metadata
license: mit
dataset_info:
features:
- name: repo
dtype: string
- name: pull_number
dtype: string
- name: instance_id
dtype: string
- name: issue_numbers
list: string
- name: base_commit
dtype: string
- name: patch
dtype: string
- name: test_patch
dtype: string
- name: problem_statement
dtype: string
- name: hints_text
dtype: string
- name: all_hints_text
dtype: string
- name: commit_urls
list: string
- name: created_at
dtype: string
- name: commit_url
dtype: string
- name: rebuild_cmds
list: string
- name: test_cmds
list: string
- name: print_cmds
list: string
- name: log_parser
dtype: string
- name: FAIL_TO_PASS
list: string
- name: PASS_TO_PASS
list: string
- name: docker_image
dtype: string
splits:
- name: c
num_bytes: 9819814
num_examples: 35
- name: cpp
num_bytes: 1763388
num_examples: 29
- name: go
num_bytes: 11323896
num_examples: 46
- name: js
num_bytes: 13750757
num_examples: 54
- name: rust
num_bytes: 28790130
num_examples: 50
- name: java
num_bytes: 46065858
num_examples: 65
- name: ts
num_bytes: 9164856
num_examples: 40
- name: cs
num_bytes: 9181522
num_examples: 41
download_size: 35463528
dataset_size: 129860221
configs:
- config_name: default
data_files:
- split: c
path: data/c-*
- split: cpp
path: data/cpp-*
- split: go
path: data/go-*
- split: js
path: data/js-*
- split: rust
path: data/rust-*
- split: java
path: data/java-*
- split: ts
path: data/ts-*
- split: cs
path: data/cs-*
This is the multi - language version of swebench-Live.
We crawled GitHub issues as SWE tasks for coding agents to solve.
We used RepoLaunch to setup docker images as the problem-solving sandboxes for each instance.
Each split is task set of one language.
Each split is filtered by prompting LLM -- discarded if LLM cannot infer test patch from problem statement or problem statement contains solution (in NL or code).