Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add support for FinMTEB benchmark #1379

Open
wants to merge 7 commits into
base: v2.0.0
Choose a base branch
from
Open

Add support for FinMTEB benchmark #1379

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
585cdeb
FinMTEB: classification, clustering, pairclassification, reranking, r…
alt-glitch Nov 4, 2024
cd86c8c
fix __init__
alt-glitch Nov 4, 2024
cdebb41
fix references not being URLs
alt-glitch Nov 4, 2024
231bd95
fix: update import statements; add summarization tasks; update pair_c…
alt-glitch Nov 7, 2024
4ccdb06
revert FinQA naming
alt-glitch Nov 7, 2024
4caa688
fix AbsTaskSTS default human/generated columns
alt-glitch Nov 7, 2024
2b38e05
refactor: update dataset column names and add references for various …
alt-glitch Nov 8, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 9 additions & 7 deletions mteb/abstasks/AbsTaskClassification.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,8 @@ class AbsTaskClassification(AbsTask):
"""

abstask_prompt = "Classify user passages."
sentence_column: str = "text"


def __init__(
self,
Expand Down Expand Up @@ -138,7 +140,7 @@ def _evaluate_subset(
)
# Bootstrap `self.samples_per_label` samples per label for each split
X_sampled, y_sampled, idxs = self._undersample_data(
train_split["text"], # type: ignore
train_split[self.sentence_column], # type: ignore
train_split["label"], # type: ignore
self.samples_per_label,
idxs,
Expand All @@ -148,7 +150,7 @@ def _evaluate_subset(
evaluator = kNNClassificationEvaluator(
X_sampled,
y_sampled,
eval_split["text"], # type: ignore
eval_split[self.sentence_column], # type: ignore
eval_split["label"], # type: ignore
task_name=self.metadata.name,
encode_kwargs=encode_kwargs,
Expand All @@ -158,7 +160,7 @@ def _evaluate_subset(
evaluator = kNNClassificationEvaluatorPytorch(
X_sampled,
y_sampled,
eval_split["text"], # type: ignore
eval_split[self.sentence_column], # type: ignore
eval_split["label"], # type: ignore
task_name=self.metadata.name,
encode_kwargs=encode_kwargs,
Expand All @@ -168,7 +170,7 @@ def _evaluate_subset(
evaluator = logRegClassificationEvaluator(
X_sampled,
y_sampled,
eval_split["text"], # type: ignore
eval_split[self.sentence_column], # type: ignore
eval_split["label"], # type: ignore
task_name=self.metadata.name,
encode_kwargs=encode_kwargs,
Expand Down Expand Up @@ -243,16 +245,16 @@ def _calculate_metrics_from_split(
self, split: str, hf_subset: str | None = None, compute_overall: bool = False
) -> ClassificationDescriptiveStatistics:
if hf_subset:
text = self.dataset[hf_subset][split]["text"]
text = self.dataset[hf_subset][split][self.sentence_column]
label = self.dataset[hf_subset][split]["label"]
elif compute_overall:
text = []
label = []
for hf_subset in self.metadata.eval_langs:
text.extend(self.dataset[hf_subset][split]["text"])
text.extend(self.dataset[hf_subset][split][self.sentence_column])
label.extend(self.dataset[hf_subset][split]["label"])
else:
text = self.dataset[split]["text"]
text = self.dataset[split][self.sentence_column]
label = self.dataset[split]["label"]

total_text_len = sum([len(t) for t in text])
Expand Down
18 changes: 10 additions & 8 deletions mteb/abstasks/AbsTaskPairClassification.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,8 @@ class AbsTaskPairClassification(AbsTask):
"""

abstask_prompt = "Retrieve text that are semantically similar to the given text."
sentence_1_column: str = "sentence1"
sentence_2_column: str = "sentence2"

def __init__(self, **kwargs):
super().__init__(**kwargs)
Expand All @@ -63,8 +65,8 @@ def _evaluate_subset(
"sentence_transformers.evaluation.PairClassificationEvaluator"
).setLevel(logging.WARN)
evaluator = PairClassificationEvaluator(
data_split["sentence1"],
data_split["sentence2"],
data_split[self.sentence_1_column],
data_split[self.sentence_2_column],
data_split["labels"],
task_name=self.metadata.name,
**kwargs,
Expand Down Expand Up @@ -93,14 +95,14 @@ def _calculate_metrics_from_split(
dataset = self.dataset[split]

sentence1 = (
dataset["sentence1"][0]
if len(dataset["sentence1"]) == 1
else dataset["sentence1"]
dataset[self.sentence_1_column][0]
if len(dataset[self.sentence_1_column]) == 1
else dataset[self.sentence_1_column]
)
sentence2 = (
dataset["sentence2"][0]
if len(dataset["sentence2"]) == 1
else dataset["sentence2"]
dataset[self.sentence_2_column][0]
if len(dataset[self.sentence_2_column]) == 1
else dataset[self.sentence_2_column]
)
labels = (
dataset["labels"][0] if len(dataset["labels"]) == 1 else dataset["labels"]
Expand Down
22 changes: 14 additions & 8 deletions mteb/abstasks/AbsTaskSTS.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,8 @@ class AbsTaskSTS(AbsTask):
"""

abstask_prompt = "Retrieve semantically similar text."
sentence_1_column: str = "sentence1"
sentence_2_column: str = "sentence2"

def __init__(self, **kwargs):
super().__init__(**kwargs)
Expand All @@ -56,8 +58,8 @@ def normalize(x):

normalized_scores = list(map(normalize, data_split["score"]))
evaluator = STSEvaluator(
data_split["sentence1"],
data_split["sentence2"],
data_split[self.sentence_1_column],
data_split[self.sentence_2_column],
normalized_scores,
task_name=self.metadata.name,
**kwargs,
Expand All @@ -74,20 +76,24 @@ def _calculate_metrics_from_split(
self, split: str, hf_subset: str | None = None, compute_overall: bool = False
) -> STSDescriptiveStatistics:
if hf_subset:
sentence1 = self.dataset[hf_subset][split]["sentence1"]
sentence2 = self.dataset[hf_subset][split]["sentence2"]
sentence1 = self.dataset[hf_subset][split][self.sentence_1_column]
sentence2 = self.dataset[hf_subset][split][self.sentence_2_column]
score = self.dataset[hf_subset][split]["score"]
elif compute_overall:
sentence1 = []
sentence2 = []
score = []
for hf_subset in self.metadata.eval_langs:
sentence1.extend(self.dataset[hf_subset][split]["sentence1"])
sentence2.extend(self.dataset[hf_subset][split]["sentence2"])
sentence1.extend(
self.dataset[hf_subset][split][self.sentence_1_column]
)
sentence2.extend(
self.dataset[hf_subset][split][self.sentence_2_column]
)
score.extend(self.dataset[hf_subset][split]["score"])
else:
sentence1 = self.dataset[split]["sentence1"]
sentence2 = self.dataset[split]["sentence2"]
sentence1 = self.dataset[split][self.sentence_1_column]
sentence2 = self.dataset[split][self.sentence_2_column]
score = self.dataset[split]["score"]

total_sentence1_len = sum([len(s) for s in sentence1])
Expand Down
23 changes: 15 additions & 8 deletions mteb/abstasks/AbsTaskSummarization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,9 @@ class AbsTaskSummarization(AbsTask):
"Given a news summary, retrieve other semantically similar summaries."
)

reference_summaries_column: str = "human_summaries"
generated_summaries_column: str = "machine_summaries"

def __init__(self, **kwargs):
super().__init__(**kwargs)

Expand All @@ -66,8 +69,8 @@ def _evaluate_subset(
for x in data_split["relevance"]
]
evaluator = self.evalutor(
machine_summaries=data_split["machine_summaries"],
human_summaries=data_split["human_summaries"],
machine_summaries=data_split[self.generated_summaries_column],
human_summaries=data_split[self.reference_summaries_column],
texts=data_split["text"],
gold_scores=normalized_scores,
task_name=self.metadata.name,
Expand All @@ -85,8 +88,12 @@ def _calculate_metrics_from_split(
) -> SummarizationDescriptiveStatistics:
if hf_subset:
text = self.dataset[hf_subset][split]["text"]
human_summaries = self.dataset[hf_subset][split]["human_summaries"]
machine_summaries = self.dataset[hf_subset][split]["machine_summaries"]
human_summaries = self.dataset[hf_subset][split][
self.reference_summaries_column
]
machine_summaries = self.dataset[hf_subset][split][
self.generated_summaries_column
]
relevance = self.dataset[hf_subset][split]["relevance"]
elif compute_overall:
text = []
Expand All @@ -97,16 +104,16 @@ def _calculate_metrics_from_split(
for hf_subset in self.metadata.eval_langs:
text.extend(self.dataset[hf_subset][split]["text"])
human_summaries.extend(
self.dataset[hf_subset][split]["human_summaries"]
self.dataset[hf_subset][split][self.reference_summaries_column]
)
machine_summaries.extend(
self.dataset[hf_subset][split]["machine_summaries"]
self.dataset[hf_subset][split][self.generated_summaries_column]
)
relevance.extend(self.dataset[hf_subset][split]["relevance"])
else:
text = self.dataset[split]["text"]
human_summaries = self.dataset[split]["human_summaries"]
machine_summaries = self.dataset[split]["machine_summaries"]
human_summaries = self.dataset[split][self.reference_summaries_column]
machine_summaries = self.dataset[split][self.generated_summaries_column]
relevance = self.dataset[split]["relevance"]

total_text_len = sum(len(x) for x in text)
Expand Down
Loading
Loading