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.

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

SPLADE++ ED w/ ONNX on BEIR #2354

Merged
merged 7 commits into from
Feb 12, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
8 changes: 5 additions & 3 deletions src/main/python/run_regression.py
Original file line number Diff line number Diff line change
Expand Up @@ -204,8 +204,10 @@ def evaluate_and_verify(yaml_data, dry_run):
expected = round(model['results'][metric['metric']][i], metric['metric_precision'])
actual = round(float(eval_out), metric['metric_precision'])

using_hnsw = True if 'VectorQueryGenerator' in model['params'] or '-encoder' in model['params'] else False

# For HNSW, we only print to third digit
if 'VectorQueryGenerator' in model['params']:
if using_hnsw:
result_str = 'expected: {0:.3f} actual: {1:.3f} - metric: {2:<8} model: {3} topics: {4}'.format(
expected, actual, metric['metric'], model['name'], topic_set['id'])
else:
Expand All @@ -216,8 +218,8 @@ def evaluate_and_verify(yaml_data, dry_run):
# For HNSW, be more tolerant, but as long as the actual score is higher than the expected score,
# let the test pass.
if is_close(expected, actual) or \
('VectorQueryGenerator' in model['params'] and is_close(expected, actual, abs_tol=0.007)) or \
('VectorQueryGenerator' in model['params'] and actual > expected):
(using_hnsw and is_close(expected, actual, abs_tol=0.007)) or \
(using_hnsw and actual > expected):
logger.info(ok_str + result_str)
# For ONNX runs, increase tolerance a bit because we observe some minor differences across OSes.
elif '-encoder' in model['params'] and is_close(expected, actual, abs_tol=0.001):
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
---
corpus: beir-v1.0.0-arguana-splade-pp-ed
corpus_path: collections/beir-v1.0.0/splade-pp-ed/arguana

index_path: indexes/lucene-index.beir-v1.0.0-arguana-splade-pp-ed/
collection_class: JsonVectorCollection
generator_class: DefaultLuceneDocumentGenerator
index_threads: 16
index_options: -impact -pretokenized
index_stats:
documents: 8674
documents (non-empty): 8674
total terms: 71992355

metrics:
- metric: nDCG@10
command: target/appassembler/bin/trec_eval
params: -c -m ndcg_cut.10
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@100
command: target/appassembler/bin/trec_eval
params: -c -m recall.100
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@1000
command: target/appassembler/bin/trec_eval
params: -c -m recall.1000
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false

topic_reader: TsvString
topics:
- name: "BEIR (v1.0.0): ArguAna"
id: test
path: topics.beir-v1.0.0-arguana.test.tsv.gz
qrel: qrels.beir-v1.0.0-arguana.test.txt

models:
- name: splade-pp-ed
display: SPLADE++ (CoCondenser-EnsembleDistil)
params: -impact -pretokenized -removeQuery -hits 1000 -encoder SpladePlusPlusEnsembleDistil
results:
nDCG@10:
- 0.5203
R@100:
- 0.9744
R@1000:
- 0.9950
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
---
corpus: beir-v1.0.0-bioasq-splade-pp-ed
corpus_path: collections/beir-v1.0.0/splade-pp-ed/bioasq

index_path: indexes/lucene-index.beir-v1.0.0-bioasq-splade-pp-ed/
collection_class: JsonVectorCollection
generator_class: DefaultLuceneDocumentGenerator
index_threads: 16
index_options: -impact -pretokenized
index_stats:
documents: 14914603
documents (non-empty): 14914603
total terms: 127381306317

metrics:
- metric: nDCG@10
command: target/appassembler/bin/trec_eval
params: -c -m ndcg_cut.10
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@100
command: target/appassembler/bin/trec_eval
params: -c -m recall.100
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@1000
command: target/appassembler/bin/trec_eval
params: -c -m recall.1000
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false

topic_reader: TsvString
topics:
- name: "BEIR (v1.0.0): BioASQ"
id: test
path: topics.beir-v1.0.0-bioasq.test.tsv.gz
qrel: qrels.beir-v1.0.0-bioasq.test.txt

models:
- name: splade-pp-ed
display: SPLADE++ (CoCondenser-EnsembleDistil)
params: -impact -pretokenized -removeQuery -hits 1000 -encoder SpladePlusPlusEnsembleDistil
results:
nDCG@10:
- 0.4980
R@100:
- 0.7385
R@1000:
- 0.8757
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
---
corpus: beir-v1.0.0-climate-fever-splade-pp-ed
corpus_path: collections/beir-v1.0.0/splade-pp-ed/climate-fever

index_path: indexes/lucene-index.beir-v1.0.0-climate-fever-splade-pp-ed/
collection_class: JsonVectorCollection
generator_class: DefaultLuceneDocumentGenerator
index_threads: 16
index_options: -impact -pretokenized
index_stats:
documents: 5416593
documents (non-empty): 5416593
total terms: 28498465299

metrics:
- metric: nDCG@10
command: target/appassembler/bin/trec_eval
params: -c -m ndcg_cut.10
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@100
command: target/appassembler/bin/trec_eval
params: -c -m recall.100
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@1000
command: target/appassembler/bin/trec_eval
params: -c -m recall.1000
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false

topic_reader: TsvString
topics:
- name: "BEIR (v1.0.0): Climate-FEVER"
id: test
path: topics.beir-v1.0.0-climate-fever.test.tsv.gz
qrel: qrels.beir-v1.0.0-climate-fever.test.txt

models:
- name: splade-pp-ed
display: SPLADE++ (CoCondenser-EnsembleDistil)
params: -impact -pretokenized -removeQuery -hits 1000 -encoder SpladePlusPlusEnsembleDistil
results:
nDCG@10:
- 0.2297
R@100:
- 0.5211
R@1000:
- 0.7183
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
---
corpus: beir-v1.0.0-cqadupstack-android-splade-pp-ed
corpus_path: collections/beir-v1.0.0/splade-pp-ed/cqadupstack-android

index_path: indexes/lucene-index.beir-v1.0.0-cqadupstack-android-splade-pp-ed/
collection_class: JsonVectorCollection
generator_class: DefaultLuceneDocumentGenerator
index_threads: 16
index_options: -impact -pretokenized
index_stats:
documents: 22998
documents (non-empty): 22998
total terms: 108476959

metrics:
- metric: nDCG@10
command: target/appassembler/bin/trec_eval
params: -c -m ndcg_cut.10
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@100
command: target/appassembler/bin/trec_eval
params: -c -m recall.100
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@1000
command: target/appassembler/bin/trec_eval
params: -c -m recall.1000
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false

topic_reader: TsvString
topics:
- name: "BEIR (v1.0.0): CQADupStack-android"
id: test
path: topics.beir-v1.0.0-cqadupstack-android.test.tsv.gz
qrel: qrels.beir-v1.0.0-cqadupstack-android.test.txt

models:
- name: splade-pp-ed
display: SPLADE++ (CoCondenser-EnsembleDistil)
params: -impact -pretokenized -removeQuery -hits 1000 -encoder SpladePlusPlusEnsembleDistil
results:
nDCG@10:
- 0.3904
R@100:
- 0.7404
R@1000:
- 0.9064
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
---
corpus: beir-v1.0.0-cqadupstack-english-splade-pp-ed
corpus_path: collections/beir-v1.0.0/splade-pp-ed/cqadupstack-english

index_path: indexes/lucene-index.beir-v1.0.0-cqadupstack-english-splade-pp-ed/
collection_class: JsonVectorCollection
generator_class: DefaultLuceneDocumentGenerator
index_threads: 16
index_options: -impact -pretokenized
index_stats:
documents: 40221
documents (non-empty): 40221
total terms: 158861979

metrics:
- metric: nDCG@10
command: target/appassembler/bin/trec_eval
params: -c -m ndcg_cut.10
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@100
command: target/appassembler/bin/trec_eval
params: -c -m recall.100
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@1000
command: target/appassembler/bin/trec_eval
params: -c -m recall.1000
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false

topic_reader: TsvString
topics:
- name: "BEIR (v1.0.0): CQADupStack-english"
id: test
path: topics.beir-v1.0.0-cqadupstack-english.test.tsv.gz
qrel: qrels.beir-v1.0.0-cqadupstack-english.test.txt

models:
- name: splade-pp-ed
display: SPLADE++ (CoCondenser-EnsembleDistil)
params: -impact -pretokenized -removeQuery -hits 1000 -encoder SpladePlusPlusEnsembleDistil
results:
nDCG@10:
- 0.4079
R@100:
- 0.6946
R@1000:
- 0.8454
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
---
corpus: beir-v1.0.0-cqadupstack-gaming-splade-pp-ed
corpus_path: collections/beir-v1.0.0/splade-pp-ed/cqadupstack-gaming

index_path: indexes/lucene-index.beir-v1.0.0-cqadupstack-gaming-splade-pp-ed/
collection_class: JsonVectorCollection
generator_class: DefaultLuceneDocumentGenerator
index_threads: 16
index_options: -impact -pretokenized
index_stats:
documents: 45301
documents (non-empty): 45301
total terms: 197713644

metrics:
- metric: nDCG@10
command: target/appassembler/bin/trec_eval
params: -c -m ndcg_cut.10
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@100
command: target/appassembler/bin/trec_eval
params: -c -m recall.100
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false
- metric: R@1000
command: target/appassembler/bin/trec_eval
params: -c -m recall.1000
separator: "\t"
parse_index: 2
metric_precision: 4
can_combine: false

topic_reader: TsvString
topics:
- name: "BEIR (v1.0.0): CQADupStack-gaming"
id: test
path: topics.beir-v1.0.0-cqadupstack-gaming.test.tsv.gz
qrel: qrels.beir-v1.0.0-cqadupstack-gaming.test.txt

models:
- name: splade-pp-ed
display: SPLADE++ (CoCondenser-EnsembleDistil)
params: -impact -pretokenized -removeQuery -hits 1000 -encoder SpladePlusPlusEnsembleDistil
results:
nDCG@10:
- 0.4957
R@100:
- 0.8131
R@1000:
- 0.9221
Loading
Loading