Merge branch 'big-refactor' of https://github.com/EleutherAI/lm-evaluation-harness into squadv2

lm_eval/filters/decontamination.py

@@ -9,7 +9,7 @@ class DecontaminationFilter(Filter):
 
     name = "track_decontamination"
 
-    def __init__(self, path):
+    def __init__(self, path) -> None:
         """
 
         TODO: make sure only ever run one time on the train set (should this be cached as a class var? keyed by value for "path").
@@ -17,7 +17,7 @@ class DecontaminationFilter(Filter):
         """
         self._decontam_results = None
 
-    def apply(self, reps):
+    def apply(self, resps, docs) -> None:
         """
         Return {"no_contamination", "only_contamination"} keys for the 2 different subsets
         """

lm_eval/filters/extraction.py

@@ -6,7 +6,9 @@ from lm_eval.api.filter import Filter
 class RegexFilter(Filter):
     """ """
 
-    def __init__(self, regex_pattern=r"#### (\-?[0-9\.\,]+)", fallback="[invalid]"):
+    def __init__(
+        self, regex_pattern: str = r"#### (\-?[0-9\.\,]+)", fallback: str = "[invalid]"
+    ) -> None:
         """
         pass a string `regex` to run `re.compile(r"regex")` on.
         `fallback` defines the output returned if no matches for the regex are located.
@@ -15,7 +17,7 @@ class RegexFilter(Filter):
         self.regex = re.compile(regex_pattern)
         self.fallback = fallback
 
-    def apply(self, resps):
+    def apply(self, resps, docs):
         # here, we assume we have a list, in which each element is
         # a list of model responses for some particular input/target pair.
         # so we process each of these (same input/target response sets)
@@ -41,12 +43,11 @@ class RegexFilter(Filter):
 class WhitespaceFilter(Filter):
     """ """
 
-    def __init__(self):
+    def __init__(self) -> None:
         pass
 
-    def apply(self, resps):
+    def apply(self, resps, docs):
         def filter_set(inst):
-
             filtered_resp = []
             for resp in inst:
                 if resp.startswith(" "):

lm_eval/filters/selection.py

@@ -4,12 +4,12 @@ from lm_eval.api.filter import Filter
 
 
 class TakeFirstFilter(Filter):
-    def __init__(self):
+    def __init__(self) -> None:
         """
         Can define custom behavior here, if an individual instantiation of a Filter class should have state.
         """
 
-    def apply(self, resps):
+    def apply(self, resps, docs):
         """
         Assuming each entry of `resps` is a list of model responses, we discard all but the first response.
         """
@@ -17,13 +17,12 @@ class TakeFirstFilter(Filter):
 
 
 class TakeKFilter(Filter):
-    def __init__(self, *args, **kwargs):
-
+    def __init__(self, *args, **kwargs) -> None:
         self.k = kwargs.pop("k")
 
         super().__init__(*args, **kwargs)
 
-    def apply(self, resps):
+    def apply(self, resps, docs):
         # check we have at least k responses per doc, else we can't take the first k
         assert (
             len(resps[0]) >= self.k
@@ -32,12 +31,12 @@ class TakeKFilter(Filter):
 
 
 class MajorityVoteFilter(Filter):
-    def __init__(self):
+    def __init__(self) -> None:
         """
         Can define custom behavior here, if an individual instantiation of a Filter class should have state.
         """
 
-    def apply(self, resps):
+    def apply(self, resps, docs):
         """
         Each entry of `resps` is a list of model responses.
         We select the response that occurs most frequently in each entry of `resps`.

lm_eval/logger.py

@@ -6,3 +6,5 @@ logging.basicConfig(
     level=logging.INFO,
 )
 eval_logger = logging.getLogger("lm-eval")
+
+SPACING = " " * 47

lm_eval/models/anthropic_llms.py

@@ -76,7 +76,7 @@ class AnthropicLM(LM):
         max_tokens_to_sample: int = 256,
         temperature: float = 0,  # defaults to 1
         **kwargs,  # top_p, top_k, etc.
-    ):
+    ) -> None:
         """Anthropic API wrapper.
 
         :param model: str
@@ -135,11 +135,10 @@ please install anthropic via `pip install lm-eval[anthropic]` or `pip install -e
     def tok_decode(self, tokens: List[int]) -> str:
         return self.tokenizer.decode(tokens)
 
-    def _loglikelihood_tokens(self, requests, disable_tqdm=False):
+    def _loglikelihood_tokens(self, requests, disable_tqdm: bool = False):
         raise NotImplementedError("No support for logits.")
 
     def greedy_until(self, requests) -> List[str]:
-
         if not requests:
             return []
 

lm_eval/models/dummy.py

@@ -5,7 +5,7 @@ from lm_eval.api.registry import register_model
 
 @register_model("dummy")
 class DummyLM(LM):
-    def __init__(self):
+    def __init__(self) -> None:
         super().__init__()
 
     @classmethod

lm_eval/models/huggingface.py

+import os
+
 import torch
 import transformers
 from transformers.models.auto.modeling_auto import (
@@ -20,7 +22,7 @@ from lm_eval.api.registry import register_model
 
 from lm_eval.utils import MultiTokenEOSCriteria, stop_sequences_criteria
 
-from accelerate import Accelerator, find_executable_batch_size
+from accelerate import Accelerator, find_executable_batch_size, DistributedType
 from typing import List, Optional, Union
 
 
@@ -67,6 +69,7 @@ class HFLM(LM):
         revision: Optional[str] = "main",
         subfolder: Optional[str] = None,
         tokenizer: Optional[str] = None,
+        truncation: Optional[bool] = False,
         max_length: Optional[int] = None,
         device: Optional[str] = "cuda",
         dtype: Optional[Union[str, torch.dtype]] = "auto",
@@ -75,6 +78,7 @@ class HFLM(LM):
         low_cpu_mem_usage: Optional[bool] = True,
         trust_remote_code: Optional[bool] = False,
         use_fast_tokenizer: Optional[bool] = True,
+        cache_dir: Optional[Union[str, os.PathLike]] = None,
         # arguments used for splitting a model across GPUs naively.
         # only used if `parallelize=True`.
         parallelize: Optional[bool] = False,
@@ -90,7 +94,7 @@ class HFLM(LM):
         bnb_4bit_compute_dtype: Optional[Union[str, torch.dtype]] = None,
         gptq: Optional[Union[bool, str]] = False,
         gptq_use_triton: Optional[bool] = False,
-    ):
+    ) -> None:
         super().__init__()
 
         assert isinstance(device, str)
@@ -103,17 +107,20 @@ class HFLM(LM):
         if not (parallelize or accelerator.num_processes > 1):
             # use user-passed device
             device_list = set(
-                ["cuda", "cpu", "mps"]
+                ["cuda", "cpu"]
                 + [f"cuda:{i}" for i in range(torch.cuda.device_count())]
+                + ["mps", "mps:0"]
             )
             if device:
                 if device not in device_list:
                     device = int(device)
                 self._device = torch.device(device)
                 eval_logger.info(f"Using device '{device}'")
-                if device == "mps":
+                if device in ("mps", "mps:0") and "dev" not in torch.__version__:
                     eval_logger.info(
-                        "MPS is still in beta and only supports float32; setting dtype to float32."
+                        "MPS: Setting dtype to float32. To use float16 with MPS, please install a nightly build of "
+                        "PyTorch: pip3 install --pre torch torchvision torchaudio --index-url "
+                        "https://download.pytorch.org/whl/nightly/cpu"
                     )
             else:
                 eval_logger.info("Device not specified")
@@ -240,6 +247,8 @@ class HFLM(LM):
             use_fast=use_fast_tokenizer,
         )
 
+        self.truncation = truncation
+
         self.vocab_size = self.tokenizer.vocab_size
         self.tokenizer.pad_token_id = self.tokenizer.eos_token_id
 
@@ -288,6 +297,13 @@ class HFLM(LM):
                     eval_logger.info(
                         "Failed to place model onto specified device. This may be because the model is quantized via `bitsandbytes`. If the desired GPU is being used, this message is safe to ignore."
                     )
+            else:
+                assert accelerator.distributed_type in [
+                    DistributedType.FSDP,
+                    DistributedType.MULTI_GPU,
+                ], "Unsupported distributed type provided. Only DDP and FSDP are supported."
+                if accelerator.distributed_type == DistributedType.FSDP:
+                    self._model = accelerator.prepare(self.model)
                 else:
                     self._model = accelerator.prepare_model(
                         self.model, evaluation_mode=True
@@ -334,7 +350,7 @@ class HFLM(LM):
         return self._DEFAULT_MAX_LENGTH
 
     @property
-    def max_gen_toks(self):
+    def max_gen_toks(self) -> int:
         return 256
 
     @property
@@ -353,7 +369,7 @@ class HFLM(LM):
     def world_size(self):
         return self._world_size
 
-    def _detect_batch_size(self, requests=None, pos=0):
+    def _detect_batch_size(self, requests=None, pos: int = 0):
         if requests:
             _, context_enc, continuation_enc = requests[pos]
             max_length = len(
@@ -419,7 +435,11 @@ class HFLM(LM):
         return encoding
 
     def tok_batch_encode(
-        self, strings: List[str], padding_side="left", left_truncate_len=None
+        self,
+        strings: List[str],
+        padding_side: str = "left",
+        left_truncate_len: int = None,
+        truncation: bool = False,
     ):
         # encode a batch of strings. converts to tensors and pads automatically, unlike tok_encode.
         old_padding_side = self.tokenizer.padding_side
@@ -432,6 +452,7 @@ class HFLM(LM):
 
         encoding = self.tokenizer(
             strings,
+            truncation=truncation,
             padding="longest",
             return_tensors="pt",
             add_special_tokens=add_special_tokens,
@@ -595,7 +616,9 @@ class HFLM(LM):
 
         return loglikelihoods
 
-    def _loglikelihood_tokens(self, requests, disable_tqdm=False, override_bs=None):
+    def _loglikelihood_tokens(
+        self, requests, disable_tqdm: bool = False, override_bs=None
+    ):
         # TODO: implement some kind of efficient-request-middleware that lumps together requests with the same context
         res = []
 
@@ -856,7 +879,9 @@ class HFLM(LM):
 
                 # encode, pad, and truncate contexts for this batch
                 context_enc, attn_masks = self.tok_batch_encode(
-                    contexts, left_truncate_len=max_ctx_len
+                    contexts,
+                    left_truncate_len=max_ctx_len,
+                    truncation=self.truncation,
                 )
                 context_enc = context_enc.to(self.device)
                 attn_masks = attn_masks.to(self.device)

lm_eval/models/openai_completions.py

@@ -69,7 +69,7 @@ class OpenaiCompletionsLM(LM):
         engine: str = "text-davinci-003",
         truncate: bool = False,
         batch_size: int = 1,
-    ):
+    ) -> None:
         """
 
         :param engine: str
@@ -99,12 +99,12 @@ class OpenaiCompletionsLM(LM):
         return self.end_of_text_token_id
 
     @property
-    def max_length(self):
+    def max_length(self) -> int:
         # Note: the OpenAI API supports up to 2049 tokens, with the first token being the first input token
         return 2048
 
     @property
-    def max_gen_toks(self):
+    def max_gen_toks(self) -> int:
         return 256
 
     @property
@@ -152,7 +152,7 @@ class OpenaiCompletionsLM(LM):
         return self._loglikelihood_tokens(new_reqs)
 
     def _loglikelihood_tokens(
-        self, requests, disable_tqdm=False
+        self, requests, disable_tqdm: bool = False
     ) -> List[Tuple[float, bool]]:
         res = []
 

lm_eval/models/textsynth.py

@@ -41,7 +41,7 @@ def textsynth_completion(**kwargs):
 
 @register_model("textsynth")
 class TextSynthLM(LM):
-    def __init__(self, engine, truncate=False):
+    def __init__(self, engine, truncate: bool = False) -> None:
         """
         :param engine: str
             TextSynth API engine (e.g. `gptj_6B`)
@@ -62,12 +62,12 @@ class TextSynthLM(LM):
         raise NotImplementedError()
 
     @property
-    def max_length(self):
+    def max_length(self) -> int:
         # NOTE: Turn on truncation to avoid errors on long inputs.
         return 2048
 
     @property
-    def max_gen_toks(self):
+    def max_gen_toks(self) -> int:
         return 256
 
     @property

lm_eval/prompts/__init__.py

+import ast
+
+from typing import Dict
 from lm_eval import utils
 from lm_eval.logger import eval_logger
 
@@ -5,7 +8,7 @@ from lm_eval.logger import eval_logger
 # Stores prompts in a dictionary indexed by 2 levels:
 # prompt category name, and prompt name.
 # This allows us to access prompts
-PROMPT_REGISTRY = {
+PROMPT_REGISTRY: Dict[str, Dict[str, str]] = {
     "qa-basic": {
         "question-newline-answer": "Question: {{question}}\nAnswer:",
         "q-newline-a": "Q: {{question}}\nA:",
@@ -13,7 +16,7 @@ PROMPT_REGISTRY = {
 }
 
 
-def get_prompt(prompt_id: str, dataset_name=None, subset_name=None):
+def get_prompt(prompt_id: str, dataset_name: str = None, subset_name: str = None):
     # unpack prompt name
     category_name, prompt_name = prompt_id.split(":")
     if subset_name is None:
@@ -63,6 +66,12 @@ def load_prompt_list(use_prompt: str, dataset_name=None, subset_name=None, **kwa
     else:
         prompts = DatasetTemplates(dataset_name=dataset_name, subset_name=subset_name)
 
-    category_name, prompt_name = use_prompt.split(":")
+    category_name, *prompt_name = use_prompt.split(":")
+    # TODO allow to multiple prompt naming
+    # if len(prompt_name) > 1:
+    #     prompt_list = []
+    #     for prompt in prompt_name:
+    #         prompt_list.append(utils.pattern_match(prompt_name, prompts.all_template_names))
+    # else:
     prompt_list = utils.pattern_match(prompt_name, prompts.all_template_names)
     return [":".join([category_name, prompt]) for prompt in prompt_list]

lm_eval/tasks/README.md

@@ -5,8 +5,8 @@ Boxes should be checked iff tasks are implemented in the refactor and tested for
 
 - [x] Glue
 - [x] SuperGlue
-- [ ] CoQA (Lintang)
-- [ ] DROP (Lintang)
+- [x] CoQA
+- [x] DROP
 - [x] ~~Lambada~~
 - [x] Lambada (Cloze variants)
 - [x] ~~Lambada (Multilingual)~~
@@ -29,7 +29,7 @@ Boxes should be checked iff tasks are implemented in the refactor and tested for
 - [x] HeadQA
 - [x] MathQA
 - [x] WebQs
-- [ ] WSC273 (Lintang)
+- [x] WSC273
 - [x] Winogrande
 - [x] ANLI
 - [x] Hendrycks Ethics (missing some tasks/metrics, see PR 660: <https://github.com/EleutherAI/lm-evaluation-harness/pull/660> for more info)
@@ -38,11 +38,11 @@ Boxes should be checked iff tasks are implemented in the refactor and tested for
 - [x] TruthfulQA (gen)
 - [ ] MuTual
 - [ ] Hendrycks Math (Hailey)
-- [ ] Asdiv
+- [x] Asdiv
 - [ ] GSM8k
 - [x] Arithmetic
 - [ ] MMMLU (Hailey)
-- [ ] Translation (WMT) suite (Hailey)
+- [x] Translation (WMT) suite
 - [x] Unscramble
 - [x] ~~Pile (perplexity)~~
 - [x] BLiMP
@@ -56,7 +56,7 @@ Boxes should be checked iff tasks are implemented in the refactor and tested for
 - [x] XWinograd
 - [x] PAWS-X
 - [x] XNLI
-- [ ] MGSM (Lintang)
+- [x] MGSM
 - [ ] SCROLLS
 - [x] Babi
 

lm_eval/tasks/__init__.py

 import os
 import yaml
-from typing import List, Union
+from typing import List, Union, Dict
 
 from lm_eval import utils
 from lm_eval import prompts
@@ -15,7 +15,7 @@ from lm_eval.api.registry import (
 )
 
 
-def register_configurable_task(config):
+def register_configurable_task(config: Dict[str, str]) -> int:
     SubClass = type(
         config["task"] + "ConfigurableTask",
         (ConfigurableTask,),
@@ -38,7 +38,7 @@ def register_configurable_task(config):
     return 0
 
 
-def check_prompt_config(config):
+def check_prompt_config(config: Dict[str, str]) -> List[Dict[str, str]]:
     all_configs = []
     if "use_prompt" in config:
         prompt_list = prompts.load_prompt_list(
@@ -69,14 +69,14 @@ def check_prompt_config(config):
     return all_configs
 
 
-def get_task_name_from_config(task_config):
+def get_task_name_from_config(task_config: Dict[str, str]) -> str:
     if "dataset_name" in task_config:
         return "{dataset_path}_{dataset_name}".format(**task_config)
     else:
         return "{dataset_path}".format(**task_config)
 
 
-def include_task_folder(task_dir):
+def include_task_folder(task_dir: str) -> None:
     """
     Calling this function
     """
@@ -128,7 +128,7 @@ def get_task_name_from_object(task_object):
 
 
 # TODO: pass num_fewshot and other cmdline overrides in a better way
-def get_task_dict(task_name_list: List[Union[str, dict, Task]], **kwargs):
+def get_task_dict(task_name_list: List[Union[str, Dict, Task]], **kwargs):
 
     config = {**kwargs}
 
@@ -136,6 +136,9 @@ def get_task_dict(task_name_list: List[Union[str, dict, Task]], **kwargs):
     task_name_from_config_dict = {}
     task_name_from_object_dict = {}
 
+    if type(task_name_list) != list:
+        task_name_list = [task_name_list]
+
     for task_element in task_name_list:
         if isinstance(task_element, str):
 
@@ -143,12 +146,20 @@ def get_task_dict(task_name_list: List[Union[str, dict, Task]], **kwargs):
                 group_name = task_element
                 for task_name in GROUP_REGISTRY[task_element]:
                     if task_name not in task_name_from_registry_dict:
+                        task_obj = get_task_dict(task_name)
+                        if task_name in task_obj.keys():
+                            task_dict = {
+                                task_name: (group_name, task_obj[task_name]),
+                            }
+                        else:
+                            task_dict = {
+                                task_name: (group_name, None),
+                                **task_obj,
+                            }
+
                         task_name_from_registry_dict = {
                             **task_name_from_registry_dict,
-                            task_name: (
-                                group_name,
-                                get_task(task_name=task_name, config=config),
-                            ),
+                            **task_dict,
                         }
             else:
                 task_name = task_element

lm_eval/tasks/asdiv/default.yaml

+task: asdiv
+dataset_path: EleutherAI/asdiv
+output_type: loglikelihood
+validation_split: validation
+doc_to_text: "{{body}}\nQuestion:{{question}}\nAnswer:"
+doc_to_target: "{{answer.split(' (')[0]}}"
+should_decontaminate: true
+doc_to_decontamination_query: "{{body}} {{question}}"
+metric_list:
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true

lm_eval/tasks/coqa/README.md

+# CoQA
+
+### Paper
+
+Title: `CoQA: A Conversational Question Answering Challenge`
+
+Abstract: https://arxiv.org/pdf/1808.07042.pdf
+
+CoQA is a large-scale dataset for building Conversational Question Answering
+systems. The goal of the CoQA challenge is to measure the ability of machines to
+understand a text passage and answer a series of interconnected questions that
+appear in a conversation.
+
+Homepage: https://stanfordnlp.github.io/coqa/
+
+### Citation
+
+```
+BibTeX-formatted citation goes here
+```
+
+### Groups and Tasks
+
+#### Groups
+
+* Not part of a group yet
+
+#### Tasks
+
+* `coqa`
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+* [ ] Is the task an existing benchmark in the literature?
+  * [ ] Have you referenced the original paper that introduced the task?
+  * [ ] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+
+If other tasks on this dataset are already supported:
+* [ ] Is the "Main" variant of this task clearly denoted?
+* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?

lm_eval/tasks/coqa/default.yaml

+task: coqa
+dataset_path: EleutherAI/coqa
+output_type: greedy_until
+training_split: train
+validation_split: validation
+doc_to_text: !function utils.doc_to_text
+doc_to_target: !function utils.doc_to_target
+process_results: !function utils.process_results
+should_decontaminate: true
+doc_to_decontamination_query: "{{story}} {{question.input_text|join('\n')}}"
+generation_kwargs:
+  until:
+    - "\nQ:"
+metric_list:
+  - metric: em
+    aggregation: mean
+    higher_is_better: true
+  - metric: f1
+    aggregation: mean
+    higher_is_better: true

lm_eval/tasks/coqa/utils.py

+from itertools import zip_longest
+
+import transformers.data.metrics.squad_metrics as squad_metrics
+
+
+def doc_to_text(doc):
+    # Given a passage p, the conversation history {q1, a1, . . . qi−1, ai−1}
+    # and a question qi, the task is to predict the answer ai
+    doc_text = doc["story"] + "\n\n"
+    for (q, a) in zip_longest(
+        doc["questions"]["input_text"], doc["answers"]["input_text"][:-1]
+    ):  # omit target answer ai
+        question = f"Q: {q}\n\n"
+        answer = f"A: {a}\n\n" if a is not None else "A:"
+        doc_text += question + answer
+    return doc_text
+
+
+def doc_to_target(doc):
+
+    turn_id = len(doc["questions"]["input_text"])
+    # Returns unique answers and valid alternatives (Some questions in CoQA have multiple valid answers).
+    answers = []
+    answer_forturn = doc["answers"]["input_text"][turn_id - 1]
+    answers.append(answer_forturn)
+
+    additional_answers = doc.get("additional_answers")
+    if additional_answers:
+        for key in additional_answers:
+            additional_answer_for_turn = additional_answers[key]["input_text"][
+                turn_id - 1
+            ]
+            if additional_answer_for_turn.lower() not in map(str.lower, answers):
+                answers.append(additional_answer_for_turn)
+    return answers
+
+
+def em(gold_list, pred):
+    # tests for exact match and on the normalised answer (compute_exact)
+    em_sum = 0.0
+    if len(gold_list) > 1:
+        for i in range(len(gold_list)):
+            gold_answers = gold_list[0:i] + gold_list[i + 1 :]
+            # predictions compared against (n) golds and take maximum
+            em_sum += max(squad_metrics.compute_exact(a, pred) for a in gold_answers)
+    else:
+        em_sum += max(squad_metrics.compute_exact(a, pred) for a in gold_list)
+
+    return em_sum / max(1, len(gold_list))
+
+
+def compute_scores(gold_list, pred):
+    # tests for exact match and on the normalised answer (compute_exact)
+    # test for overlap (compute_f1)
+    f1_sum = 0.0
+    em_sum = 0.0
+    if len(gold_list) > 1:
+        for i in range(len(gold_list)):
+            gold_answers = gold_list[0:i] + gold_list[i + 1 :]
+            # predictions compared against (n) golds and take maximum
+            em_sum += max(squad_metrics.compute_exact(a, pred) for a in gold_answers)
+            f1_sum += max(squad_metrics.compute_f1(a, pred) for a in gold_answers)
+    else:
+        em_sum += max(squad_metrics.compute_exact(a, pred) for a in gold_list)
+        f1_sum += max(squad_metrics.compute_f1(a, pred) for a in gold_list)
+
+    return {
+        "em": em_sum / max(1, len(gold_list)),
+        "f1": f1_sum / max(1, len(gold_list)),
+    }
+
+
+def process_results(doc, results):
+
+    gold_list = doc_to_target(doc)
+    pred = results[0].strip().split("\n")[0]
+
+    scores = compute_scores(gold_list, pred)
+    return scores

lm_eval/tasks/drop/README.md

+# DROP
+
+### Paper
+
+Title: `DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs`
+
+Abstract: https://aclanthology.org/attachments/N19-1246.Supplementary.pdf
+
+DROP is a QA dataset which tests comprehensive understanding of paragraphs. In
+this crowdsourced, adversarially-created, 96k question-answering benchmark, a
+system must resolve multiple references in a question, map them onto a paragraph,
+and perform discrete operations over them (such as addition, counting, or sorting).
+
+Homepage: https://allenai.org/data/drop
+
+Acknowledgement: This implementation is based on the official evaluation for `DROP`:
+https://github.com/allenai/allennlp-reading-comprehension/blob/master/allennlp_rc/eval/drop_eval.py
+
+### Citation
+
+```
+@misc{dua2019drop,
+    title={DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs},
+    author={Dheeru Dua and Yizhong Wang and Pradeep Dasigi and Gabriel Stanovsky and Sameer Singh and Matt Gardner},
+    year={2019},
+    eprint={1903.00161},
+    archivePrefix={arXiv},
+    primaryClass={cs.CL}
+}
+```
+
+### Groups and Tasks
+
+#### Groups
+
+* Not part of a group yet.
+
+#### Tasks
+
+* `drop`
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+* [ ] Is the task an existing benchmark in the literature?
+  * [ ] Have you referenced the original paper that introduced the task?
+  * [ ] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+
+If other tasks on this dataset are already supported:
+* [ ] Is the "Main" variant of this task clearly denoted?
+* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?

lm_eval/tasks/drop/default.yaml

+task: drop
+dataset_path: EleutherAI/drop
+output_type: greedy_until
+training_split: train
+validation_split: validation
+process_docs: !function utils.process_docs
+doc_to_text: "{{passage}} {{question}}"
+doc_to_target: "{{ answer|join(',')}}"
+target_delimiter: ""
+process_results: !function utils.process_results
+should_decontaminate: true
+doc_to_decontamination_query: "{{passage}} {{question}}"
+generation_kwargs:
+  until:
+    - "."
+metric_list:
+  - metric: em
+    aggregation: mean
+    higher_is_better: true
+  - metric: f1
+    aggregation: mean
+    higher_is_better: true

lm_eval/tasks/drop/utils.py

+import re
+import string
+
+import numpy as np
+from scipy.optimize import linear_sum_assignment
+
+_ARTICLES = re.compile(r"\b(a|an|the)\b", re.UNICODE)
+
+
+def process_docs(dataset):
+    def _process(doc):
+        return {
+            "id": doc["query_id"],
+            "passage": doc["passage"],
+            "question": doc["question"],
+            "answers": get_answers(doc),
+        }
+
+    return dataset.map(_process)
+
+
+def get_answers(doc):
+    def _flatten_validated_answers(validated_answers):
+        """Flattens a dict of lists of validated answers.
+        {"number": ['1', '8'], ...}
+        -> [{"number": ['1'], ...}, {"number": ['8'], ...}]
+        """
+        valid_answers = []
+        for i in range(len(validated_answers["number"])):
+            valid_answers.append(
+                {
+                    "number": validated_answers["number"][i],
+                    "date": validated_answers["date"][i],
+                    "spans": validated_answers["spans"][i],
+                }
+            )
+        return valid_answers
+
+    answers = []
+    answers_set = set()
+    candidates = [doc["answer"]] + _flatten_validated_answers(doc["validated_answers"])
+    for candidate in candidates:
+        answer = parse_answer(candidate)
+        if answer in answers_set:
+            continue
+        answers_set.add(answer)
+        answers.append(answer)
+    return answers
+
+
+def parse_answer(answer):
+    # NOTE: Everything is returned as a tuple for uniformity and hashability.
+    if answer["number"] != "":
+        return (str(answer["number"]),)
+    if answer["spans"] != []:
+        return tuple(answer["spans"])
+    return (
+        " ".join(
+            [answer["date"]["day"], answer["date"]["month"], answer["date"]["year"]]
+        ).strip(),
+    )
+
+
+def process_results(doc, results):
+
+    preds, golds = results, doc["answers"]
+    max_em = 0
+    max_f1 = 0
+    for gold_answer in golds:
+        exact_match, f1_score = get_metrics(preds, gold_answer)
+        if gold_answer[0].strip():
+            max_em = max(max_em, exact_match)
+            max_f1 = max(max_f1, f1_score)
+    return {"em": max_em, "f1": max_f1}
+
+
+def get_metrics(predicted, gold):
+    """
+    Takes a predicted answer and a gold answer (that are both either a string or a list of
+    strings), and returns exact match and the DROP F1 metric for the prediction.  If you are
+    writing a script for evaluating objects in memory (say, the output of predictions during
+    validation, or while training), this is the function you want to call, after using
+    :func:`answer_json_to_strings` when reading the gold answer from the released data file.
+    """
+    predicted_bags = _answer_to_bags(predicted)
+    gold_bags = _answer_to_bags(gold)
+
+    if set(predicted_bags[0]) == set(gold_bags[0]) and len(predicted_bags[0]) == len(
+        gold_bags[0]
+    ):
+        exact_match = 1.0
+    else:
+        exact_match = 0.0
+
+    f1_per_bag = _align_bags(predicted_bags[1], gold_bags[1])
+    f1 = np.mean(f1_per_bag)
+    f1 = round(f1, 2)
+    return exact_match, f1
+
+
+def _answer_to_bags(answer):
+    if isinstance(answer, (list, tuple)):
+        raw_spans = answer
+    else:
+        raw_spans = [answer]
+    normalized_spans = []
+    token_bags = []
+    for raw_span in raw_spans:
+        normalized_span = _normalize(raw_span)
+        normalized_spans.append(normalized_span)
+        token_bags.append(set(normalized_span.split()))
+    return normalized_spans, token_bags
+
+
+def _align_bags(predicted, gold):
+    """
+    Takes gold and predicted answer sets and first finds the optimal 1-1 alignment
+    between them and gets maximum metric values over all the answers.
+    """
+    scores = np.zeros([len(gold), len(predicted)])
+    for gold_index, gold_item in enumerate(gold):
+        for pred_index, pred_item in enumerate(predicted):
+            if _match_numbers_if_present(gold_item, pred_item):
+                scores[gold_index, pred_index] = _compute_f1(pred_item, gold_item)
+    row_ind, col_ind = linear_sum_assignment(-scores)
+
+    max_scores = np.zeros([max(len(gold), len(predicted))])
+    for row, column in zip(row_ind, col_ind):
+        max_scores[row] = max(max_scores[row], scores[row, column])
+    return max_scores
+
+
+def _compute_f1(predicted_bag, gold_bag):
+    intersection = len(gold_bag.intersection(predicted_bag))
+    if not predicted_bag:
+        precision = 1.0
+    else:
+        precision = intersection / float(len(predicted_bag))
+    if not gold_bag:
+        recall = 1.0
+    else:
+        recall = intersection / float(len(gold_bag))
+    f1 = (
+        (2 * precision * recall) / (precision + recall)
+        if not (precision == 0.0 and recall == 0.0)
+        else 0.0
+    )
+    return f1
+
+
+def _match_numbers_if_present(gold_bag, predicted_bag):
+    gold_numbers = set()
+    predicted_numbers = set()
+    for word in gold_bag:
+        if _is_number(word):
+            gold_numbers.add(word)
+    for word in predicted_bag:
+        if _is_number(word):
+            predicted_numbers.add(word)
+    if (not gold_numbers) or gold_numbers.intersection(predicted_numbers):
+        return True
+    return False
+
+
+def _is_number(text):
+    try:
+        float(text)
+        return True
+    except ValueError:
+        return False
+
+
+def _remove_articles(text):
+    return _ARTICLES.sub(" ", text)
+
+
+def _white_space_fix(text):
+    return " ".join(text.split())
+
+
+def _remove_punc(text):
+    exclude = set(string.punctuation)
+    if not _is_number(text):
+        return "".join(ch for ch in text if ch not in exclude)
+    else:
+        return text
+
+
+def _fix_number(text):
+    return str(float(text)) if _is_number(text) else text
+
+
+def _tokenize(text):
+    return re.split(" |-", text)
+
+
+def _normalize(answer):
+    tokens = [
+        _white_space_fix(_remove_articles(_fix_number(_remove_punc(token.lower()))))
+        for token in _tokenize(answer)
+    ]
+    tokens = [token for token in tokens if token.strip()]
+    normalized = " ".join(tokens).strip()
+    return normalized

lm_eval/tasks/glue/mnli/utils.py

-def doc_to_text(doc):
+def doc_to_text(doc) -> str:
     return "{}\nQuestion: {} True, False or Neither?\nAnswer:".format(
         doc["premise"],
         doc["hypothesis"].strip()