structure

CODE_OF_CONDUCT.md

@@ -0,0 +1,4 @@
+## Code of Conduct
+This project has adopted the [Amazon Open Source Code of Conduct](https://aws.github.io/code-of-conduct).
+For more information see the [Code of Conduct FAQ](https://aws.github.io/code-of-conduct-faq) or contact
+opensource-codeofconduct@amazon.com with any additional questions or comments.

CONTRIBUTING.md

@@ -0,0 +1,59 @@
+# Contributing Guidelines
+
+Thank you for your interest in contributing to our project. Whether it's a bug report, new feature, correction, or additional
+documentation, we greatly value feedback and contributions from our community.
+
+Please read through this document before submitting any issues or pull requests to ensure we have all the necessary
+information to effectively respond to your bug report or contribution.
+
+
+## Reporting Bugs/Feature Requests
+
+We welcome you to use the GitHub issue tracker to report bugs or suggest features.
+
+When filing an issue, please check existing open, or recently closed, issues to make sure somebody else hasn't already
+reported the issue. Please try to include as much information as you can. Details like these are incredibly useful:
+
+* A reproducible test case or series of steps
+* The version of our code being used
+* Any modifications you've made relevant to the bug
+* Anything unusual about your environment or deployment
+
+
+## Contributing via Pull Requests
+Contributions via pull requests are much appreciated. Before sending us a pull request, please ensure that:
+
+1. You are working against the latest source on the *main* branch.
+2. You check existing open, and recently merged, pull requests to make sure someone else hasn't addressed the problem already.
+3. You open an issue to discuss any significant work - we would hate for your time to be wasted.
+
+To send us a pull request, please:
+
+1. Fork the repository.
+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.
+3. Ensure local tests pass.
+4. Commit to your fork using clear commit messages.
+5. Send us a pull request, answering any default questions in the pull request interface.
+6. Pay attention to any automated CI failures reported in the pull request, and stay involved in the conversation.
+
+GitHub provides additional document on [forking a repository](https://help.github.com/articles/fork-a-repo/) and
+[creating a pull request](https://help.github.com/articles/creating-a-pull-request/).
+
+
+## Finding contributions to work on
+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.
+
+
+## Code of Conduct
+This project has adopted the [Amazon Open Source Code of Conduct](https://aws.github.io/code-of-conduct).
+For more information see the [Code of Conduct FAQ](https://aws.github.io/code-of-conduct-faq) or contact
+opensource-codeofconduct@amazon.com with any additional questions or comments.
+
+
+## Security issue notifications
+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.
+
+
+## Licensing
+
+See the [LICENSE](LICENSE) file for our project's licensing. We will ask you to confirm the licensing of your contribution.

LICENSE

@@ -0,0 +1,175 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.

NOTICE

@@ -0,0 +1 @@
+Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.

evaluations.py

@@ -0,0 +1,100 @@
+'''
+Adapted from https://github.com/lupantech/ScienceQA
+'''
+
+import re
+from rouge import Rouge
+from nltk.translate.bleu_score import sentence_bleu
+from sentence_transformers import util
+
+########################
+## BLEU
+########################
+def tokenize(text):
+    tokens = re.split(r'\s|\.', text)
+    tokens = [t for t in tokens if len(t) > 0]
+    return tokens
+
+
+def bleu_score(reference, hypothesis, gram):
+    reference_tokens = tokenize(reference)
+    hypothesis_tokens = tokenize(hypothesis)
+
+    if gram == 1:
+        bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1., ))  # BELU-1
+    elif gram == 2:
+        bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1. / 2., 1. / 2.))  # BELU-2
+    elif gram == 3:
+        bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1. / 3., 1. / 3., 1. / 3.))  # BELU-3
+    elif gram == 4:
+        bleu = sentence_bleu([reference_tokens], hypothesis_tokens, (1. / 4., 1. / 4., 1. / 4., 1. / 4.))  # BELU-4
+
+    return bleu
+
+
+def caculate_bleu(results, data, gram):
+    bleus = []
+    for qid, output in results.items():
+        prediction = output
+        target = data[qid]
+        target = target.strip()
+        if target == "":
+            continue
+        bleu = bleu_score(target, prediction, gram)
+        bleus.append(bleu)
+
+    avg_bleu = sum(bleus) / len(bleus)
+
+    return avg_bleu
+
+
+########################
+## Rouge-L
+########################
+def score_rouge(str1, str2):
+    rouge = Rouge(metrics=["rouge-l"])
+    scores = rouge.get_scores(str1, str2, avg=True)
+    rouge_l = scores['rouge-l']['f']
+    return rouge_l
+
+
+def caculate_rouge(results, data):
+    rouges = []
+    for qid, output in results.items():
+        prediction = output
+        target = data[qid]
+        target = target.strip()
+        if prediction == "":
+            continue
+        if target == "":
+            continue
+        rouge = score_rouge(target, prediction)
+        rouges.append(rouge)
+
+    avg_rouge = sum(rouges) / len(rouges)
+    return avg_rouge
+
+
+########################
+## Sentence Similarity
+########################
+def similariry_score(str1, str2, model):
+    # compute embedding for both lists
+    embedding_1 = model.encode(str1, convert_to_tensor=True)
+    embedding_2 = model.encode(str2, convert_to_tensor=True)
+    score = util.pytorch_cos_sim(embedding_1, embedding_2).item()
+    return score
+
+
+def caculate_similariry(results, data, model):
+    scores = []
+    for qid, output in results.items():
+        prediction = output
+        target = data[qid]
+        target = target.strip()
+
+        score = similariry_score(target, prediction, model)
+        scores.append(score)
+
+    avg_score = sum(scores) / len(scores)
+    return avg_score

main.py

@@ -0,0 +1,383 @@
+import os
+import numpy as np
+import torch
+import os
+import re
+import json
+import argparse
+import random
+from transformers import T5Tokenizer, DataCollatorForSeq2Seq, Seq2SeqTrainingArguments, Seq2SeqTrainer, T5ForConditionalGeneration
+from model import T5ForConditionalGeneration, T5ForMultimodalGeneration
+from utils_data import img_shape, load_data_std, load_data_img, ScienceQADatasetStd, ScienceQADatasetImg
+from utils_prompt import *
+from utils_evaluate import get_scores
+from rich.table import Column, Table
+from rich import box
+from rich.console import Console
+console = Console(record=True)
+from torch import cuda
+import nltk
+import evaluate
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--data_root', type=str, default='data')
+    parser.add_argument('--output_dir', type=str, default='experiments')
+    parser.add_argument('--model', type=str, default='allenai/unifiedqa-t5-base')
+    parser.add_argument('--options', type=list, default=["A", "B", "C", "D", "E"])
+    parser.add_argument('--epoch', type=int, default=20)
+    parser.add_argument('--lr', type=float, default=5e-5)
+    parser.add_argument('--bs', type=int, default=16)
+    parser.add_argument('--input_len', type=int, default=512)
+    parser.add_argument('--output_len', type=int, default=64)
+    parser.add_argument('--eval_bs', type=int, default=16)
+    parser.add_argument('--eval_acc', type=int, default=None, help='evaluate accumulation step')
+    parser.add_argument('--train_split', type=str, default='train', choices=['train', 'trainval', 'minitrain'])
+    parser.add_argument('--val_split', type=str, default='val', choices=['test', 'val', 'minival'])
+    parser.add_argument('--test_split', type=str, default='test', choices=['test', 'minitest'])
+    
+    parser.add_argument('--use_generate', action='store_true', help='only for baseline to improve inference speed')
+    parser.add_argument('--final_eval', action='store_true', help='only evaluate the model at the final epoch')
+    parser.add_argument('--user_msg', type=str, default="baseline", help='experiment type in the save_dir')
+    parser.add_argument('--img_type', type=str, default=None, choices=['detr', 'clip', 'resnet'], help='type of image features')
+    parser.add_argument('--eval_le', type=str, default=None, help='generated rationale for the dev set')
+    parser.add_argument('--test_le', type=str, default=None, help='generated rationale for the test set')
+    parser.add_argument('--evaluate_dir', type=str, default=None, help='the directory of model for evaluation')
+    parser.add_argument('--caption_file', type=str, default='data/captions.json')
+    parser.add_argument('--use_caption', action='store_true', help='use image captions or not')
+    parser.add_argument('--prompt_format', type=str, default='QCM-A', help='prompt format template',
+                        choices=['QCM-A', 'QCM-LE', 'QCMG-A', 'QCM-LEA', 'QCM-ALE'])
+    parser.add_argument('--seed', type=int, default=42, help='random seed')
+
+    args = parser.parse_args()
+    return args
+        
+def T5Trainer(
+    dataframe, args,
+):
+    torch.manual_seed(args.seed)  # pytorch random seed
+    np.random.seed(args.seed)  # numpy random seed
+    torch.backends.cudnn.deterministic = True
+    
+    if args.evaluate_dir is not None:
+        args.model = args.evaluate_dir
+
+    tokenizer = T5Tokenizer.from_pretrained(args.model)
+
+    console.log(f"""[Model]: Loading {args.model}...\n""")
+    console.log(f"[Data]: Reading data...\n")
+    problems = dataframe['problems']
+    qids = dataframe['qids']
+    train_qids = qids['train']
+    test_qids = qids['test']
+    val_qids = qids['val']
+    
+    if args.evaluate_dir is not None:
+        save_dir = args.evaluate_dir
+    else:
+        model_name = args.model.replace("/","-")
+        gpu_count = torch.cuda.device_count()
+        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}"
+        if not os.path.exists(save_dir):
+            os.mkdir(save_dir)
+
+    padding_idx = tokenizer._convert_token_to_id(tokenizer.pad_token)
+    if args.img_type is not None:
+        patch_size = img_shape[args.img_type]
+        model = T5ForMultimodalGeneration.from_pretrained(args.model, patch_size=patch_size, padding_idx=padding_idx, save_dir=save_dir) 
+        name_maps = dataframe['name_maps'] 
+        image_features = dataframe['image_features']
+        train_set = ScienceQADatasetImg(
+            problems,
+            train_qids,
+            name_maps,
+            tokenizer,
+            args.input_len,
+            args.output_len,
+            args,
+            image_features,
+        )
+        eval_set = ScienceQADatasetImg(
+            problems,
+            val_qids,
+            name_maps,
+            tokenizer,
+            args.input_len,
+            args.output_len,
+            args,
+            image_features,
+            args.eval_le,
+        )
+        test_set = ScienceQADatasetImg(
+            problems,
+            test_qids,
+            name_maps,
+            tokenizer,
+            args.input_len,
+            args.output_len,
+            args,
+            image_features,
+            args.test_le,
+        )
+    else:
+        model = T5ForConditionalGeneration.from_pretrained(args.model) 
+        train_set = ScienceQADatasetStd(
+            problems,
+            train_qids,
+            tokenizer,
+            args.input_len,
+            args.output_len,
+            args,
+        )
+        eval_set = ScienceQADatasetStd(
+            problems,
+            val_qids,
+            tokenizer,
+            args.input_len,
+            args.output_len,
+            args,
+            args.eval_le,
+        )
+        
+        test_set = ScienceQADatasetStd(
+            problems,
+            test_qids,
+            tokenizer,
+            args.input_len,
+            args.output_len,
+            args,
+            args.test_le,
+        )
+
+    datacollator = DataCollatorForSeq2Seq(tokenizer)
+    print("model parameters: ", model.num_parameters())
+    def extract_ans(ans):
+        pattern = re.compile(r'The answer is \(([A-Z])\)')
+        res = pattern.findall(ans)
+        
+        if len(res) == 1:
+            answer = res[0]  # 'A', 'B', ...
+        else:
+            answer = "FAILED" 
+        return answer  
+
+    # accuracy for answer inference
+    def compute_metrics_acc(eval_preds):
+        if args.use_generate:
+            preds, targets = eval_preds
+            if isinstance(preds, tuple):
+                preds = preds[0]
+        else:
+            preds = eval_preds.predictions[0]
+            targets = eval_preds.label_ids
+            preds = preds.argmax(axis=2)
+        preds = tokenizer.batch_decode(preds, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+        targets = tokenizer.batch_decode(targets, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+        correct = 0
+        assert len(preds) == len(targets)
+        for idx, pred in enumerate(preds):
+            reference = targets[idx]
+            reference = extract_ans(reference)
+            extract_pred = extract_ans(pred)
+            best_option = extract_pred
+            if reference == best_option:
+                correct +=1 
+        return {'accuracy': 1.0*correct/len(targets)}
+    
+    # rougel for rationale generation
+    metric = evaluate.load("rouge")
+    def postprocess_text(preds, labels):
+        preds = [pred.strip() for pred in preds]
+        labels = [label.strip() for label in labels]
+        preds = ["\n".join(nltk.sent_tokenize(pred)) for pred in preds]
+        labels = ["\n".join(nltk.sent_tokenize(label)) for label in labels]
+        return preds, labels
+
+    def compute_metrics_rougel(eval_preds):
+        if args.use_generate:
+            preds, targets = eval_preds
+            if isinstance(preds, tuple):
+                preds = preds[0]
+        else:
+            preds = eval_preds.predictions[0]
+            targets = eval_preds.label_ids
+            preds = preds.argmax(axis=2)
+        preds = tokenizer.batch_decode(preds, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+        targets = tokenizer.batch_decode(targets, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+
+        decoded_preds, decoded_labels = postprocess_text(preds, targets)
+
+        result = metric.compute(predictions=decoded_preds, references=decoded_labels, use_stemmer=True)
+        result = {k: round(v * 100, 4) for k, v in result.items()}
+        prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
+        result["gen_len"] = np.mean(prediction_lens)
+        return result
+
+    # only use the last model for evaluation to save time
+    if args.final_eval:
+        training_args = Seq2SeqTrainingArguments(
+            save_dir,
+            do_train=True if args.evaluate_dir is None else False,
+            do_eval=False,
+            evaluation_strategy="no",
+            logging_strategy="steps",
+            save_strategy="epoch",
+            save_total_limit = 2,
+            learning_rate= args.lr,
+            eval_accumulation_steps=args.eval_acc,
+            per_device_train_batch_size=args.bs,
+            per_device_eval_batch_size=args.eval_bs,
+            weight_decay=0.01,
+            num_train_epochs=args.epoch,
+            predict_with_generate=args.use_generate,
+            report_to="none",
+        )
+    # evaluate at each epoch
+    else:
+        training_args = Seq2SeqTrainingArguments(
+            save_dir,
+            do_train=True if args.evaluate_dir is None else False,
+            do_eval=True,
+            evaluation_strategy="epoch",
+            logging_strategy="steps",
+            save_strategy="epoch",
+            save_total_limit = 2,
+            learning_rate= args.lr,
+            eval_accumulation_steps=args.eval_acc,
+            per_device_train_batch_size=args.bs,
+            per_device_eval_batch_size=args.eval_bs,
+            weight_decay=0.01,
+            num_train_epochs=args.epoch,
+            metric_for_best_model="accuracy" if args.prompt_format != "QCM-LE" else "rougeL",
+            predict_with_generate=args.use_generate,
+            load_best_model_at_end=True,
+            report_to="none",
+        )
+
+    trainer = Seq2SeqTrainer(
+        model=model,
+        args=training_args,
+        train_dataset=train_set,
+        eval_dataset=eval_set,
+        data_collator=datacollator,
+        tokenizer=tokenizer,
+        compute_metrics = compute_metrics_acc if args.prompt_format != "QCM-LE" else compute_metrics_rougel
+    )
+
+    if args.evaluate_dir is None:
+        trainer.train()
+        trainer.save_model(save_dir)
+        
+    metrics = trainer.evaluate(eval_dataset = test_set)
+    trainer.log_metrics("test", metrics)
+    trainer.save_metrics("test", metrics)
+
+    predict_results = trainer.predict(test_dataset=test_set, max_length=args.output_len) 
+    if trainer.is_world_process_zero():
+        if args.use_generate:
+            preds, targets = predict_results.predictions, predict_results.label_ids
+        else:
+            preds = predict_results.predictions[0]
+            targets = predict_results.label_ids
+            preds = preds.argmax(axis=2)
+
+        preds = tokenizer.batch_decode(
+            preds, skip_special_tokens=True, clean_up_tokenization_spaces=True
+        )
+        targets = tokenizer.batch_decode(
+            targets, skip_special_tokens=True, clean_up_tokenization_spaces=True
+        )
+
+        results_ans = {}
+        results_rationale = {}
+        results_reference = {}
+        
+        num_fail = 0
+        for idx, qid in enumerate(test_qids):
+            pred = preds[int(idx)]
+            ref = targets[int(idx)]
+            extract_pred = extract_ans(pred)
+            if extract_pred != "FAILED":
+                if extract_pred in args.options:
+                    extract_pred = args.options.index(extract_pred)
+                else:
+                    extract_pred = random.choice(range(0,len(args.options)))
+            else:
+                num_fail += 1
+                extract_pred = random.choice(range(len(args.options))) # random choose one option
+            results_ans[str(qid)] = extract_pred
+            results_rationale[str(qid)] = pred
+            results_reference[str(qid)] = ref
+
+        scores = get_scores(results_ans, results_rationale, results_reference, os.path.join(args.data_root, "scienceqa/problems.json"))
+        preds = [pred.strip() for pred in preds]
+        output_data = {
+                "num_fail": num_fail,
+                "scores": scores,
+                "preds": preds,
+                 "labels": targets}
+        output_prediction_file = os.path.join(save_dir,"predictions_ans_test.json")
+        with open(output_prediction_file, "w") as writer:
+            writer.write(json.dumps(output_data, indent=4))
+    
+    # generate the rationale for the eval set
+    if args.prompt_format == "QCM-LE":
+        torch.cuda.empty_cache()
+        del predict_results, preds, targets
+        predict_results = trainer.predict(test_dataset=eval_set, max_length=args.output_len) 
+        if trainer.is_world_process_zero():
+            if args.use_generate:
+                preds, targets = predict_results.predictions, predict_results.label_ids
+            else:
+                preds = predict_results.predictions[0]
+                targets = predict_results.label_ids
+                preds = preds.argmax(axis=2)
+
+            preds = tokenizer.batch_decode(
+                preds, skip_special_tokens=True, clean_up_tokenization_spaces=True
+            )
+            targets = tokenizer.batch_decode(
+                targets, skip_special_tokens=True, clean_up_tokenization_spaces=True
+            )
+            preds = [pred.strip() for pred in preds]
+            output_data = {"preds": preds,
+                 "labels": targets}
+            output_prediction_file = os.path.join(save_dir,"predictions_ans_eval.json")
+            with open(output_prediction_file, "w") as writer:
+                writer.write(json.dumps(output_data, indent=4))
+    
+
+if __name__ == '__main__':
+
+    # training logger to log training progress
+    training_logger = Table(
+        Column("Epoch", justify="center"),
+        Column("Steps", justify="center"),
+        Column("Loss", justify="center"),
+        title="Training Status",
+        pad_edge=False,
+        box=box.ASCII,
+    )
+    
+    args = parse_args()
+    print("args",args)
+    print('====Input Arguments====')
+    print(json.dumps(vars(args), indent=2, sort_keys=False))
+
+    random.seed(args.seed)
+    
+    if not os.path.exists(args.output_dir):
+            os.mkdir(args.output_dir)
+
+    if args.img_type is not None:
+        problems, qids, name_maps, image_features = load_data_img(args)  # probelms, test question ids, shot example ids
+        dataframe = {'problems':problems, 'qids':qids, 'name_maps': name_maps, 'image_features': image_features}
+    else:
+        problems, qids = load_data_std(args)  # probelms, test question ids, shot example ids
+        dataframe = {'problems':problems, 'qids':qids}
+
+    T5Trainer(
+        dataframe=dataframe,
+        args = args
+    )

model.py

@@ -0,0 +1,194 @@
+'''
+Adapted from https://github.com/huggingface/transformers
+'''
+
+from transformers import T5Config, T5ForConditionalGeneration
+from transformers.models.t5.modeling_t5 import T5Stack, __HEAD_MASK_WARNING_MSG, T5EncoderModel
+import copy
+import math
+import os
+import warnings
+from typing import Optional, Tuple, Union
+import torch
+from torch import nn
+from torch.nn import CrossEntropyLoss
+from transformers.modeling_outputs import (
+    BaseModelOutput,
+    Seq2SeqLMOutput,
+)
+
+class T5ForMultimodalGeneration(T5ForConditionalGeneration):
+    _keys_to_ignore_on_load_missing = [
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
+        r"lm_head.weight",
+    ]
+    _keys_to_ignore_on_load_unexpected = [
+        r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
+    ]
+
+    def __init__(self, config: T5Config, patch_size, padding_idx, save_dir):
+        super().__init__(config)
+        self.model_dim = config.d_model
+        
+        self.padding_idx = padding_idx
+        self.out = open(os.path.join(save_dir, 'gate.txt'), 'w')
+
+        self.shared = nn.Embedding(config.vocab_size, config.d_model)
+        self.patch_num, self.patch_dim = patch_size
+
+        self.image_dense = nn.Linear(self.patch_dim, config.d_model)
+        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)
+        self.gate_dense = nn.Linear(2*config.hidden_size, config.hidden_size)
+        self.sigmoid = nn.Sigmoid()
+
+        encoder_config = copy.deepcopy(config)
+        encoder_config.is_decoder = False
+        encoder_config.use_cache = False
+        encoder_config.is_encoder_decoder = False
+        self.encoder = T5Stack(encoder_config, self.shared)
+
+        decoder_config = copy.deepcopy(config)
+        decoder_config.is_decoder = True
+        decoder_config.is_encoder_decoder = False
+        decoder_config.num_layers = config.num_decoder_layers
+        self.decoder = T5Stack(decoder_config, self.shared)
+
+        self.lm_head = nn.Linear(config.d_model, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+        # Model parallel
+        self.model_parallel = False
+        self.device_map = None
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        image_ids=None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        decoder_input_ids: Optional[torch.LongTensor] = None,
+        decoder_attention_mask: Optional[torch.BoolTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        decoder_head_mask: Optional[torch.FloatTensor] = None,
+        cross_attn_head_mask: Optional[torch.Tensor] = None,
+        encoder_outputs: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.FloatTensor], Seq2SeqLMOutput]:
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # FutureWarning: head_mask was separated into two input args - head_mask, decoder_head_mask
+        if head_mask is not None and decoder_head_mask is None:
+            if self.config.num_layers == self.config.num_decoder_layers:
+                warnings.warn(__HEAD_MASK_WARNING_MSG, FutureWarning)
+                decoder_head_mask = head_mask
+
+        # Encode if needed (training, first prediction pass)
+        if encoder_outputs is None:
+            # Convert encoder inputs in embeddings if needed
+            encoder_outputs = self.encoder(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                inputs_embeds=inputs_embeds,
+                head_mask=head_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            )
+
+        elif return_dict and not isinstance(encoder_outputs, BaseModelOutput):
+            encoder_outputs = BaseModelOutput(
+                last_hidden_state=encoder_outputs[0],
+                hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
+                attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
+            )
+
+
+        hidden_states = encoder_outputs[0]
+        
+        image_embedding = self.image_dense(image_ids)
+        image_att, _ = self.mha_layer(hidden_states, image_embedding, image_embedding)
+
+        merge = torch.cat([hidden_states, image_att], dim=-1)
+        gate = self.sigmoid(self.gate_dense(merge))
+        hidden_states = (1 - gate) * hidden_states + gate * image_att
+
+        if self.model_parallel:
+            torch.cuda.set_device(self.decoder.first_device)
+
+        if labels is not None and decoder_input_ids is None and decoder_inputs_embeds is None:
+            # get decoder inputs from shifting lm labels to the right
+            decoder_input_ids = self._shift_right(labels)
+
+        # Set device for model parallelism
+        if self.model_parallel:
+            torch.cuda.set_device(self.decoder.first_device)
+            hidden_states = hidden_states.to(self.decoder.first_device)
+            if decoder_input_ids is not None:
+                decoder_input_ids = decoder_input_ids.to(self.decoder.first_device)
+            if attention_mask is not None:
+                attention_mask = attention_mask.to(self.decoder.first_device)
+            if decoder_attention_mask is not None:
+                decoder_attention_mask = decoder_attention_mask.to(self.decoder.first_device)
+
+        # Decode
+        decoder_outputs = self.decoder(
+            input_ids=decoder_input_ids,
+            attention_mask=decoder_attention_mask,
+            inputs_embeds=decoder_inputs_embeds,
+            past_key_values=past_key_values,
+            encoder_hidden_states=hidden_states,
+            encoder_attention_mask=attention_mask,
+            head_mask=decoder_head_mask,
+            cross_attn_head_mask=cross_attn_head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = decoder_outputs[0]
+
+        # Set device for model parallelism
+        if self.model_parallel:
+            torch.cuda.set_device(self.encoder.first_device)
+            self.lm_head = self.lm_head.to(self.encoder.first_device)
+            sequence_output = sequence_output.to(self.lm_head.weight.device)
+
+        if self.config.tie_word_embeddings:
+            # Rescale output before projecting on vocab
+            # See https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/transformer/transformer.py#L586
+            sequence_output = sequence_output * (self.model_dim**-0.5)
+
+        lm_logits = self.lm_head(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss(ignore_index=-100)
+            loss = loss_fct(lm_logits.view(-1, lm_logits.size(-1)), labels.view(-1))
+            # TODO(thom): Add z_loss https://github.com/tensorflow/mesh/blob/fa19d69eafc9a482aff0b59ddd96b025c0cb207d/mesh_tensorflow/layers.py#L666
+
+        if not return_dict:
+            output = (lm_logits,) + decoder_outputs[1:] + encoder_outputs
+            return ((loss,) + output) if loss is not None else output
+
+        return Seq2SeqLMOutput(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=decoder_outputs.past_key_values,
+            decoder_hidden_states=decoder_outputs.hidden_states,
+            decoder_attentions=decoder_outputs.attentions,
+            cross_attentions=decoder_outputs.cross_attentions,
+            encoder_last_hidden_state=encoder_outputs.last_hidden_state,
+            encoder_hidden_states=encoder_outputs.hidden_states,
+            encoder_attentions=encoder_outputs.attentions,
+        )

requirements.txt

@@ -0,0 +1,11 @@
+huggingface-hub==0.0.12
+numpy==1.23.2
+openai==0.23.0
+pandas==1.4.3
+rouge==1.0.1
+sentence-transformers==2.2.2
+transformers==4.21.1
+nltk==3.6.6
+evaluate==0.4.0
+rouge==1.0.1
+rouge_score==0.1.2

run_inference.sh

@@ -0,0 +1,17 @@
+# rationale generation
+CUDA_VISIBLE_DEVICES=0,1 python main.py \
+    --model allenai/unifiedqa-t5-base \
+    --user_msg rationale --img_type detr \
+    --bs 8 --eval_bs 4 --eval_acc 10 --output_len 512 \
+    --final_eval --prompt_format QCM-LE \
+    --evaluate_dir models/rationale
+
+# answer inference
+CUDA_VISIBLE_DEVICES=0,1 python main.py \
+    --model allenai/unifiedqa-t5-base \
+    --user_msg answer --img_type detr \
+    --bs 8 --eval_bs 4 --eval_acc 10 --output_len 64 \
+    --final_eval --prompt_format QCMG-A \
+    --eval_le models/rationale/predictions_ans_eval.json \
+    --test_le models/rationale/predictions_ans_test.json \
+    --evaluate_dir models/answer

run_training.sh

@@ -0,0 +1,15 @@
+# rationale generation
+CUDA_VISIBLE_DEVICES=0,1 python main.py \
+    --model allenai/unifiedqa-t5-base \
+    --user_msg rationale --img_type detr \
+    --bs 8 --eval_bs 4 --eval_acc 10 --output_len 512 \
+    --final_eval --prompt_format QCM-LE
+
+# answer inference
+CUDA_VISIBLE_DEVICES=0,1 python main.py \
+    --model allenai/unifiedqa-t5-base \
+    --user_msg answer --img_type detr \
+    --bs 8 --eval_bs 4 --eval_acc 10 --output_len 64 \
+    --final_eval --prompt_format QCMG-A \
+    --eval_le experiments/rationale_allenai-unifiedqa-t5-base_detr_QCM-LE_lr5e-05_bs16_op512_ep20/predictions_ans_eval.json \
+    --test_le experiments/rationale_allenai-unifiedqa-t5-base_detr_QCM-LE_lr5e-05_bs16_op512_ep20/predictions_ans_test.json

utils_data.py

@@ -0,0 +1,228 @@
+import os
+from torch.utils.data import Dataset
+import os
+import json
+import numpy as np
+import torch
+from utils_prompt import *
+
+img_shape = {
+    "resnet": (512, 2048),
+    "clip": (49, 2048),
+    "detr": (100, 256),
+}
+
+def load_data_std(args):
+    problems = json.load(open(os.path.join(args.data_root, 'scienceqa/problems.json')))
+    pid_splits = json.load(open(os.path.join(args.data_root, 'scienceqa/pid_splits.json')))
+    captions = json.load(open(args.caption_file))["captions"]
+
+    for qid in problems:
+        problems[qid]['caption'] = captions[qid] if qid in captions else ""
+
+    train_qids = pid_splits['%s' % (args.train_split)]
+    val_qids = pid_splits['%s' % (args.val_split)]
+    test_qids = pid_splits['%s' % (args.test_split)]
+    print(f"number of train problems: {len(train_qids)}\n")
+    print(f"number of val problems: {len(val_qids)}\n")
+    print(f"number of test problems: {len(test_qids)}\n")
+
+    qids = {'train': train_qids, 'val':val_qids,'test':test_qids}
+    return problems, qids,
+
+def load_data_img(args):
+    problems = json.load(open(os.path.join(args.data_root, 'scienceqa/problems.json')))
+    pid_splits = json.load(open(os.path.join(args.data_root, 'scienceqa/pid_splits.json')))
+    captions = json.load(open(args.caption_file))["captions"]
+    name_maps = json.load(open('vision_features/name_map.json'))
+
+    # check
+    if args.img_type == "resnet":
+        image_features = np.load('vision_features/resnet.npy')
+        image_features = np.expand_dims(image_features, axis=1)
+        image_features = image_features.repeat(512, axis=1)
+    elif args.img_type == "clip":
+        image_features = np.load('vision_features/clip.npy')
+    elif args.img_type == "detr":
+        image_features = np.load('vision_features/detr.npy')
+    else:
+        image_features = np.load('vision_features/detr.npy')
+    print("img_features size: ", image_features.shape)
+
+    for qid in problems:
+        problems[qid]['caption'] = captions[qid] if qid in captions else ""
+
+    train_qids = pid_splits['%s' % (args.train_split)]
+    val_qids = pid_splits['%s' % (args.val_split)]
+    test_qids = pid_splits['%s' % (args.test_split)]
+    print(f"number of train problems: {len(train_qids)}\n")
+    print(f"number of val problems: {len(val_qids)}\n")
+    print(f"number of test problems: {len(test_qids)}\n")
+
+    qids = {'train': train_qids, 'val':val_qids,'test':test_qids}
+    return problems, qids, name_maps, image_features
+
+class ScienceQADatasetStd(Dataset):
+    """
+    Creating a custom dataset for reading the dataset and
+    loading it into the dataloader to pass it to the
+    neural network for finetuning the model
+
+    """
+
+    def __init__(
+        self, problems, qids, tokenizer, source_len, target_len, args, test_le=None
+    ):
+        self.tokenizer = tokenizer
+        self.data = {qid : problems[qid] for qid in qids}
+        self.source_len = source_len
+        self.summ_len = target_len
+        self.target_text = []
+        self.source_text = []
+        if test_le is not None:
+            test_le_data =json.load(open(test_le))["preds"]
+        else:
+            test_le_data = None
+        idx = 0
+        for qid in self.data:
+            if test_le_data is not None:
+                curr_le_data = test_le_data[idx]
+                idx += 1
+            else:
+                curr_le_data = None
+            prompt, target = build_train_pair(problems, qid, args, curr_le_data)
+            self.target_text.append(target)
+            self.source_text.append(prompt)
+
+    def __len__(self):
+        return len(self.target_text)
+
+    def __getitem__(self, index):
+        source_text = str(self.source_text[index])
+        target_text = str(self.target_text[index])
+
+        # cleaning data so as to ensure data is in string type
+        source_text = " ".join(source_text.split())
+        target_text = " ".join(target_text.split())
+
+        source = self.tokenizer.batch_encode_plus(
+            [source_text],
+            max_length=self.source_len,
+            pad_to_max_length=True,
+            truncation=True,
+            padding="max_length",
+            return_tensors="pt",
+        )
+        target = self.tokenizer.batch_encode_plus(
+            [target_text],
+            max_length=self.summ_len,
+            pad_to_max_length=True,
+            truncation=True,
+            padding="max_length",
+            return_tensors="pt",
+        )
+        source_ids = source["input_ids"].squeeze()
+        source_mask = source["attention_mask"].squeeze()
+        target_ids = target["input_ids"].squeeze().tolist()
+        
+        return {
+            "input_ids": source_ids,
+            "attention_mask": source_mask,
+            "labels": target_ids,
+        }
+
+
+class ScienceQADatasetImg(Dataset):
+    """
+    Creating a custom dataset for reading the dataset and
+    loading it into the dataloader to pass it to the
+    neural network for finetuning the model
+
+    """
+
+    def __init__(
+        self, problems, qids, name_maps, tokenizer, source_len, target_len, args, image_features, test_le=None
+    ):
+        """
+        Initializes a Dataset class
+
+        Args:
+            dataframe (pandas.DataFrame): Input dataframe
+            tokenizer (transformers.tokenizer): Transformers tokenizer
+            source_len (int): Max length of source text
+            target_len (int): Max length of target text
+            source_text (str): column name of source text
+            target_text (str): column name of target text
+        """
+        self.tokenizer = tokenizer
+        self.data = {qid : problems[qid] for qid in qids}
+        self.source_len = source_len
+        self.summ_len = target_len
+        self.target_text = []
+        self.source_text = []
+        self.image_ids = []
+        if test_le is not None:
+            test_le_data =json.load(open(test_le))["preds"]
+        else:
+            test_le_data = None
+        idx = 0
+        for qid in self.data:
+            if test_le_data is not None:
+                curr_le_data = test_le_data[idx]
+                idx += 1
+            else:
+                curr_le_data = None
+            prompt, target = build_train_pair(problems, qid, args, curr_le_data)
+            self.target_text.append(target)
+            self.source_text.append(prompt)
+            if str(qid) in name_maps:
+                i_vectors = image_features[int(name_maps[str(qid)])]
+                self.image_ids.append(i_vectors)
+            else:
+                shape = img_shape[args.img_type]
+                self.image_ids.append(np.zeros(shape))
+    
+    def __len__(self):
+        """returns the length of dataframe"""
+
+        return len(self.target_text)
+
+    def __getitem__(self, index):
+        """return the input ids, attention masks and target ids"""
+
+        source_text = str(self.source_text[index])
+        target_text = str(self.target_text[index])
+        image_ids = self.image_ids[index]
+
+        # cleaning data so as to ensure data is in string type
+        source_text = " ".join(source_text.split())
+        target_text = " ".join(target_text.split())
+
+        source = self.tokenizer.batch_encode_plus(
+            [source_text],
+            max_length=self.source_len,
+            pad_to_max_length=True,
+            truncation=True,
+            padding="max_length",
+            return_tensors="pt",
+        )
+        target = self.tokenizer.batch_encode_plus(
+            [target_text],
+            max_length=self.summ_len,
+            pad_to_max_length=True,
+            truncation=True,
+            padding="max_length",
+            return_tensors="pt",
+        )
+        source_ids = source["input_ids"].squeeze()
+        source_mask = source["attention_mask"].squeeze()
+        target_ids = target["input_ids"].squeeze().tolist()
+
+        image_ids = torch.tensor(image_ids).squeeze()
+        
+        return {
+            "input_ids": source_ids,
+            "attention_mask": source_mask,
+            "image_ids": image_ids,
+            "labels": target_ids,
+        }

utils_evaluate.py

@@ -0,0 +1,108 @@
+'''
+Adapted from https://github.com/lupantech/ScienceQA
+'''
+
+import os
+import json
+import argparse
+import warnings
+import pandas as pd
+from sentence_transformers import SentenceTransformer
+from evaluations import caculate_bleu, caculate_rouge, caculate_similariry
+
+warnings.filterwarnings('ignore')
+
+def get_acc_with_contion(res_pd, key, values):
+    if isinstance(values, list):
+        total_pd = res_pd[res_pd[key].isin(values)]
+    else:
+        total_pd = res_pd[res_pd[key] == values]
+    correct_pd = total_pd[total_pd['true_false'] == True]
+    acc = "{:.2f}".format(len(correct_pd) / len(total_pd) * 100)
+    return acc
+
+
+def get_scores(result_data, rationale_data, results_reference, data_file):
+    # read result file
+    results = result_data
+    num = len(results)
+    assert num == 4241
+    #print("number of questions:", num)
+
+    # read data file
+    sqa_data = json.load(open(data_file))
+
+    # construct pandas data
+    sqa_pd = pd.DataFrame(sqa_data).T
+    res_pd = sqa_pd[sqa_pd['split'] == 'test']  # test set
+
+    # update data
+    for index, row in res_pd.iterrows():
+
+        res_pd.loc[index, 'no_context'] = True if (not row['hint'] and not row['image']) else False
+        res_pd.loc[index, 'has_text'] = True if row['hint'] else False
+        res_pd.loc[index, 'has_image'] = True if row['image'] else False
+        res_pd.loc[index, 'has_text_image'] = True if (row['hint'] and row['image']) else False
+
+        label = row['answer']
+        pred = int(results[index])
+        res_pd.loc[index, 'pred'] = pred
+        res_pd.loc[index, 'true_false'] = (label == pred)
+
+    # accuracy scores
+    acc_average = len(res_pd[res_pd['true_false'] == True]) / num * 100
+    #assert result_file.split('_')[-1] == "{:.3f}.json".format(acc_average)
+
+
+    # rationale quality
+
+    ## BLEU
+    bleu1 = caculate_bleu(rationale_data, results_reference, gram=1)
+    bleu4 = caculate_bleu(rationale_data, results_reference, gram=4)
+
+    ## Rouge-L
+    rouge = caculate_rouge(rationale_data, results_reference)
+
+    ## Similarity
+    model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2').cuda()
+    similariry = caculate_similariry(rationale_data, results_reference, model)
+
+    scores = {
+            "answer":{
+                'acc_natural':
+                get_acc_with_contion(res_pd, 'subject', 'natural science'),
+                'acc_social':
+                get_acc_with_contion(res_pd, 'subject', 'social science'),
+                'acc_language':
+                get_acc_with_contion(res_pd, 'subject', 'language science'),
+                'acc_has_text':
+                get_acc_with_contion(res_pd, 'has_text', True),
+                'acc_has_image':
+                get_acc_with_contion(res_pd, 'has_image', True),
+                'acc_no_context':
+                get_acc_with_contion(res_pd, 'no_context', True),
+                'acc_grade_1_6':
+                get_acc_with_contion(res_pd, 'grade', ['grade1', 'grade2', 'grade3', 'grade4', 'grade5', 'grade6']),
+                'acc_grade_7_12':
+                get_acc_with_contion(res_pd, 'grade', ['grade7', 'grade8', 'grade9', 'grade10', 'grade11', 'grade12']),
+                'acc_average':
+                "{:.2f}".format(acc_average),
+            },
+            "rationale":{
+                'bleu1': bleu1 * 100,
+                'bleu4': bleu4 * 100,
+                'rouge': rouge * 100,
+                'similariry': similariry * 100,
+            }
+    }
+
+    return scores
+
+
+def print_scores(scores):
+    latex_output = ""
+    for key, score in scores.items():
+        print(f"{key[4:]}: \t{score}")
+        latex_output += f"& {score} "
+    latex_output += "\\\\"
+    print(latex_output)

utils_prompt.py

@@ -0,0 +1,240 @@
+'''
+Adapted from https://github.com/lupantech/ScienceQA
+'''
+
+from dataclasses import dataclass
+from typing import List, Optional
+
+def get_question_text(problem):
+    question = problem['question']
+    return question
+
+
+def get_context_text(problem, use_caption):
+    txt_context = problem['hint']
+    img_context = problem['caption'] if use_caption else ""
+    context = " ".join([txt_context, img_context]).strip()
+    if context == "":
+        context = "N/A"
+    return context
+
+
+def get_choice_text(probelm, options):
+    choices = probelm['choices']
+    choice_list = []
+    for i, c in enumerate(choices):
+        choice_list.append("({}) {}".format(options[i], c))
+    choice_txt = " ".join(choice_list)
+    #print(choice_txt)
+    return choice_txt
+
+def get_origin_answer(problem, options):
+    return problem['choices'][problem['answer']]
+
+def get_answer(problem, options):
+    return options[problem['answer']]
+
+
+def get_lecture_text(problem):
+    # \\n: GPT-3 can generate the lecture with more tokens.
+    lecture = problem['lecture'].replace("\n", "\\n")
+    return lecture
+
+
+def get_solution_text(problem):
+    # \\n: GPT-3 can generate the solution with more tokens
+    solution = problem['solution'].replace("\n", "\\n")
+    return solution
+
+
+def create_one_example(format, question, context, choice, answer, lecture, solution, test_example=True, WithOutput = False, curr_le_data=None):
+
+    input_format, output_format = format.split("-")
+
+    ## Inputs
+    if input_format == "CQM":
+        input = f"Context: {context}\nQuestion: {question}\nOptions: {choice}\n"
+    elif input_format == "QCM":
+        input = f"Question: {question}\nContext: {context}\nOptions: {choice}\n"
+    elif input_format == "QM":
+        input = f"Question: {question}\nOptions: {choice}\n"
+    elif input_format == "QC":
+        input = f"Question: {question}\nContext: {context}\n"
+    elif input_format == "QCMG":
+        if curr_le_data is not None:
+            input = f"Question: {question}\nContext: {context}\nOptions: {choice}\n{curr_le_data}\n"
+        else:
+            input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nSolution: {lecture} {solution}\n"
+    elif input_format == "CQMG":
+        if curr_le_data is not None:
+            input = f"Context: {context}\nQuestion: {question}\nOptions: {choice}\n{curr_le_data}\n"
+        else:
+            input = f"Context: {context}\nQuestion: {question}\nOptions: {choice}\nSolution: {lecture} {solution}\n"
+    # upper bound experiment
+    elif input_format == "QCML":
+        input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nBECAUSE: {lecture}\n"
+    elif input_format == "QCME":
+        input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nBECAUSE: {solution}\n"
+    elif input_format == "QCMLE":
+        input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nBECAUSE: {lecture} {solution}\n"
+
+    elif input_format == "QCLM":
+        input = f"Question: {question}\nContext: {context}\nBECAUSE: {lecture}\nOptions: {choice}\n"
+    elif input_format == "QCEM":
+        input = f"Question: {question}\nContext: {context}\nBECAUSE: {solution}\nOptions: {choice}\n"
+    elif input_format == "QCLEM":
+        input = f"Question: {question}\nContext: {context}\nBECAUSE: {lecture} {solution}\nOptions: {choice}\n"
+    elif input_format == "QCMA":
+        input = f"Question: {question}\nContext: {context}\nOptions: {choice}\nAnswer: The answer is {answer}.\n"
+    elif input_format == "QCA":
+        input = f"Question: {question}\nContext: {context}\nAnswer: The answer is {answer}. \nBECAUSE:"
+
+    # Outputs
+    if test_example:
+        if output_format == 'A':
+            output = "Answer:"
+        elif output_format == 'E':
+            output = "Solution:"
+        else:
+            output = "Solution:"
+    elif output_format == 'A':
+        output = f"Answer: The answer is {answer}."
+
+    elif output_format == 'AL':
+        output = f"Answer: The answer is {answer}. BECAUSE: {solution}"
+    elif output_format == 'AE':
+        output = f"Answer: The answer is {answer}. BECAUSE: {lecture}"
+    elif output_format == 'ALE':
+        output = f"Answer: The answer is {answer}. BECAUSE: {lecture} {solution}"
+    elif output_format == 'AEL':
+        output = f"Answer: The answer is {answer}. BECAUSE: {solution} {lecture}"
+
+    elif output_format == 'LA':
+        output = f"Answer: {lecture} The answer is {answer}."
+    elif output_format == 'EA':
+        output = f"Answer: {solution} The answer is {answer}."
+    elif output_format == 'LEA':
+        output = f"Answer: {lecture} {solution} The answer is {answer}."
+    elif output_format == 'ELA':
+        output = f"Answer: {solution} {lecture} The answer is {answer}."
+
+    elif output_format == 'LE':
+        output = f"Solution: {lecture} {solution}."
+
+    elif output_format == 'E':
+        output = f"Solution: {solution}"
+        
+    
+    if WithOutput:
+        if output.endswith("BECAUSE:"):
+            output = output.replace("BECAUSE:", "").strip()
+        if output_format == 'E':
+            text = input + f'Solution:'
+        elif output_format == 'A':
+            text = input + f'Answer:'
+        else: 
+            text = input + f'Solution:'
+        text = text.replace("  ", " ").strip()
+        output = output.replace("  ", " ").strip()
+        return text, output
+        
+        
+    text = input + output
+    text = text.replace("  ", " ").strip()
+    if text.endswith("BECAUSE:"):
+        text = text.replace("BECAUSE:", "").strip()
+    return text
+
+
+def build_prompt(problems, shot_qids, test_qid, args):
+
+    examples = []
+
+    # n-shot training examples
+    for qid in shot_qids:
+        question = get_question_text(problems[qid])
+        context = get_context_text(problems[qid], args.use_caption)
+        choice = get_choice_text(problems[qid], args.options)
+        answer = get_answer(problems[qid], args.options)
+        lecture = get_lecture_text(problems[qid])
+        solution = get_solution_text(problems[qid])
+
+        train_example = create_one_example(args.prompt_format,
+                                           question,
+                                           context,
+                                           choice,
+                                           answer,
+                                           lecture,
+                                           solution,
+                                           test_example=False)
+        examples.append(train_example)
+
+    # test example
+    question = get_question_text(problems[test_qid])
+    context = get_context_text(problems[test_qid], args.use_caption)
+    choice = get_choice_text(problems[test_qid], args.options)
+    answer = get_answer(problems[test_qid], args.options)
+    lecture = get_lecture_text(problems[test_qid])
+    solution = get_solution_text(problems[test_qid])
+
+    test_example = create_one_example(args.prompt_format,
+                                      question,
+                                      context,
+                                      choice,
+                                      answer,
+                                      lecture,
+                                      solution,
+                                      test_example=True)
+    examples.append(test_example)
+
+    # create the prompt input
+    prompt_input = '\n\n'.join(examples)
+
+    return prompt_input
+
+def build_train_pair(problems, test_qid, args, curr_le_data=None):
+
+    examples = []
+
+    # test example
+    question = get_question_text(problems[test_qid])
+    context = get_context_text(problems[test_qid], args.use_caption)
+    choice = get_choice_text(problems[test_qid], args.options)
+    
+    lecture = get_lecture_text(problems[test_qid])
+    solution = get_solution_text(problems[test_qid])
+
+    # answer_text = get_origin_answer(problems[test_qid], args.options)
+    answer_option = get_answer(problems[test_qid], args.options)
+    answer = "(" + answer_option + ")"
+    
+    test_example, target = create_one_example(args.prompt_format,
+                                      question,
+                                      context,
+                                      choice,
+                                      answer,
+                                      lecture,
+                                      solution,
+                                      test_example=False,WithOutput = True, curr_le_data=curr_le_data)
+    examples.append(test_example)
+    
+    target = target.replace("Answer:", "").strip()
+    # create the prompt input
+    prompt_input = '\n\n'.join(examples)
+
+    return prompt_input, target
+
+@dataclass(frozen=True)
+class InputFeatures:
+    """
+    A single set of features of data.
+    Property names are the same names as the corresponding inputs to a model.
+    """
+
+    input_ids: List[List[int]]
+    attention_mask: Optional[List[List[int]]]
+    token_type_ids: Optional[List[List[int]]]
+    le_input_ids: List[List[int]]
+    le_attention_mask: Optional[List[List[int]]]
+    le_token_type_ids: Optional[List[List[int]]]
+    label: Optional[int]