build: apply a typical project boilerplate of BigScience repo

.github/workflows/code_quality.yml

@@ -0,0 +1,23 @@
+name: Code quality
+
+on:
+    push:
+        branches: [ main ]
+    pull_request:
+        branches: [ main ]
+
+jobs:
+    build:
+        runs-on: ubuntu-latest
+        steps:
+            - uses: actions/checkout@v2
+            - name: Set up Python 3.8
+              uses: actions/setup-python@v2
+              with:
+                  python-version: 3.8
+            - name: Install dependencies
+              run: |
+                  python -m pip install -U pip
+                  python -m pip install -r requirements-dev.txt
+            - name: Check code quality
+              run: make quality

.github/workflows/test.yml

@@ -0,0 +1,25 @@
+name: Test
+
+on:
+    push:
+        branches: [main]
+    pull_request:
+        branches: [main]
+
+jobs:
+    test:
+        name: Test
+        runs-on: ubuntu-latest
+        steps:
+            - uses: actions/checkout@v2
+            - name: Set up Python 3.8
+              uses: actions/setup-python@v2
+              with:
+                  python-version: 3.8
+            - name: Install dependencies
+              run: |
+                  python -m pip install -U pip
+                  python -m pip install .
+                  python -m pip install -r requirements-dev.txt
+            - name: Test
+              run: python -m pytest tests

Makefile

@@ -0,0 +1,12 @@
+.PHONY: quality style
+
+check_dirs := .
+
+quality: # Check that source code meets quality standards
+	black --check $(check_dirs)
+	isort --check-only $(check_dirs)
+	flake8 $(check_dirs) --max-line-length 119
+
+style: # Format source code automatically
+	black $(check_dirs)
+	isort $(check_dirs)

evaluation/eval.py

@@ -1,39 +1,31 @@
+import os
 from dataclasses import dataclass, field
 from datetime import datetime
-from typing import Optional, List
-import os
+from typing import List, Optional
 
 import torch
-from transformers import (
-    HfArgumentParser,
-    AutoTokenizer,
-    AutoModelForCausalLM,
-    TrainingArguments,
-    set_seed,
-)
-import evaluation.tasks  # needed for AutoTask.__subclass__() to work correctly
+from transformers import AutoModelForCausalLM, AutoTokenizer, HfArgumentParser, TrainingArguments, set_seed
+
+import evaluation.tasks  # noqa: F401; needed for AutoTask.__subclass__() to work correctly
 from evaluation.tasks.auto_task import AutoTask
 from evaluation.utils.log import get_logger
 
 
 @dataclass
 class EvaluationArguments:
     """
-        Arguments for any adjustable params in this evaluation script
+    Arguments for any adjustable params in this evaluation script
     """
+
     model_name_or_path: str = field(
         metadata={"help": "The model checkpoint that we want to evaluate, could be name or the path."}
     )
-    eval_tasks: List[str] = field(
-        metadata={"help": "A list of tasks to run the evaluation on, e.g. tydiqa_secondary"}
-    )
+    eval_tasks: List[str] = field(metadata={"help": "A list of tasks to run the evaluation on, e.g. tydiqa_secondary"})
     config_name: Optional[str] = field(
-        default=None,
-        metadata={"help": "Pretrained config name or path if not the same as model_name."}
+        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name."}
     )
     tokenizer_name: Optional[str] = field(
-        default=None,
-        metadata={"help": "Pretrained tokenizer name or path if not the same as model_name."}
+        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name."}
     )
     tag: Optional[str] = field(
         default=None,
@@ -50,8 +42,8 @@ def main():
     eval_args, train_args = parser.parse_args_into_dataclasses()
 
     if not eval_args.eval_tasks:
-        raise ValueError('Must provide at least one eval task!')
-    
+        raise ValueError("Must provide at least one eval task!")
+
     # initialize device
     device = torch.device(train_args.device)
 
@@ -65,7 +57,8 @@ def main():
     tokenizer.padding_side = "left"
 
     model = AutoModelForCausalLM.from_pretrained(
-        eval_args.model_name_or_path, pad_token_id=tokenizer.eos_token,
+        eval_args.model_name_or_path,
+        pad_token_id=tokenizer.eos_token,
     )
     model.config.pad_token_id = model.config.eos_token_id
     model.resize_token_embeddings(len(tokenizer))

evaluation/tasks/__init__.py

@@ -2,6 +2,7 @@
 # source: https://stackoverflow.com/questions/3365740/how-to-import-all-submodules
 import pkgutil
 
+
 __all__ = []
 for loader, module_name, is_pkg in pkgutil.walk_packages(__path__):
     __all__.append(module_name)

evaluation/tasks/auto_task.py

@@ -1,6 +1,6 @@
+import os
 from abc import ABC, abstractmethod
 from typing import Dict
-import os
 
 import torch
 from transformers import AutoTokenizer, PreTrainedModel, PreTrainedTokenizerFast

evaluation/tasks/tydiqa_primary/tydiqa_primary.py

@@ -4,6 +4,7 @@
 from jinja2 import Template
 from torch.utils.data import Dataset
 
+
 TEMPLATE = Template(
     """
     {%- set _blank=["passage", "text", "text snippet", "context"]|random -%}
@@ -16,30 +17,31 @@
     {{"\n"}}{{context}}
     {%- endif -%}
     {{"\n"}}Answer: 
-    """
+    """  # noqa W291
 )
 
+
 class TyDiQADataset(Dataset):
     def __init__(self, data, tokenizer, target_langs):
         super(TyDiQADataset, self).__init__()
         self.items = []
-        
+
         for sample_id, sample in enumerate(data):
             lang = sample["id"].split("-")[0]
             if lang in target_langs:
                 # Filter out samples in languages that are not used during training
                 prompt = TEMPLATE.render(
-                    id       = sample["id"],
-                    context  = sample["context"],
-                    question = sample["question"],
+                    id=sample["id"],
+                    context=sample["context"],
+                    question=sample["question"],
                 )
                 prompt = prompt.strip()  # Remove trailing white space and newline
 
                 # Tokenize and construct this sample
                 inputs = tokenizer(
                     prompt,
                     padding=True,
-                    return_tensors='pt',
+                    return_tensors="pt",
                 )
                 self.items.append(
                     {
@@ -48,12 +50,12 @@ def __init__(self, data, tokenizer, target_langs):
                         "input_ids": inputs["input_ids"],
                         "attention_mask": inputs["attention_mask"],
                         "input_len": inputs["attention_mask"].shape[1],
-                        "target_answer": [ans.lower() for ans in sample["answers"]['text']],
+                        "target_answer": [ans.lower() for ans in sample["answers"]["text"]],
                     }
                 )
-    
+
     def __len__(self):
         return len(self.items)
-    
+
     def __getitem__(self, index):
         return self.items[index]

evaluation/tasks/tydiqa_secondary/tydiqa_secondary.py

@@ -1,14 +1,13 @@
 # Module for any additional processing required for the TyDi QA dataset
 # HuggingFace dataset link: https://huggingface.co/datasets/tydiqa
-from typing import Dict
-
+from datasets import load_dataset
 from jinja2 import Template
 from torch.utils.data import Dataset
-from datasets import load_dataset
 from tqdm import tqdm
 
 from evaluation.tasks.auto_task import AutoTask
 
+
 TEMPLATE = Template(
     """
     {%- set _blank=["passage", "text", "text snippet", "context"]|random -%}
@@ -21,7 +20,7 @@
     {{"\n"}}{{context}}
     {%- endif -%}
     {{"\n"}}Answer: 
-    """
+    """  # noqa W291
 )
 
 
@@ -31,23 +30,23 @@ def __init__(self, tokenizer, target_langs):
         assert tokenizer.pad_token == tokenizer.eos_token
         tydiqa = load_dataset("tydiqa", "secondary_task", split="validation")
         self.items = []
-        
+
         for sample in tydiqa:
             lang = sample["id"].split("-")[0]
             if lang in target_langs:
                 # Filter out samples in languages that are not used during training
                 prompt = TEMPLATE.render(
-                    id       = sample["id"],
-                    context  = sample["context"],
-                    question = sample["question"],
+                    id=sample["id"],
+                    context=sample["context"],
+                    question=sample["question"],
                 )
                 prompt = prompt.strip()  # Remove trailing white space and newline
 
                 # Tokenize and construct this sample
                 inputs = tokenizer(
                     prompt,
                     padding=True,
-                    return_tensors='pt',
+                    return_tensors="pt",
                 )
                 self.items.append(
                     {
@@ -56,27 +55,27 @@ def __init__(self, tokenizer, target_langs):
                         "input_ids": inputs["input_ids"],
                         "attention_mask": inputs["attention_mask"],
                         "input_len": inputs["attention_mask"].shape[1],
-                        "target_answer": [ans.lower() for ans in sample["answers"]['text']],
+                        "target_answer": [ans.lower() for ans in sample["answers"]["text"]],
                     }
                 )
-    
+
     def __len__(self):
         return len(self.items)
-    
+
     def __getitem__(self, index):
         return self.items[index]
 
 
 class TydiqaSecondaryTask(AutoTask):
     @staticmethod
     def get_display_name() -> str:
-        return 'tydiqa_secondary'
+        return "tydiqa_secondary"
 
     def evaluate(self) -> None:
         dataset = TyDiQADataset(self.tokenizer, target_langs=self.task_config["target_langs"])
 
         substring_matches = 0
-        for sample in tqdm(dataset, desc=f'Evaluating {self.get_display_name()}'):
+        for sample in tqdm(dataset, desc=f"Evaluating {self.get_display_name()}"):
             output = self.model.generate(
                 input_ids=sample["input_ids"].to(self.device),
                 attention_mask=sample["attention_mask"].to(self.device),
@@ -91,6 +90,4 @@ def evaluate(self) -> None:
             substring_match = any([target_answer in predicted_answer.lower() for target_answer in target_answers])
             substring_matches += substring_match
 
-        self.metrics = {
-            "substring_matches": substring_matches / len(dataset) * 100
-        }
+        self.metrics = {"substring_matches": substring_matches / len(dataset) * 100}

evaluation/utils/log.py

@@ -4,11 +4,11 @@
 def get_logger():
     logger = logging.getLogger("evaluation")
     formatter = logging.Formatter(
-        '%(asctime)s - %(name)s - %(levelname)s - %(message)s', 
+        "%(asctime)s - %(name)s - %(levelname)s - %(message)s",
         datefmt="%m/%d/%Y %H:%M:%S",
     )
     handler = logging.StreamHandler()
     handler.setFormatter(formatter)
     logger.addHandler(handler)
     logger.setLevel(logging.INFO)
-    return logger
+    return logger