 |
+
preds = eval_preds.predictions[0]
|
204 |
+
targets = eval_preds.label_ids
|
205 |
+
preds = preds.argmax(axis=2)
|
206 |
+
preds = tokenizer.batch_decode(preds, skip_special_tokens=True, clean_up_tokenization_spaces=True)
|
207 |
+
targets = tokenizer.batch_decode(targets, skip_special_tokens=True, clean_up_tokenization_spaces=True)
|
208 |
+
|
209 |
+
decoded_preds, decoded_labels = postprocess_text(preds, targets)
|
210 |
+
|
211 |
+
result = metric.compute(predictions=decoded_preds, references=decoded_labels, use_stemmer=True)
|
212 |
+
result = {k: round(v * 100, 4) for k, v in result.items()}
|
213 |
+
prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
|
214 |
+
result["gen_len"] = np.mean(prediction_lens)
|
215 |
+
return result
|
216 |
+
|
217 |
+
# only use the last model for evaluation to save time
|
218 |
+
if args.final_eval:
|
219 |
+
training_args = Seq2SeqTrainingArguments(
|
220 |
+
save_dir,
|
221 |
+
do_train=True if args.evaluate_dir is None else False,
|
222 |
+
do_eval=False,
|
223 |
+
evaluation_strategy="no",
|
224 |
+
logging_strategy="steps",
|
225 |
+
save_strategy="epoch",
|
226 |
+
save_total_limit = 2,
|
227 |
+
learning_rate= args.lr,
|
228 |
+
eval_accumulation_steps=args.eval_acc,
|
229 |
+
per_device_train_batch_size=args.bs,
|
230 |
+
per_device_eval_batch_size=args.eval_bs,
|
231 |
+
weight_decay=0.01,
|
232 |
+
num_train_epochs=args.epoch,
|
233 |
+
predict_with_generate=args.use_generate,
|
234 |
+
report_to="none",
|
235 |
+
)
|
236 |
+
# evaluate at each epoch
|
237 |
+
else:
|
238 |
+
training_args = Seq2SeqTrainingArguments(
|
239 |
+
save_dir,
|
240 |
+
do_train=True if args.evaluate_dir is None else False,
|
241 |
+
do_eval=True,
|
242 |
+
evaluation_strategy="epoch",
|
243 |
+
logging_strategy="steps",
|
244 |
+
save_strategy="epoch",
|
245 |
+
save_total_limit = 2,
|
246 |
+
learning_rate= args.lr,
|
247 |
+
eval_accumulation_steps=args.eval_acc,
|
248 |
+
per_device_train_batch_size=args.bs,
|
249 |
+
per_device_eval_batch_size=args.eval_bs,
|
250 |
+
weight_decay=0.01,
|
251 |
+
num_train_epochs=args.epoch,
|
252 |
+
metric_for_best_model="accuracy" if args.prompt_format != "QCM-LE" else "rougeL",
|
253 |
+
predict_with_generate=args.use_generate,
|
254 |
+
load_best_model_at_end=True,
|
255 |
+
report_to="none",
|
256 |
+
)
|
257 |
+
|
258 |
+
trainer = Seq2SeqTrainer(
|
259 |
+
model=model,
|
260 |
+
args=training_args,
|
261 |
+
train_dataset=train_set,
|
262 |
+
eval_dataset=eval_set,
|
263 |
+
data_collator=datacollator,
|
264 |
+
tokenizer=tokenizer,
|
265 |
+
compute_metrics = compute_metrics_acc if args.prompt_format != "QCM-LE" else compute_metrics_rougel
|
266 |
+
)
|
267 |
+
|
268 |
+
if args.evaluate_dir is None:
|
269 |
+
trainer.train()
|
270 |
+
trainer.save_model(save_dir)
|
271 |
+
|
272 |
+
metrics = trainer.evaluate(eval_dataset = test_set)
|
273 |
+
trainer.log_metrics("test", metrics)
|
274 |
+
trainer.save_metrics("test", metrics)
|
275 |
+
|
276 |
+
predict_results = trainer.predict(test_dataset=test_set, max_length=args.output_len)
|
277 |
+
if trainer.is_world_process_zero():
|
278 |
+
if args.use_generate:
|
279 |
+
preds, targets = predict_results.predictions, predict_results.label_ids
|
280 |
+
else:
|
281 |
+
preds = predict_results.predictions[0]
|
282 |
+
targets = predict_results.label_ids
|
283 |
+
preds = preds.argmax(axis=2)
|
284 |
+
|
285 |
+
preds = tokenizer.batch_decode(
|
286 |
+
preds, skip_special_tokens=True, clean_up_tokenization_spaces=True
|
287 |
+
)
|
288 |
+
targets = tokenizer.batch_decode(
|
289 |
+
targets, skip_special_tokens=True, clean_up_tokenization_spaces=True
|
290 |
+
)
|
291 |
+
|
292 |
+
results_ans = {}
|
293 |
+
results_rationale = {}
|
294 |
+
results_reference = {}
|
295 |
+
|
296 |
+
num_fail = 0
|
297 |
+
for idx, qid in enumerate(test_qids):
|
298 |
+
pred = preds[int(idx)]
|
299 |
+
ref = targets[int(idx)]
|
300 |
+
extract_pred = extract_ans(pred)
|
301 |
+
if extract_pred != "FAILED":
|
302 |
+
if extract_pred in args.options:
|
303 |
+
extract_pred = args.options.index(extract_pred)
|
304 |
+
else:
|
305 |
+
extract_pred = random.choice(range(0,len(args.options)))
|
306 |
+
else:
|
307 |
+
num_fail += 1
|
308 |
+
extract_pred = random.choice(range(len(args.options))) # random choose one option
|
309 |
+
results_ans[str(qid)] = extract_pred
|
310 |
+
results_rationale[str(qid)] = pred
|
311 |
+
results_reference[str(qid)] = ref
|
312 |
+
|
313 |
+
scores = get_scores(results_ans, results_rationale, results_reference, os.path.join(args.data_root, "scienceqa/problems.json"))
|
314 |
+
preds = [pred.strip() for pred in preds]
|
315 |
+
output_data = {
|
316 |
+
"num_fail": num_fail,
|
317 |
+
"scores": scores,
|
318 |
+
"preds": preds,
|
319 |
+
"labels": targets}
|
320 |
+
output_prediction_file = os.path.join(save_dir,"predictions_ans_test.json")
|
321 |
+
with open(output_prediction_file, "w") as writer:
|
322 |
+
writer.write(json.dumps(output_data, indent=4))
|
323 |
+
|
324 |
+
# generate the rationale for the eval set
|
325 |
+
if args.prompt_format == "QCM-LE":
|
326 |
+
torch.cuda.empty_cache()
|
327 |
+
del predict_results, preds, targets
|
328 |
+
predict_results = trainer.predict(test_dataset=eval_set, max_length=args.output_len)
|
329 |
+
if trainer.is_world_process_zero():
|
330 |
+
if args.use_generate:
|
331 |
+
preds, targets = predict_results.predictions, predict_results.label_ids
|
332 |
+
else:
|
333 |
+
preds = predict_results.predictions[0]
|
334 |
+
targets = predict_results.label_ids
|
335 |
+
preds = preds.argmax(axis=2)
|
336 |
+
|
337 |
+
preds = tokenizer.batch_decode(
|
338 |
+
preds, skip_special_tokens=True, clean_up_tokenization_spaces=True
|
339 |
+
)
|
340 |
+
targets = tokenizer.batch_decode(
|
341 |
+
targets, skip_special_tokens=True, clean_up_tokenization_spaces=True
|
342 |
+
)
|
343 |
+
preds = [pred.strip() for pred in preds]
|
344 |
+
output_data = {"preds": preds,
|
345 |
+
"labels": targets}
|
346 |
+
output_prediction_file = os.path.join(save_dir,"predictions_ans_eval.json")
|
347 |
+
with open(output_prediction_file, "w") as writer:
|
348 |
+
writer.write(json.dumps(output_data, indent=4))
|
349 |
+
|
350 |
+
|
351 |
+
if __name__ == '__main__':
|
352 |
+
|
353 |
+
# training logger to log training progress
|
354 |
+
training_logger = Table(
|
355 |
+
Column("Epoch", justify="center"),
|
356 |
+
Column("Steps", justify="center"),
|
357 |
+
Column("Loss", justify="center"),
|
358 |
+
title="Training Status",
|
359 |
+
pad_edge=False,
|
360 |
+
box=box.ASCII,
|
361 |
+
)
|
362 |
+
|
363 |
+
args = parse_args()
|
364 |
+
print("args",args)
|
365 |
+
print('====Input Arguments====')
|
366 |
+
print(json.dumps(vars(args), indent=2, sort_keys=False))
|
367 |
+
|
368 |
+
random.seed(args.seed)
|
369 |
+
|
370 |
+
if not os.path.exists(args.output_dir):
|
371 |
+
os.mkdir(args.output_dir)
|
372 |
+
|
373 |
+
if args.img_type is not None:
|
374 |
+
problems, qids, name_maps, image_features = load_data_img(args) # probelms, test question ids, shot example ids
|
375 |
+
dataframe = {'problems':problems, 'qids':qids, 'name_maps': name_maps, 'image_features': image_features}
|
376 |
+
else:
|
377 |
+
problems, qids = load_data_std(args) # probelms, test question ids, shot example ids
|
378 |
+
dataframe = {'problems':problems, 'qids':qids}
|
379 |
+
|
380 |
+
T5Trainer(
|
381 |
+
dataframe=dataframe,
|
382 |
+
args = args
|
383 |
+
)
|
model.py
ADDED
@@ -0,0 +1,194 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
'''
|
2 |
+
Adapted from https://github.com/huggingface/transformers
|
3 |
+
'''
|
4 |
+
|
5 |
+
from transformers import T5Config, T5ForConditionalGeneration
|
6 |
+
from transformers.models.t5.modeling_t5 import T5Stack, __HEAD_MASK_WARNING_MSG, T5EncoderModel
|
7 |
+
import copy
|
8 |
+
import math
|
9 |
+
import os
|
10 |
+
import warnings
|
11 |
+
from typing import Optional, Tuple, Union
|
12 |
+
import torch
|
13 |
+
from torch import nn
|
14 |
+
from torch.nn import CrossEntropyLoss
|
15 |
+
from transformers.modeling_outputs import (
|
16 |
+
BaseModelOutput,
|
17 |
+
Seq2SeqLMOutput,
|
18 |
+
)
|
19 |
+
|
20 |
+
class T5ForMultimodalGeneration(T5ForConditionalGeneration):
|
21 |
+
_keys_to_ignore_on_load_missing = [
|
22 |
+
r"encoder.embed_tokens.weight",
|
23 |
+
r"decoder.embed_tokens.weight",
|
24 |
+
r"lm_head.weight",
|
25 |
+
]
|
26 |
+
_keys_to_ignore_on_load_unexpected = [
|
27 |
+
r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
|
28 |
+
]
|
29 |
+
|
30 |
+
def __init__(self, config: T5Config, patch_size, padding_idx, save_dir):
|
31 |
+
super().__init__(config)
|
32 |
+
self.model_dim = config.d_model
|
33 |
+
|
34 |
+
self.padding_idx = padding_idx
|
35 |
+
self.out = open(os.path.join(save_dir, 'gate.txt'), 'w')
|
36 |
+
|
37 |
+
self.shared = nn.Embedding(config.vocab_size, config.d_model)
|
38 |
+
self.patch_num, self.patch_dim = patch_size
|
39 |
+
|
40 |
+
self.image_dense = nn.Linear(self.patch_dim, config.d_model)
|
41 |
+
self.mha_layer = torch.nn.MultiheadAttention(embed_dim=config.hidden_size, kdim=config.hidden_size, vdim=config.hidden_size, num_heads=1, batch_first=True)
|
42 |
+
self.gate_dense = nn.Linear(2*config.hidden_size, config.hidden_size)
|
43 |
+
self.sigmoid = nn.Sigmoid()
|
44 |
+
|
45 |
+
encoder_config = copy.deepcopy(config)
|
46 |
+
encoder_config.is_decoder = False
|
47 |
+
encoder_config.use_cache = False
|
48 |
+
encoder_config.is_encoder_decoder = False
|
49 |
+
self.encoder = T5Stack(encoder_config, self.shared)
|
50 |
+
|
51 |
+
decoder_config = copy.deepcopy(config)
|
52 |
+
decoder_config.is_decoder = True
|
53 |
+
decoder_config.is_encoder_decoder = False
|
54 |
+
decoder_config.num_layers = config.num_decoder_layers
|
55 |
+
self.decoder = T5Stack(decoder_config, self.shared)
|
56 |
+
|
57 |
+
self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False)
|
58 |
+
|
59 |
+
# Initialize weights and apply final processing
|
60 |
+
self.post_init()
|
61 |
+
|
62 |
+
# Model parallel
|
63 |
+
self.model_parallel = False
|
64 |
+
self.device_map = None
|
65 |
+
|
66 |
+
def forward(
|
67 |
+
self,
|
68 |
+
input_ids: Optional[torch.LongTensor] = None,
|
69 |
+
image_ids=None,
|
70 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
71 |
+
decoder_input_ids: Optional[torch.LongTensor] = None,
|
72 |
+
decoder_attention_mask: Optional[torch.BoolTensor] = None,
|
73 |
+
head_mask: Optional[torch.FloatTensor] = None,
|
74 |
+
decoder_head_mask: Optional[torch.FloatTensor] = None,
|
75 |
+
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
76 |
+
encoder_outputs: Optional[Tuple[Tuple[torch.Tensor]]] = None,
|
77 |
+
past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
|
78 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
79 |
+
decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
|
80 |
+
labels: Optional[torch.LongTensor] = None,
|
81 |
+
use_cache: Optional[bool] = None,
|
82 |
+
output_attentions: Optional[bool] = None,
|
83 |
+
output_hidden_states: Optional[bool] = None,
|
84 |
+
return_dict: Optional[bool] = None,
|
85 |
+
) -> Union[Tuple[torch.FloatTensor], Seq2SeqLMOutput]:
|
86 |
+
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
87 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
88 |
+
|
89 |
+
# FutureWarning: head_mask was separated into two input args - head_mask, decoder_head_mask
|
90 |
+
if head_mask is not None and decoder_head_mask is None:
|
91 |
+
if self.config.num_layers == self.config.num_decoder_layers:
|
92 |
+
warnings.warn(__HEAD_MASK_WARNING_MSG, FutureWarning)
|
93 |
+
decoder_head_mask = head_mask
|
94 |
+
|
95 |
+
# Encode if needed (training, first prediction pass)
|
96 |
+
if encoder_outputs is None:
|
97 |
+
# Convert encoder inputs in embeddings if needed
|
98 |
+
encoder_outputs = self.encoder(
|
99 |
+
input_ids=input_ids,
|
100 |
+
attention_mask=attention_mask,
|
101 |
+
inputs_embeds=inputs_embeds,
|
102 |
+
head_mask=head_mask,
|
103 |
+
output_attentions=output_attentions,
|
104 |
+
output_hidden_states=output_hidden_states,
|
105 |
+
return_dict=return_dict,
|
106 |
+
)
|
107 |
+
|
108 |
+
elif return_dict and not isinstance(encoder_outputs, BaseModelOutput):
|
109 |
+
encoder_outputs = BaseModelOutput(
|
110 |
+
last_hidden_state=encoder_outputs[0],
|
111 |
+
hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
|
112 |
+
attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
|
113 |
+
)
|
114 |
+
|
115 |
+
|
116 |
+
hidden_states = encoder_outputs[0]
|
117 |
+
|
118 |
+
image_embedding = self.image_dense(image_ids)
|
119 |
+
image_att, _ = self.mha_layer(hidden_states, image_embedding, image_embedding)
|
120 |
+
|
121 |
+
merge = torch.cat([hidden_states, image_att], dim=-1)
|
122 |
+
gate = self.sigmoid(self.gate_dense(merge))
|
123 |
+
hidden_states = (1 - gate) * hidden_states + gate * image_att
|
124 |
+
|
125 |
+
if self.model_parallel:
|
126 |
+
torch.cuda.set_device(self.decoder.first_device)
|
127 |
+
|
128 |
+
if labels is not None and decoder_input_ids is None and decoder_inputs_embeds is None:
|
129 |
+
# get decoder inputs from shifting lm labels to the right
|
130 |
+
decoder_input_ids = self._shift_right(labels)
|
131 |
+
|
132 |
+
# Set device for model parallelism
|
133 |
+
if self.model_parallel:
|
134 |
+
torch.cuda.set_device(self.decoder.first_device)
|
135 |
+
hidden_states = hidden_states.to(self.decoder.first_device)
|
136 |
+
if decoder_input_ids is not None:
|
137 |
+
decoder_input_ids = decoder_input_ids.to(self.decoder.first_device)
|
138 |
+
if attention_mask is not None:
|
139 |
+
attention_mask = attention_mask.to(self.decoder.first_device)
|
140 |
+
if decoder_attention_mask is not None:
|
141 |
+
decoder_attention_mask = decoder_attention_mask.to(self.decoder.first_device)
|
142 |
+
|
143 |
+
# Decode
|
144 |
+
decoder_outputs = self.decoder(
|
145 |
+
input_ids=decoder_input_ids,
|
146 |
+
attention_mask=decoder_attention_mask,
|
147 |
+
inputs_embeds=decoder_inputs_embeds,
|
148 |
+
past_key_values=past_key_values,
|
149 |
+
encoder_hidden_states=hidden_states,
|
150 |
+
encoder_attention_mask=attention_mask,
|
151 |
+
head_mask=decoder_head_mask,
|
152 |
+
cross_attn_head_mask=cross_attn_head_mask,
|
153 |
+
use_cache=use_cache,
|
154 |
+
output_attentions=output_attentions,
|
155 |
+
output_hidden_states=output_hidden_states,
|
156 |
+
return_dict=return_dict,
|
157 |
+
)
|
158 |
+
|
159 |
+
sequence_output = decoder_outputs[0]
|
160 |
+
|
161 |
+
# Set device for model parallelism
|
162 |
+
if self.model_parallel:
|
163 |
+
torch.cuda.set_device(self.encoder.first_device)
|
164 |
+
self.lm_head = self.lm_head.to(self.encoder.first_device)
|
165 |
+
sequence_output = sequence_output.to(self.lm_head.weight.device)
|
166 |
+
|
167 |
+
if self.config.tie_word_embeddings:
|
168 |
+
# Rescale output before projecting on vocab
|
169 |
+
# See https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/transformer/transformer.py#L586
|
170 |
+
sequence_output = sequence_output * (self.model_dim**-0.5)
|
171 |
+
|
172 |
+
lm_logits = self.lm_head(sequence_output)
|
173 |
+
|
174 |
+
loss = None
|
175 |
+
if labels is not None:
|
176 |
+
loss_fct = CrossEntropyLoss(ignore_index=-100)
|
177 |
+
loss = loss_fct(lm_logits.view(-1, lm_logits.size(-1)), labels.view(-1))
|
178 |
+
# TODO(thom): Add z_loss https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/layers.py#L666
|
179 |
+
|
180 |
+
if not return_dict:
|
181 |
+
output = (lm_logits,) + decoder_outputs[1:] + encoder_outputs
|
182 |
+
return ((loss,) + output) if loss is not None else output
|
183 |
+
|
184 |
+
return Seq2SeqLMOutput(
|
185 |
+
loss=loss,
|
186 |
+
logits=lm_logits,
|
187 |
+
past_key_values=decoder_outputs.past_key_values,
|
188 |
+
decoder_hidden_states=decoder_outputs.hidden_states,
|
189 |
+
decoder_attentions=decoder_outputs.attentions,
|
190 |
+
cross_attentions=decoder_outputs.cross_attentions,
|
191 |
+
encoder_last_hidden_state=encoder_outputs.last_hidden_state,
|
192 |
+
encoder_hidden_states=encoder_outputs.hidden_states,
|
193 |
+
encoder_attentions=encoder_outputs.attentions,
|
194 |
+
)
|
requirements.txt
ADDED
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
huggingface-hub==0.0.12
|
2 |
+
numpy==1.23.2
|
3 |
+
openai==0.23.0
|
4 |
+
pandas==1.4.3
|
5 |
+
rouge==1.0.1
|
6 |
+
sentence-transformers==2.2.2
|
7 |
+
transformers==4.21.1
|
8 |
+
nltk==3.6.6
|
9 |
+
evaluate==0.4.0
|
10 |
+
rouge==1.0.1
|
11 |
+
rouge_score==0.1.2
|
run_inference.sh
ADDED
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# rationale generation
|
2 |
+
CUDA_VISIBLE_DEVICES=0,1 python main.py \
|
3 |
+
--model allenai/unifiedqa-t5-base \
|
4 |
+
--user_msg rationale --img_type detr \
|
5 |
+
--bs 8 --eval_bs 4 --eval_acc 10 --output_len 512 \
|
6 |
+
--final_eval --prompt_format QCM-LE \
|
7 |
+
--evaluate_dir models/rationale
|
8 |
+
|
9 |
+
# answer inference
|
10 |
+
CUDA_VISIBLE_DEVICES=0,1 python main.py \
|
11 |
+
--model allenai/unifiedqa-t5-base \
|
12 |
+
--user_msg answer --img_type detr \
|
13 |
+
--bs 8 --eval_bs 4 --eval_acc 10 --output_len 64 \
|
14 |
+
--final_eval --prompt_format QCMG-A \
|
15 |
+
--eval_le models/rationale/predictions_ans_eval.json \
|
16 |
+
--test_le models/rationale/predictions_ans_test.json \
|
17 |
+
--evaluate_dir models/answer
|
run_training.sh
ADDED
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# rationale generation
|
2 |
+
CUDA_VISIBLE_DEVICES=0,1 python main.py \
|
3 |
+
--model allenai/unifiedqa-t5-base \
|
4 |
+
--user_msg rationale --img_type detr \
|
5 |
+
--bs 8 --eval_bs 4 --eval_acc 10 --output_len 512 \
|
6 |
+
--final_eval --prompt_format QCM-LE
|
7 |
+
|
8 |
+
# answer inference
|
9 |
+
CUDA_VISIBLE_DEVICES=0,1 python main.py \
|
10 |
+
--model allenai/unifiedqa-t5-base \
|
11 |
+
--user_msg answer --img_type detr \
|
12 |
+
--bs 8 --eval_bs 4 --eval_acc 10 --output_len 64 \
|
13 |
+
--final_eval --prompt_format QCMG-A \
|
14 |
+
--eval_le experiments/rationale_allenai-unifiedqa-t5-base_detr_QCM-LE_lr5e-05_bs16_op512_ep20/predictions_ans_eval.json \
|
15 |
+
--test_le experiments/rationale_allenai-unifiedqa-t5-base_detr_QCM-LE_lr5e-05_bs16_op512_ep20/predictions_ans_test.json
|
utils_data.py
ADDED
@@ -0,0 +1,228 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from torch.utils.data import Dataset
|
3 |
+
import os
|
4 |
+
import json
|
5 |
+
import numpy as np
|
6 |
+
import torch
|
7 |
+
from utils_prompt import *
|
8 |
+
|
9 |
+
img_shape = {
|
10 |
+
"resnet": (512, 2048),
|
11 |
+
"clip": (49, 2048),
|
12 |
+
"detr": (100, 256),
|
13 |
+
}
|
14 |
+
|
15 |
+
def load_data_std(args):
|
16 |
+
problems = json.load(open(os.path.join(args.data_root, 'scienceqa/problems.json')))
|
17 |
+
pid_splits = json.load(open(os.path.join(args.data_root, 'scienceqa/pid_splits.json')))
|
18 |
+
captions = json.load(open(args.caption_file))["captions"]
|
19 |
+
|
20 |
+
for qid in problems:
|
21 |
+
problems[qid]['caption'] = captions[qid] if qid in captions else ""
|
22 |
+
|
23 |
+
train_qids = pid_splits['%s' % (args.train_split)]
|
24 |
+
val_qids = pid_splits['%s' % (args.val_split)]
|
25 |
+
test_qids = pid_splits['%s' % (args.test_split)]
|
26 |
+
print(f"number of train problems: {len(train_qids)}\n")
|
27 |
+
print(f"number of val problems: {len(val_qids)}\n")
|
28 |
+
print(f"number of test problems: {len(test_qids)}\n")
|
29 |
+
|
30 |
+
qids = {'train': train_qids, 'val':val_qids,'test':test_qids}
|
31 |
+
return problems, qids,
|
32 |
+
|
33 |
+
def load_data_img(args):
|
34 |
+
problems = json.load(open(os.path.join(args.data_root, 'scienceqa/problems.json')))
|
35 |
+
pid_splits = json.load(open(os.path.join(args.data_root, 'scienceqa/pid_splits.json')))
|
36 |
+
captions = json.load(open(args.caption_file))["captions"]
|
37 |
+
name_maps = json.load(open('vision_features/name_map.json'))
|
38 |
+
|
39 |
+
# check
|
40 |
+
if args.img_type == "resnet":
|
41 |
+
image_features = np.load('vision_features/resnet.npy')
|
42 |
+
image_features = np.expand_dims(image_features, axis=1)
|
43 |
+
image_features = image_features.repeat(512, axis=1)
|
44 |
+
elif args.img_type == "clip":
|
45 |
+
image_features = np.load('vision_features/clip.npy')
|
46 |
+
elif args.img_type == "detr":
|
47 |
+
image_features = np.load('vision_features/detr.npy')
|
48 |
+
else:
|
49 |
+
image_features = np.load('vision_features/detr.npy')
|
50 |
+
print("img_features size: ", image_features.shape)
|
51 |
+
|
52 |
+
for qid in problems:
|
53 |
+
problems[qid]['caption'] = captions[qid] if qid in captions else ""
|
54 |
+
|
55 |
+
train_qids = pid_splits['%s' % (args.train_split)]
|
56 |
+
val_qids = pid_splits['%s' % (args.val_split)]
|
57 |
+
test_qids = pid_splits['%s' % (args.test_split)]
|
58 |
+
print(f"number of train problems: {len(train_qids)}\n")
|
59 |
+
print(f"number of val problems: {len(val_qids)}\n")
|
60 |
+
print(f"number of test problems: {len(test_qids)}\n")
|
61 |
+
|
62 |
+
qids = {'train': train_qids, 'val':val_qids,'test':test_qids}
|
63 |
+
return problems, qids, name_maps, image_features
|
64 |
+
|
65 |
+
class ScienceQADatasetStd(Dataset):
|
66 |
+
"""
|
67 |
+
Creating a custom dataset for reading the dataset and
|
68 |
+
loading it into the dataloader to pass it to the
|
69 |
+
neural network for finetuning the model
|
70 |
+
|
71 |
+
"""
|
72 |
+
|
73 |
+
def __init__(
|
74 |
+
self, problems, qids, tokenizer, source_len, target_len, args, test_le=None
|
75 |
+
):
|
76 |
+
self.tokenizer = tokenizer
|
77 |
+
self.data = {qid : problems[qid] for qid in qids}
|
78 |
+
self.source_len = source_len
|
79 |
+
self.summ_len = target_len
|
80 |
+
self.target_text = []
|
81 |
+
self.source_text = []
|
82 |
+
if test_le is not None:
|
83 |
+
test_le_data =json.load(open(test_le))["preds"]
|
84 |
+
else:
|
85 |
+
test_le_data = None
|
86 |
+
idx = 0
|
87 |
+
for qid in self.data:
|
88 |
+
if test_le_data is not None:
|
89 |
+
curr_le_data = test_le_data[idx]
|
90 |
+
idx += 1
|
91 |
+
else:
|
92 |
+
curr_le_data = None
|
93 |
+
prompt, target = build_train_pair(problems, qid, args, curr_le_data)
|
94 |
+
self.target_text.append(target)
|
95 |
+
self.source_text.append(prompt)
|
96 |
+
|
97 |
+
def __len__(self):
|
98 |
+
return len(self.target_text)
|
99 |
+
|
100 |
+
def __getitem__(self, index):
|
101 |
+
source_text = str(self.source_text[index])
|
102 |
+
target_text = str(self.target_text[index])
|
103 |
+
|
104 |
+
# cleaning data so as to ensure data is in string type
|
105 |
+
source_text = " ".join(source_text.split())
|
106 |
+
target_text = " ".join(target_text.split())
|
107 |
+
|
108 |
+
source = self.tokenizer.batch_encode_plus(
|
109 |
+
[source_text],
|
110 |
+
max_length=self.source_len,
|
111 |
+
pad_to_max_length=True,
|
112 |
+
truncation=True,
|
113 |
+
padding="max_length",
|
114 |
+
return_tensors="pt",
|
115 |
+
)
|
116 |
+
target = self.tokenizer.batch_encode_plus(
|
117 |
+
[target_text],
|
118 |
+
max_length=self.summ_len,
|
119 |
+
pad_to_max_length=True,
|
120 |
+
truncation=True,
|
121 |
+
padding="max_length",
|
122 |
+
return_tensors="pt",
|
123 |
+
)
|
124 |
+
source_ids = source["input_ids"].squeeze()
|
125 |
+
source_mask = source["attention_mask"].squeeze()
|
126 |
+
target_ids = target["input_ids"].squeeze().tolist()
|
127 |
+
|
128 |
+
return {
|
129 |
+
"input_ids": source_ids,
|
130 |
+
"attention_mask": source_mask,
|
131 |
+
"labels": target_ids,
|
132 |
+
}
|
133 |
+
|
134 |
+
|
135 |
+
class ScienceQADatasetImg(Dataset):
|
136 |
+
"""
|
137 |
+
Creating a custom dataset for reading the dataset and
|
138 |
+
loading it into the dataloader to pass it to the
|
139 |
+
neural network for finetuning the model
|
140 |
+
|
141 |
+
"""
|
142 |
+
|
143 |
+
def __init__(
|
144 |
+
self, problems, qids, name_maps, tokenizer, source_len, target_len, args, image_features, test_le=None
|
145 |
+
):
|
146 |
+
"""
|
147 |
+
Initializes a Dataset class
|
148 |
+
|
149 |
+
Args:
|
150 |
+
dataframe (pandas.DataFrame): Input dataframe
|
151 |
+
tokenizer (transformers.tokenizer): Transformers tokenizer
|
152 |
+
source_len (int): Max length of source text
|
153 |
+
target_len (int): Max length of target text
|
154 |
+
source_text (str): column name of source text
|
155 |
+
target_text (str): column name of target text
|
156 |
+
"""
|
157 |
+
self.tokenizer = tokenizer
|
158 |
+
self.data = {qid : problems[qid] for qid in qids}
|
159 |
+
self.source_len = source_len
|
160 |
+
self.summ_len = target_len
|
161 |
+
self.target_text = []
|
162 |
+
self.source_text = []
|
163 |
+
self.image_ids = []
|
164 |
+
if test_le is not None:
|
165 |
+
test_le_data =json.load(open(test_le))["preds"]
|
166 |
+
else:
|
167 |
+
test_le_data = None
|
168 |
+
idx = 0
|
169 |
+
for qid in self.data:
|
170 |
+
if test_le_data is not None:
|
171 |
+
curr_le_data = test_le_data[idx]
|
172 |
+
idx += 1
|
173 |
+
else:
|
174 |
+
curr_le_data = None
|
175 |
+
prompt, target = build_train_pair(problems, qid, args, curr_le_data)
|
176 |
+
self.target_text.append(target)
|
177 |
+
self.source_text.append(prompt)
|
178 |
+
if str(qid) in name_maps:
|
179 |
+
i_vectors = image_features[int(name_maps[str(qid)])]
|
180 |
+
self.image_ids.append(i_vectors)
|
181 |
+
else:
|
182 |
+
shape = img_shape[args.img_type]
|
183 |
+
self.image_ids.append(np.zeros(shape))
|
184 |
+
|
185 |
+
def __len__(self):
|
186 |
+
"""returns the length of dataframe"""
|
187 |
+
|
188 |
+
return len(self.target_text)
|
189 |
+
|
190 |
+
def __getitem__(self, index):
|
191 |
+
"""return the input ids, attention masks and target ids"""
|
192 |
+
|
193 |
+
source_text = str(self.source_text[index])
|
194 |
+
target_text = str(self.target_text[index])
|
195 |
+
image_ids = self.image_ids[index]
|
196 |
+
|
197 |
+
# cleaning data so as to ensure data is in string type
|
198 |
+
source_text = " ".join(source_text.split())
|
199 |
+
target_text = " ".join(target_text.split())
|
200 |
+
|
201 |
+
source = self.tokenizer.batch_encode_plus(
|
202 |
+
[source_text],
|
203 |
+
max_length=self.source_len,
|
204 |
+
pad_to_max_length=True,
|
205 |
+
truncation=True,
|
206 |
+
padding="max_length",
|
207 |
+
return_tensors="pt",
|
208 |
+
)
|
209 |
+
target = self.tokenizer.batch_encode_plus(
|
210 |
+
[target_text],
|
211 |
+
max_length=self.summ_len,
|
212 |
+
pad_to_max_length=True,
|
213 |
+
truncation=True,
|
214 |
+
padding="max_length",
|
215 |
+
return_tensors="pt",
|
216 |
+
)
|
217 |
+
source_ids = source["input_ids"].squeeze()
|
218 |
+
source_mask = source["attention_mask"].squeeze()
|
219 |
+
target_ids = target["input_ids"].squeeze().tolist()
|
220 |
+
|
221 |
+
image_ids = torch.tensor(image_ids).squeeze()
|
222 |
+
|
223 |
+
return {
|
224 |
+
"input_ids": source_ids,
|
225 |
+
"attention_mask": source_mask,
|
226 |
+
"image_ids": image_ids,
|
227 |
+
"labels": target_ids,
|
228 |
+
}
|
utils_evaluate.py
ADDED
@@ -0,0 +1,108 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
'''
|
2 |
+
Adapted from https://github.com/lupantech/ScienceQA
|
3 |
+
'''
|
4 |
+
|
5 |
+
import os
|
6 |
+
import json
|
7 |
+
import argparse
|
8 |
+
import warnings
|
9 |
+
import pandas as pd
|
10 |
+
from sentence_transformers import SentenceTransformer
|
11 |
+
from evaluations import caculate_bleu, caculate_rouge, caculate_similariry
|
12 |
+
|
13 |
+
warnings.filterwarnings('ignore')
|
14 |
+
|
15 |
+
def get_acc_with_contion(res_pd, key, values):
|
16 |
+
if isinstance(values, list):
|
17 |
+
total_pd = res_pd[res_pd[key].isin(values)]
|
18 |
+
else:
|
19 |
+
total_pd = res_pd[res_pd[key] == values]
|
20 |
+
correct_pd = total_pd[total_pd['true_false'] == True]
|
21 |
+
acc = "{:.2f}".format(len(correct_pd) / len(total_pd) * 100)
|
22 |
+
return acc
|
23 |
+
|
24 |
+
|
25 |
+
def get_scores(result_data, rationale_data, results_reference, data_file):
|
26 |
+
# read result file
|
27 |
+
results = result_data
|
28 |
+
num = len(results)
|
29 |
+
assert num == 4241
|
30 |
+
#print("number of questions:", num)
|
31 |
+
|
32 |
+
# read data file
|
33 |
+
sqa_data = json.load(open(data_file))
|
34 |
+
|
35 |
+
# construct pandas data
|
36 |
+
sqa_pd = pd.DataFrame(sqa_data).T
|
37 |
+
res_pd = sqa_pd[sqa_pd['split'] == 'test'] # test set
|
38 |
+
|
39 |
+
# update data
|
40 |
+
for index, row in res_pd.iterrows():
|
41 |
+
|
42 |
+
res_pd.loc[index, 'no_context'] = True if (not row['hint'] and not row['image']) else False
|
43 |
+
res_pd.loc[index, 'has_text'] = True if row['hint'] else False
|
44 |
+
res_pd.loc[index, 'has_image'] = True if row['image'] else False
|
45 |
+
res_pd.loc[index, 'has_text_image'] = True if (row['hint'] and row['image']) else False
|
46 |
+
|
47 |
+
label = row['answer']
|
48 |
+
pred = int(results[index])
|
49 |
+
res_pd.loc[index, 'pred'] = pred
|
50 |
+
res_pd.loc[index, 'true_false'] = (label == pred)
|
51 |
+
|
52 |
+
# accuracy scores
|
53 |
+
acc_average = len(res_pd[res_pd['true_false'] == True]) / num * 100
|
54 |
+
#assert result_file.split('_')[-1] == "{:.3f}.json".format(acc_average)
|
55 |
+
|
56 |
+
|
57 |
+
# rationale quality
|
58 |
+
|
59 |
+
## BLEU
|
60 |
+
bleu1 = caculate_bleu(rationale_data, results_reference, gram=1)
|
61 |
+
bleu4 = caculate_bleu(rationale_data, results_reference, gram=4)
|
62 |
+
|
63 |
+
## Rouge-L
|
64 |
+
rouge = caculate_rouge(rationale_data, results_reference)
|
65 |
+
|
66 |
+
## Similarity
|
67 |
+
model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2').cuda()
|
68 |
+
similariry = caculate_similariry(rationale_data, results_reference, model)
|
69 |
+
|
70 |
+
scores = {
|
71 |
+
"answer":{
|
72 |
+
'acc_natural':
|
73 |
+
get_acc_with_contion(res_pd, 'subject', 'natural science'),
|
74 |
+
'acc_social':
|
75 |
+
get_acc_with_contion(res_pd, 'subject', 'social science'),
|
76 |
+
'acc_language':
|
77 |
+
get_acc_with_contion(res_pd, 'subject', 'language science'),
|
78 |
+
'acc_has_text':
|
79 |
+
get_acc_with_contion(res_pd, 'has_text', True),
|
80 |
+
'acc_has_image':
|
81 |
+
get_acc_with_contion(res_pd, 'has_image', True),
|
82 |
+
'acc_no_context':
|
83 |
+
get_acc_with_contion(res_pd, 'no_context', True),
|
84 |
+
'acc_grade_1_6':
|
85 |
+
get_acc_with_contion(res_pd, 'grade', ['grade1', 'grade2', 'grade3', 'grade4', 'grade5', 'grade6']),
|
86 |
+
'acc_grade_7_12':
|
87 |
+
get_acc_with_contion(res_pd, 'grade', ['grade7', 'grade8', 'grade9', 'grade10', 'grade11', 'grade12']),
|
88 |
+
'acc_average':
|
89 |
+
"{:.2f}".format(acc_average),
|
90 |
+
},
|
91 |
+
"rationale":{
|
92 |
+
'bleu1': bleu1 * 100,
|
93 |
+
'bleu4': bleu4 * 100,
|
94 |
+
'rouge': rouge * 100,
|
95 |
+
'similariry': similariry * 100,
|
96 |
+
}
|
97 |
+
}
|
98 |
+
|
99 |
+
return scores
|
100 |
+
|
101 |
+
|
102 |
+
def print_scores(scores):
|
103 |
+
latex_output = ""
|
104 |
+
for key, score in scores.items():
|
105 |
+
print(f"{key[4:]}: \t{score}")
|
106 |
+
latex_output += f"& {score} "
|
107 |
+
latex_output += "\\\\"
|
108 |
+
print(latex_output)
|
utils_prompt.py
ADDED
@@ -0,0 +1,240 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
'''
|
2 |
+
Adapted from https://github.com/lupantech/ScienceQA
|
3 |
+
'''
|
4 |
+
|
5 |
+
from dataclasses import dataclass
|
6 |
+
from typing import List, Optional
|
7 |
+
|
8 |
+
def get_question_text(problem):
|
9 |
+
question = problem['question']
|
10 |
+
return question
|
11 |
+
|
12 |
+
|
13 |
+
def get_context_text(problem, use_caption):
|
14 |
+
txt_context = problem['hint']
|
15 |
+
img_context = problem['caption'] if use_caption else ""
|
16 |
+
context = " ".join([txt_context, img_context]).strip()
|
17 |
+
if context == "":
|
18 |
+
context = "N/A"
|
19 |
+
return context
|
20 |
+
|
21 |
+
|
22 |
+
def get_choice_text(probelm, options):
|
23 |
+
choices = probelm['choices']
|
24 |
+
choice_list = []
|
25 |
+
for i, c in enumerate(choices):
|
26 |
+
choice_list.append("({}) {}".format(options[i], c))
|
27 |
+
choice_txt = " ".join(choice_list)
|
28 |
+
#print(choice_txt)
|
29 |
+
return choice_txt
|
30 |
+
|
31 |
+
def get_origin_answer(problem, options):
|
32 |
+
return problem['choices'][problem['answer']]
|
33 |
+
|
34 |
+
def get_answer(problem, options):
|
35 |
+
return options[problem['answer']]
|
36 |
+
|
37 |
+
|
38 |
+
def get_lecture_text(problem):
|
39 |
+
# \\n: GPT-3 can generate the lecture with more tokens.
|
40 |
+
lecture = problem['lecture'].replace("\n", "\\n")
|
41 |
+
return lecture
|
42 |
+
|
43 |
+
|
44 |
+
def get_solution_text(problem):
|
45 |
+
# \\n: GPT-3 can generate the solution with more tokens
|
46 |
+
solution = problem['solution'].replace("\n", "\\n")
|
47 |
+
return solution
|
48 |
+
|
49 |
+
|
50 |
+
def create_one_example(format, question, context, choice, answer, lecture, solution, test_example=True, WithOutput = False, curr_le_data=None):
|
51 |
+
|
52 |
+
input_format, output_format = format.split("-")
|
53 |
+
|
54 |
+
## Inputs
|
55 |
+
if input_format == "CQM":
|
56 |
+
input = f"Context: {context}\nQuestion: {question}\nOptions: {choice}\n"
|
57 |
+
elif input_format == "QCM":
|
58 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\n"
|
59 |
+
elif input_format == "QM":
|
60 |
+
input = f"Question: {question}\nOptions: {choice}\n"
|
61 |
+
elif input_format == "QC":
|
62 |
+
input = f"Question: {question}\nContext: {context}\n"
|
63 |
+
elif input_format == "QCMG":
|
64 |
+
if curr_le_data is not None:
|
65 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\n{curr_le_data}\n"
|
66 |
+
else:
|
67 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nSolution: {lecture} {solution}\n"
|
68 |
+
elif input_format == "CQMG":
|
69 |
+
if curr_le_data is not None:
|
70 |
+
input = f"Context: {context}\nQuestion: {question}\nOptions: {choice}\n{curr_le_data}\n"
|
71 |
+
else:
|
72 |
+
input = f"Context: {context}\nQuestion: {question}\nOptions: {choice}\nSolution: {lecture} {solution}\n"
|
73 |
+
# upper bound experiment
|
74 |
+
elif input_format == "QCML":
|
75 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nBECAUSE: {lecture}\n"
|
76 |
+
elif input_format == "QCME":
|
77 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nBECAUSE: {solution}\n"
|
78 |
+
elif input_format == "QCMLE":
|
79 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nBECAUSE: {lecture} {solution}\n"
|
80 |
+
|
81 |
+
elif input_format == "QCLM":
|
82 |
+
input = f"Question: {question}\nContext: {context}\nBECAUSE: {lecture}\nOptions: {choice}\n"
|
83 |
+
elif input_format == "QCEM":
|
84 |
+
input = f"Question: {question}\nContext: {context}\nBECAUSE: {solution}\nOptions: {choice}\n"
|
85 |
+
elif input_format == "QCLEM":
|
86 |
+
input = f"Question: {question}\nContext: {context}\nBECAUSE: {lecture} {solution}\nOptions: {choice}\n"
|
87 |
+
elif input_format == "QCMA":
|
88 |
+
input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nAnswer: The answer is {answer}.\n"
|
89 |
+
elif input_format == "QCA":
|
90 |
+
input = f"Question: {question}\nContext: {context}\nAnswer: The answer is {answer}. \nBECAUSE:"
|
91 |
+
|
92 |
+
# Outputs
|
93 |
+
if test_example:
|
94 |
+
if output_format == 'A':
|
95 |
+
output = "Answer:"
|
96 |
+
elif output_format == 'E':
|
97 |
+
output = "Solution:"
|
98 |
+
else:
|
99 |
+
output = "Solution:"
|
100 |
+
elif output_format == 'A':
|
101 |
+
output = f"Answer: The answer is {answer}."
|
102 |
+
|
103 |
+
elif output_format == 'AL':
|
104 |
+
output = f"Answer: The answer is {answer}. BECAUSE: {solution}"
|
105 |
+
elif output_format == 'AE':
|
106 |
+
output = f"Answer: The answer is {answer}. BECAUSE: {lecture}"
|
107 |
+
elif output_format == 'ALE':
|
108 |
+
output = f"Answer: The answer is {answer}. BECAUSE: {lecture} {solution}"
|
109 |
+
elif output_format == 'AEL':
|
110 |
+
output = f"Answer: The answer is {answer}. BECAUSE: {solution} {lecture}"
|
111 |
+
|
112 |
+
elif output_format == 'LA':
|
113 |
+
output = f"Answer: {lecture} The answer is {answer}."
|
114 |
+
elif output_format == 'EA':
|
115 |
+
output = f"Answer: {solution} The answer is {answer}."
|
116 |
+
elif output_format == 'LEA':
|
117 |
+
output = f"Answer: {lecture} {solution} The answer is {answer}."
|
118 |
+
elif output_format == 'ELA':
|
119 |
+
output = f"Answer: {solution} {lecture} The answer is {answer}."
|
120 |
+
|
121 |
+
elif output_format == 'LE':
|
122 |
+
output = f"Solution: {lecture} {solution}."
|
123 |
+
|
124 |
+
elif output_format == 'E':
|
125 |
+
output = f"Solution: {solution}"
|
126 |
+
|
127 |
+
|
128 |
+
if WithOutput:
|
129 |
+
if output.endswith("BECAUSE:"):
|
130 |
+
output = output.replace("BECAUSE:", "").strip()
|
131 |
+
if output_format == 'E':
|
132 |
+
text = input + f'Solution:'
|
133 |
+
elif output_format == 'A':
|
134 |
+
text = input + f'Answer:'
|
135 |
+
else:
|
136 |
+
text = input + f'Solution:'
|
137 |
+
text = text.replace(" ", " ").strip()
|
138 |
+
output = output.replace(" ", " ").strip()
|
139 |
+
return text, output
|
140 |
+
|
141 |
+
|
142 |
+
text = input + output
|
143 |
+
text = text.replace(" ", " ").strip()
|
144 |
+
if text.endswith("BECAUSE:"):
|
145 |
+
text = text.replace("BECAUSE:", "").strip()
|
146 |
+
return text
|
147 |
+
|
148 |
+
|
149 |
+
def build_prompt(problems, shot_qids, test_qid, args):
|
150 |
+
|
151 |
+
examples = []
|
152 |
+
|
153 |
+
# n-shot training examples
|
154 |
+
for qid in shot_qids:
|
155 |
+
question = get_question_text(problems[qid])
|
156 |
+
context = get_context_text(problems[qid], args.use_caption)
|
157 |
+
choice = get_choice_text(problems[qid], args.options)
|
158 |
+
answer = get_answer(problems[qid], args.options)
|
159 |
+
lecture = get_lecture_text(problems[qid])
|
160 |
+
solution = get_solution_text(problems[qid])
|
161 |
+
|
162 |
+
train_example = create_one_example(args.prompt_format,
|
163 |
+
question,
|
164 |
+
context,
|
165 |
+
choice,
|
166 |
+
answer,
|
167 |
+
lecture,
|
168 |
+
solution,
|
169 |
+
test_example=False)
|
170 |
+
examples.append(train_example)
|
171 |
+
|
172 |
+
# test example
|
173 |
+
question = get_question_text(problems[test_qid])
|
174 |
+
context = get_context_text(problems[test_qid], args.use_caption)
|
175 |
+
choice = get_choice_text(problems[test_qid], args.options)
|
176 |
+
answer = get_answer(problems[test_qid], args.options)
|
177 |
+
lecture = get_lecture_text(problems[test_qid])
|
178 |
+
solution = get_solution_text(problems[test_qid])
|
179 |
+
|
180 |
+
test_example = create_one_example(args.prompt_format,
|
181 |
+
question,
|
182 |
+
context,
|
183 |
+
choice,
|
184 |
+
answer,
|
185 |
+
lecture,
|
186 |
+
solution,
|
187 |
+
test_example=True)
|
188 |
+
examples.append(test_example)
|
189 |
+
|
190 |
+
# create the prompt input
|
191 |
+
prompt_input = '\n\n'.join(examples)
|
192 |
+
|
193 |
+
return prompt_input
|
194 |
+
|
195 |
+
def build_train_pair(problems, test_qid, args, curr_le_data=None):
|
196 |
+
|
197 |
+
examples = []
|
198 |
+
|
199 |
+
# test example
|
200 |
+
question = get_question_text(problems[test_qid])
|
201 |
+
context = get_context_text(problems[test_qid], args.use_caption)
|
202 |
+
choice = get_choice_text(problems[test_qid], args.options)
|
203 |
+
|
204 |
+
lecture = get_lecture_text(problems[test_qid])
|
205 |
+
solution = get_solution_text(problems[test_qid])
|
206 |
+
|
207 |
+
# answer_text = get_origin_answer(problems[test_qid], args.options)
|
208 |
+
answer_option = get_answer(problems[test_qid], args.options)
|
209 |
+
answer = "(" + answer_option + ")"
|
210 |
+
|
211 |
+
test_example, target = create_one_example(args.prompt_format,
|
212 |
+
question,
|
213 |
+
context,
|
214 |
+
choice,
|
215 |
+
answer,
|
216 |
+
lecture,
|
217 |
+
solution,
|
218 |
+
test_example=False,WithOutput = True, curr_le_data=curr_le_data)
|
219 |
+
examples.append(test_example)
|
220 |
+
|
221 |
+
target = target.replace("Answer:", "").strip()
|
222 |
+
# create the prompt input
|
223 |
+
prompt_input = '\n\n'.join(examples)
|
224 |
+
|
225 |
+
return prompt_input, target
|
226 |
+
|
227 |
+
@dataclass(frozen=True)
|
228 |
+
class InputFeatures:
|
229 |
+
"""
|
230 |
+
A single set of features of data.
|
231 |
+
Property names are the same names as the corresponding inputs to a model.
|
232 |
+
"""
|
233 |
+
|
234 |
+
input_ids: List[List[int]]
|
235 |
+
attention_mask: Optional[List[List[int]]]
|
236 |
+
token_type_ids: Optional[List[List[int]]]
|
237 |
+
le_input_ids: List[List[int]]
|
238 |
+
le_attention_mask: Optional[List[List[int]]]
|
239 |
+
le_token_type_ids: Optional[List[List[int]]]
|
240 |
+
label: Optional[int]
|
vision_features/mm-cot.png
ADDED