Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code:   DatasetGenerationError
Exception:    IndexError
Message:      list index out of range
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1859, in _prepare_split_single
                  original_shard_lengths[original_shard_id] += len(table)
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
              IndexError: list index out of range
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1880, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

text
string
33d32
< import com.jcabi.aspects.Tv;
58a58,60
> private static final int TEN = 10;
> private static final int EIGHT = 8;
>
63,65c65,67
< final String name = RandomStringUtils.randomAlphanumeric(Tv.TEN);
< final String attr = RandomStringUtils.randomAlphabetic(Tv.EIGHT);
< final String value = RandomStringUtils.randomAlphanumeric(Tv.TEN);
---
> final String name = RandomStringUtils.randomAlphanumeric(TEN);
> final String attr = RandomStringUtils.randomAlphabetic(EIGHT);
> final String value = RandomStringUtils.randomAlphanumeric(TEN);
111c113
< for (int idx = 0; idx < Tv.TEN; ++idx) {
---
> for (int idx = 0; idx < TEN; ++idx) {
120c122
< Matchers.iterableWithSize(Tv.TEN)
---
> Matchers.iterableWithSize(TEN)
33:cannot find symbol symbol: class Tv location: package com.jcabi.aspects
/jcabi-simpledb/src/test/java/com/jcabi/simpledb/RegionITCase.java
Act as an Automatic Program Repair (APR) tool, reply only with code, without explanation.
You are specialized in breaking dependency updates, in which the failure is caused by an external dependency.
To solve the failure you can only work on the client code.
```java
/*
* Copyright (c) 2012-2022, jcabi.com
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met: 1) Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer. 2) Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials provided
* with the distribution. 3) Neither the name of the jcabi.com nor
* the names of its contributors may be used to endorse or promote
* products derived from this software without specific prior written
* permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT
* NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
* FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
* THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
* INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
* OF THE POSSIBILITY OF SUCH DAMAGE.
*/
package com.jcabi.simpledb;
import com.amazonaws.services.simpledb.model.SelectRequest;
import com.jcabi.aspects.Tv;
import org.apache.commons.lang3.RandomStringUtils;
import org.hamcrest.MatcherAssert;
import org.hamcrest.Matchers;
import org.junit.jupiter.api.Assumptions;
import org.junit.jupiter.api.Test;
/**
* Integration case for {@link Region}.
*
* @since 0.1
*/
final class RegionITCase {
/**
* SimpleDB key.
*/
private static final String KEY =
System.getProperty("failsafe.sdb.key");
/**
* SimpleDB secret key.
*/
private static final String SECRET =
System.getProperty("failsafe.sdb.secret");
@Test
void putsAndRemovesIndividualItems() {
final Domain domain = this.domain();
try {
final String name = RandomStringUtils.randomAlphanumeric(Tv.TEN);
final String attr = RandomStringUtils.randomAlphabetic(Tv.EIGHT);
final String value = RandomStringUtils.randomAlphanumeric(Tv.TEN);
for (int idx = 0; idx < 2; ++idx) {
domain.item(name).put(attr, value);
MatcherAssert.assertThat(
domain.item(name), Matchers.hasKey(attr)
);
domain.item(name).remove(attr);
MatcherAssert.assertThat(
End of preview.

Bacardi Breaking-Update Repair

Results from evaluating five self-hosted, open-weight LLMs on the Bacardi benchmark: automatically repairing Java projects broken by upstream dependency updates. Each model is run against the same 103-case breaking-dependency-update benchmark, across all 8 Bacardi prompt pipelines, served locally via vLLM on the Berzelius (NSC) HPC cluster.

  • Benchmark: 103 real-world Java "breaking update" commits (from the chains-project/breaking-updates corpus), one repair attempt per case (MAX_ATTEMPT=1).
  • Pipelines: 8 Bacardi prompt strategies — baseline, baseline-cot, baseline-cot-buggy, baseline-buggy-line, baseline-api-diff, baseline-api-diff-buggy, baseline-cot-api-diff, baseline-cot-api-diff-buggy — combining chain-of-thought prompting, buggy-line pointers, and API-diff context in different combinations.
  • Decoding: temperature 0.0 for four of the five models. Qwen3.5-4B could not be run at temperature 0.0 — see the note below — and was instead run 5 times at temperature 0.7 per pipeline to characterize run-to-run variance.

Results by pipeline (temperature 0.0)

Cells show successful repairs / cases attempted and the success rate. Qwen3.5-4B is not in this table — see its own section below, since it could not be run at temperature 0.0 and is not directly comparable to the other four.

Pipeline Ministral 3 14B Ministral 3 3B Gemma 4 E4B Gemma 4 12B
baseline 9/103 (8.7%) 2/103 (1.9%) 8/103 (7.8%) 17/103 (16.5%)
baseline-cot 9/103 (8.7%) 2/103 (1.9%) 6/103 (5.8%) 15/103 (14.6%)
baseline-cot-buggy 14/103 (13.6%) 3/103 (2.9%) 5/103 (4.9%) 17/103 (16.5%)
baseline-buggy-line 11/102 (10.8%) 3/103 (2.9%) 7/103 (6.8%) 18/103 (17.5%)
baseline-api-diff 14/102 (13.7%) 8/103 (7.8%) 15/103 (14.6%) 21/103 (20.4%)
baseline-api-diff-buggy 16/103 (15.5%) 7/103 (6.8%) 17/103 (16.5%) 22/103 (21.4%)
baseline-cot-api-diff 15/103 (14.6%) 9/103 (8.7%) 15/103 (14.6%) 17/103 (16.5%)
baseline-cot-api-diff-buggy 13/103 (12.6%) 4/103 (3.9%) 13/102 (12.7%) 19/103 (18.4%)
Total 101/822 (12.3%) 38/824 (4.6%) 86/822 (10.5%) 146/824 (17.7%)

Model cards

Model Params License Context Decoding
Ministral 3 14B 14B Apache 2.0 128K temp 0.0
Ministral 3 3B 3B Apache 2.0 256K temp 0.0
Gemma 4 E4B ~4B effective Apache 2.0 128K temp 0.0
Gemma 4 12B 12B Apache 2.0 256K temp 0.0

Qwen3.5-4B (separate: temperature 0.7, 5 rounds)

Both Qwen3.5-4B and Qwen3.5-9B loop inside their own reasoning trace at temperature 0.0 and never emit a usable fix on a meaningful fraction of cases (e.g. repeating "Wait, I need to check... Wait, I think... Wait, I found..." until the generation budget runs out). This reproduces on random cases, not just cherry-picked ones, and affects both model sizes, so it looks like a property of the Qwen3.5 reasoning template rather than a size-specific issue. temperature 0.7 avoids it, so Qwen3.5-4B was run 5 independent times per pipeline at temperature 0.7 instead of once at temperature 0.0, both to get a usable result at all and to characterize how much run-to-run variance that introduces. It is kept out of the main table above because it isn't decoded the same way as the other four models.

Per-round breakdown (successful repairs / cases attempted, out of 103):

Pipeline Round 1 Round 2 Round 3 Round 4 Round 5
baseline 3/103 8/103 6/103 8/103 7/103
baseline-cot 11/103 10/103 10/103 13/103 9/103
baseline-cot-buggy 9/103 9/103 10/103 13/103 11/103
baseline-buggy-line 8/103 9/103 8/103 7/103 9/103
baseline-api-diff 14/103 13/103 13/103 10/103 15/103
baseline-api-diff-buggy 11/103 13/103 10/103 14/103 11/103
baseline-cot-api-diff 12/103 12/103 13/103 11/103 13/103
baseline-cot-api-diff-buggy 13/103 12/103 12/103 12/103 15/103
Total 81/824 86/824 82/824 88/824 90/824

Aggregate across all 5 rounds:

Pipeline Successes / Attempts Rate
baseline 32/515 6.2%
baseline-cot 53/515 10.3%
baseline-cot-buggy 52/515 10.1%
baseline-buggy-line 41/515 8.0%
baseline-api-diff 65/515 12.6%
baseline-api-diff-buggy 59/515 11.5%
baseline-cot-api-diff 61/515 11.8%
baseline-cot-api-diff-buggy 64/515 12.5%
Total 427/4120 (10.4%)

The aggregate rate (sum of successes / sum of attempts) is the best point estimate for comparing against the four temperature 0.0 models, but on its own it hides exactly the thing this experiment set out to measure. Reported alone, it implies a level of precision the underlying single-round numbers don't have.

Run-to-run variability (mean and sample standard deviation of the 5 per-round success counts, out of 103 each):

Pipeline Mean ± SD Range (min–max) Coefficient of variation
baseline 6.4 ± 2.1 3–8 32.4%
baseline-cot 10.6 ± 1.5 9–13 14.3%
baseline-cot-buggy 10.4 ± 1.7 9–13 16.1%
baseline-buggy-line 8.2 ± 0.8 7–9 10.2%
baseline-api-diff 13.0 ± 1.9 10–15 14.4%
baseline-api-diff-buggy 11.8 ± 1.6 10–14 13.9%
baseline-cot-api-diff 12.2 ± 0.8 11–13 6.9%
baseline-cot-api-diff-buggy 12.8 ± 1.3 12–15 10.2%

baseline — the plain prompt with no extra context — is both the worst-performing pipeline and by far the least stable one (CV 32.4%, roughly 2-3x every other pipeline): a single temperature 0.7 run of baseline could have landed anywhere from 3 to 8 successes out of 103 purely from decoding randomness. Pipelines that give the model more context (baseline-api-diff, baseline-cot-api-diff) are not just more accurate on average, they're also markedly more consistent between rounds — extra context appears to reduce the model's sensitivity to sampling noise, not just shift its average outcome. This is why the per-round table and the mean±SD table above should be read together with the aggregate rate, not in place of it, whenever this dataset is used for comparison against temperature 0.0 results from other models.

Fields (per-attempt records)

Each repair attempt is recorded with: model, pipeline, round (Qwen3.5-4B only, 1-5), case_hash (the breaking-update commit), success, failure_category (BUILD_SUCCESS, COMPILATION_FAILURE, DEPENDENCY_RESOLUTION_FAILURE, ERROR_MODEL_RESPONSE, TEST_FAILURE, ...), prefix_errors/postfix_errors (compile errors before/after the model's fix), fixed_errors/unfixed_errors/new_errors, plus the full per-case artifacts: the prompt sent to the model, its raw and extracted response, the generated diff, and pre/post compile error logs.

Infrastructure

All five models were served locally via vLLM on A100 GPUs (Berzelius / NSC, Sweden) — no third-party API calls, no metered cost beyond the cluster's own compute-hour allocation.

Downloads last month
403