SWE 任务和轨迹进度看板

最后更新时间:2026-06-18 19:49:27 BJT | 下次刷新:2026-06-18 20:49:28 BJT | 刷新间隔:3600 秒

收集 PR 总数
557,067
1h 0 / 24h -386
有效 SWE 总数
49,258
1h +41 / 24h +464
整体处理成功率
10.6%
Valid SWE / 已处理 466,474
difficulty_score 均值
5.89
median 5.8,count 49,201

语言进度

语言收集 PR过去 1h过去 24h有效 SWE过去 1h过去 24h已处理处理成功率
Cc29,4690-819,663+1+1729,469
32.8%
C++cpp45,6200-864,0320+1419,000
21.2%
Gogo126,439007,884+11+10686,523
9.1%
Javajava84,8680-624,004+1+5875,499
5.3%
JavaScriptjs36,4090-1577,063+2+5436,599
19.3%
Pythonpy98,883004,890+14+7994,543
5.2%
Rustrust68,650005,414+6+4568,611
7.9%
TypeScriptts66,729006,308+6+9156,230
11.2%

运行参数

语言评估模型 (OPENAI)填充模型 (ANTHROPIC)并发数min_source_filesmax_source_files
Cgpt-5.4claude-sonnet-4-612215
C++glm-5claude-sonnet-4-68215
GoQwen3.6-35B-A3BQwen3.6-35B-A3B12210
Javaclaude-haiku-4-5-20251001claude-sonnet-4-68210
JavaScriptQwen3.6-35B-A3BQwen3.6-35B-A3B12210
Pythonglm-5claude-sonnet-4-612315
RustQwen3.6-35B-A3BQwen3.6-35B-A3B8210
TypeScriptQwen3.6-35B-A3BQwen3.6-35B-A3B12210

失败原因统计

语言已处理有效 SWE失败trivial_prvalidationinfra_errortimeoutworkflow_error其他
C29,4699,66319,80614,6273545,12117301
C++19,0004,03214,9682,0477314,398156285266
Go86,5237,88478,63921,3407,10847,4241,508731511
Java75,4994,00471,49516,9376,58544,4561,2931,0281,602
JavaScript36,5997,06329,53615,08183314,4825531480
Python94,5434,89089,65325,8905,77058,303925357120
Rust68,6115,41463,19719,0544,54736,9671,2348091,234
TypeScript56,2306,30849,92213,4193,07931,3331,69275511

trivial_pr:PR 被 LLM 评估为过于简单(如仅修改配置、文档、依赖版本等),不适合作为 SWE 任务。

validation:任务生成后验证失败(NOP agent 未返回 reward=0 或 ORACLE agent 未返回 reward=1)。

infra_error:基础设施错误(Docker 构建失败、网络超时、磁盘空间不足等)。

timeout:处理超时(单个 PR 总超时或 Claude Code session 超时)。

workflow_error:工作流程错误(PR 元数据获取失败、worktree 创建失败、patch 生成失败等)。

fix.patch 复杂度

语言Valid SWE CountAvg fix.patch linesAvg fix.patch hunksAvg fix.patch files
C9,663334.7117.935.84
C++4,032287.0313.735.10
Go7,884215.9712.804.39
Java4,004163.6410.514.23
JavaScript7,06377.256.292.79
Python4,890152.2610.973.81
Rust5,414226.7813.184.10
TypeScript6,308158.759.594.14

统计方法说明

难度打分 difficulty_score

读取每个有效任务目录的 solution/fix.patchtests/instruction.md,由 src/swegen/scoring.py 使用零 API 静态评分。

当前公式采用 log-scale 连续评分,避免中等规模 patch 过早变成 hard。权重为:patch_scope 38%logic_complexity 32%context_breadth 15%test_complexity 10%instruction_complexity 5%

label 阈值:easy <= 4.0medium <= 7.0hard > 7.0

Tags 生成与展示

tags 不是看板现场计算的,而是在 swegen 构建任务时由 LLM 根据 PR 信息生成,并写入 task.toml[metadata].tags

prompt 要求 tags 按三段式生成:编程语言、项目层级/领域、框架/库名或具体主题。看板只读取已有 task.toml 并统计每个语言的 tag 出现次数和占比。

fix.patch 统计

patch 统计来自每个有效任务的 solution/fix.patch,并按语言扩展名过滤代码文件,口径与 upload_march_swe_to_hf.py 的 code-only 统计保持一致。

Avg fix.patch lines 统计代码文件 diff 中新增/删除行数;Avg fix.patch hunks 统计 @@ hunk 数;Avg fix.patch files 统计涉及的代码文件数。

difficulty_label 分布

语言easy / medium / hardeasymediumhard
C
863 / 6388 / 2402
8636,3882,402
C++
431 / 2501 / 1092
4312,5011,092
Go
614 / 5695 / 1569
6145,6951,569
Java
441 / 2594 / 964
4412,594964
JavaScript
1063 / 5235 / 763
1,0635,235763
Python
264 / 3145 / 1458
2643,1451,458
Rust
379 / 3201 / 1832
3793,2011,832
TypeScript
561 / 4672 / 1074
5614,6721,074

difficulty_score 概览

语言countminp25medianmeanp75max
C9,6532.44.96.05.977.09.2
C++4,0242.54.96.05.997.29.1
Go7,8782.64.95.85.866.89.1
Java3,9992.84.85.95.907.09.2
JavaScript7,0612.64.45.25.366.29.2
Python4,8672.65.26.26.217.39.1
Rust5,4122.75.26.36.267.49.0
TypeScript6,3072.74.75.65.726.69.1

全局 Top Tags

library22,930 (46.6%)
backend15,138 (30.8%)
cli6,669 (13.5%)
frontend3,011 (6.1%)
testing2,145 (4.4%)
http1,523 (3.1%)
react1,377 (2.8%)
framework1,204 (2.4%)
kubernetes691 (1.4%)
async660 (1.3%)
embedded640 (1.3%)
networking607 (1.2%)
cpp588 (1.2%)
typescript373 (0.8%)
parsing353 (0.7%)
eslint339 (0.7%)
graphql330 (0.7%)
postgresql298 (0.6%)
compiler283 (0.6%)
security283 (0.6%)
git280 (0.6%)
database276 (0.6%)
aws259 (0.5%)
json245 (0.5%)
redis233 (0.5%)
ruby220 (0.4%)
api220 (0.4%)
fastapi220 (0.4%)
angular207 (0.4%)
cryptography204 (0.4%)

每语言 Tags 分布

C c

library5,064 (52.4%)
backend2,498 (25.9%)
cli1,114 (11.5%)
embedded614 (6.4%)
cpp585 (6.1%)
testing403 (4.2%)
networking399 (4.1%)
framework285 (2.9%)
postgresql212 (2.2%)
ruby209 (2.2%)
http201 (2.1%)
firmware185 (1.9%)
kernel178 (1.8%)
quic169 (1.7%)
rust140 (1.4%)
bluetooth129 (1.3%)
tls128 (1.3%)
python114 (1.2%)
scheduler111 (1.1%)
cryptography107 (1.1%)

C++ cpp

library2,779 (69.0%)
backend728 (18.1%)
testing488 (12.1%)
cli356 (8.8%)
framework185 (4.6%)
http129 (3.2%)
boost114 (2.8%)
async91 (2.3%)
parsing76 (1.9%)
qt63 (1.6%)
compiler54 (1.3%)
serialization54 (1.3%)
geometry51 (1.3%)
networking49 (1.2%)
formatting45 (1.1%)
ros244 (1.1%)
frontend40 (1.0%)
json40 (1.0%)
logging37 (0.9%)
templates36 (0.9%)

Go go

backend4,146 (52.6%)
library1,953 (24.8%)
cli1,920 (24.4%)
kubernetes632 (8.0%)
http513 (6.5%)
testing219 (2.8%)
docker129 (1.6%)
aws117 (1.5%)
grpc116 (1.5%)
terraform112 (1.4%)
prometheus100 (1.3%)
networking93 (1.2%)
database87 (1.1%)
framework86 (1.1%)
api81 (1.0%)
git76 (1.0%)
security69 (0.9%)
aws-sdk65 (0.8%)
blockchain55 (0.7%)
configuration53 (0.7%)

Java java

backend1,943 (48.6%)
library1,809 (45.2%)
testing209 (5.2%)
spring157 (3.9%)
framework150 (3.7%)
aem147 (3.7%)
http136 (3.4%)
android118 (2.9%)
json67 (1.7%)
cli63 (1.6%)
maven58 (1.4%)
sling57 (1.4%)
kafka53 (1.3%)
flink48 (1.2%)
concurrency47 (1.2%)
grpc47 (1.2%)
frontend46 (1.1%)
mybatis37 (0.9%)
nacos34 (0.8%)
security34 (0.8%)

JavaScript js

library3,804 (53.9%)
backend1,222 (17.3%)
frontend998 (14.1%)
cli843 (11.9%)
react343 (4.9%)
typescript311 (4.4%)
testing291 (4.1%)
eslint260 (3.7%)
framework196 (2.8%)
http179 (2.5%)
fastify146 (2.1%)
webpack144 (2.0%)
mongoose103 (1.5%)
nodejs101 (1.4%)
svelte99 (1.4%)
express91 (1.3%)
stylelint64 (0.9%)
vue64 (0.9%)
lighthouse63 (0.9%)
async61 (0.9%)

Python py

library2,053 (42.2%)
backend2,036 (41.8%)
cli674 (13.8%)
fastapi198 (4.1%)
django137 (2.8%)
testing109 (2.2%)
pytorch101 (2.1%)
ansible100 (2.1%)
framework98 (2.0%)
async95 (2.0%)
http91 (1.9%)
aws72 (1.5%)
aiohttp65 (1.3%)
pydantic57 (1.2%)
openai55 (1.1%)
click51 (1.0%)
litellm49 (1.0%)
beets48 (1.0%)
flask48 (1.0%)
frontend43 (0.9%)

Rust rust

library3,031 (56.0%)
backend1,168 (21.6%)
cli1,113 (20.6%)
testing305 (5.6%)
async218 (4.0%)
http204 (3.8%)
compiler119 (2.2%)
git117 (2.2%)
macros91 (1.7%)
parsing90 (1.7%)
graphql79 (1.5%)
blockchain74 (1.4%)
substrate74 (1.4%)
serde71 (1.3%)
framework62 (1.1%)
sql62 (1.1%)
datafusion57 (1.1%)
cryptography56 (1.0%)
lsp55 (1.0%)
database53 (1.0%)

TypeScript ts

library2,437 (38.6%)
frontend1,743 (27.6%)
backend1,397 (22.1%)
react1,026 (16.3%)
cli586 (9.3%)
angular201 (3.2%)
graphql161 (2.6%)
framework142 (2.3%)
javascript138 (2.2%)
electron130 (2.1%)
fullstack124 (2.0%)
testing121 (1.9%)
vue96 (1.5%)
nextjs82 (1.3%)
github-actions81 (1.3%)
eslint79 (1.3%)
express79 (1.3%)
vscode71 (1.1%)
http70 (1.1%)
mcp69 (1.1%)
轨迹文件数
27
轨迹总条数
79,279
平均消息轮数
67.0
平均 composite_score
0.7449
基于 29,596 条有分数的轨迹

轨迹文件总览

数据集脚手架模型Owner轨迹数文件大小平均轮数平均 Token平均 Tool Calls平均 Score
swegenClaude Codeglm5chaofan1,096446 MB72.739,64837.90.7218
swegenOpenCodeglm5chaofan860197 MB64.830,57734.00.7337
swegenOpenHands SDKglm5chaofan1,140280 MB119.649,81159.80.7533
swegenTerminus-2glm5chaofan1,331142 MB53.226,96841.20.7048
swerebench_oraclesolvedClaude Codeglm5chaofan4,053601 MB44.122,11722.7
swerebench_oraclesolvedOpenCodeglm5chaofan3,431405 MB41.319,70421.5
swerebench_oraclesolvedOpenHands-AIglm5chaofan2,808638 MB135.136,43467.2
swerebench_oraclesolvedTerminus-2glm5chaofan2,920291 MB55.725,47336.9
swegen_selfmade_260301_260414OpenHands SDKglm5jierun4,555993 MB98.541,83949.1
swegen_selfmade_260301_260414Terminus-2glm5jierun4,805659 MB75.834,44854.4
swegen_selfmade_260415_260505OpenHands SDKglm5jierun3,954901 MB108.843,86754.2
swerebench_oraclesolvedClaude Codeglm5jierun5,013779 MB44.724,12323.3
swerebench_oraclesolvedOpenCodeglm5jierun3,521464 MB51.822,44727.0
swerebench_oraclesolvedOpenHands SDKglm5jierun2,733507 MB95.835,56747.5
swerebench_oraclesolvedTerminus-2glm5jierun2,998296 MB55.525,25737.0
swerebench_othersClaude Codeglm5jierun2,885457 MB43.825,19423.10.7178
swerebench_othersOpenCodeglm5jierun2,637360 MB49.123,63725.80.7210
swerebench_othersOpenHands SDKglm5jierun1,843340 MB90.235,60244.80.7868
swerebench_othersTerminus-2glm5jierun2,270225 MB50.624,99035.10.7844
swerebenchv2_python_oraclesolvedClaude Codeglm5jierun2,728445 MB47.526,06425.1
swerebenchv2_python_oraclesolvedOpenCodeglm5jierun1,654240 MB60.723,89831.3
swerebenchv2_python_oraclesolvedOpenHands SDKglm5jierun1,521313 MB103.040,23251.0
swerebenchv2_python_oraclesolvedTerminus-2glm5jierun1,633174 MB55.727,01337.6
v2nopy_fullClaude Codeglm5jierun5,128818 MB42.726,94122.80.7281
v2nopy_fullOpenCodeglm5jierun3,465455 MB47.023,17224.80.7353
v2nopy_fullOpenHands SDKglm5jierun3,266620 MB92.738,25246.00.7580
v2nopy_fullTerminus-2glm5jierun3,675411 MB64.128,36545.60.7799

质量评分统计

数据集脚手架compositeefficiencystyletool_masterycompletionprecision
swegenClaude Code0.72180.91250.34140.86780.62130.7722
swegenOpenCode0.73370.93220.32720.88860.62880.7923
swegenOpenHands SDK0.75330.89750.35000.85370.77370.7632
swegenTerminus-20.70480.92410.42280.89830.39600.8866
swerebench_othersClaude Code0.71780.88640.31970.79420.64050.8924
swerebench_othersOpenCode0.72100.89520.31710.80930.65090.8621
swerebench_othersOpenHands SDK0.78680.88860.36930.76980.93330.8525
swerebench_othersTerminus-20.78440.85390.42920.81720.82030.9323
v2nopy_fullClaude Code0.72810.91860.32130.88860.55650.8996
v2nopy_fullOpenCode0.73530.92780.32540.90080.57660.8772
v2nopy_fullOpenHands SDK0.75800.90020.35550.84400.75230.8370
v2nopy_fullTerminus-20.77990.92220.43100.90160.69300.8808

按数据集对比

数据集总轨迹数平均轮数平均 Token平均 Tool Calls平均 Score
swegen4,42777.436,69143.80.7271
swegen_selfmade_260301_2604149,36086.838,04451.8
swegen_selfmade_260415_2605053,954108.843,86754.2
swerebench_oraclesolved27,47761.825,72433.3
swerebench_others9,63555.726,71130.80.7475
swerebenchv2_python_oraclesolved7,53663.428,65434.4
v2nopy_full15,53459.228,81533.50.7483

源数据路径目录

数据集脚手架Owner文件路径大小条数
swegenClaude Codechaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swegen_cc_2135.jsonl446 MB2,135
swegenOpenCodechaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swegen_oc_1177.jsonl197 MB1,177
swegenOpenHands SDKchaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swegen_ohsdk_1140.jsonl280 MB1,140
swegenTerminus-2chaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swegen_t2_1331.jsonl142 MB1,331
swegen_selfmade_260301_260414OpenHands SDKjierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swegen_selfmade_260301_260414_oh_sdk_4555.jsonl993 MB4,555
swegen_selfmade_260301_260414Terminus-2jierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swegen_selfmade_260301_260414_t2_4805.jsonl659 MB4,805
swegen_selfmade_260415_260505OpenHands SDKjierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swegen_selfmade_260415_260505_oh_sdk_3954.jsonl901 MB3,954
swerebench_oraclesolvedClaude Codechaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swerebench_oraclesolved_cc_4053.jsonl601 MB4,053
swerebench_oraclesolvedClaude Codejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_oraclesolved_cc_5013.jsonl779 MB5,013
swerebench_oraclesolvedOpenCodechaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swerebench_oraclesolved_oc_3431.jsonl405 MB3,431
swerebench_oraclesolvedOpenCodejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_oraclesolved_oc_3521.jsonl464 MB3,521
swerebench_oraclesolvedOpenHands-AIchaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swerebench_oraclesolved_oh_2808.jsonl638 MB2,808
swerebench_oraclesolvedOpenHands SDKjierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_oraclesolved_oh_sdk_2733.jsonl507 MB2,733
swerebench_oraclesolvedTerminus-2chaofan/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/chaofan_glm5_swerebench_oraclesolved_t2_2920.jsonl291 MB2,920
swerebench_oraclesolvedTerminus-2jierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_oraclesolved_t2_2998.jsonl296 MB2,998
swerebench_othersClaude Codejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_others_cc_2885.jsonl457 MB2,885
swerebench_othersOpenCodejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_others_oc_2637.jsonl360 MB2,637
swerebench_othersOpenHands SDKjierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_others_oh_sdk_1843.jsonl340 MB1,843
swerebench_othersTerminus-2jierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebench_others_t2_2270.jsonl225 MB2,270
swerebenchv2_python_oraclesolvedClaude Codejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebenchv2_python_oraclesolved_cc_2728.jsonl445 MB2,728
swerebenchv2_python_oraclesolvedOpenCodejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebenchv2_python_oraclesolved_oc_1654.jsonl240 MB1,654
swerebenchv2_python_oraclesolvedOpenHands SDKjierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebenchv2_python_oraclesolved_oh_sdk_1521.jsonl313 MB1,521
swerebenchv2_python_oraclesolvedTerminus-2jierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_swerebenchv2_python_oraclesolved_t2_1633.jsonl174 MB1,633
v2nopy_fullClaude Codejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_v2nopy_full_cc_5128.jsonl818 MB5,128
v2nopy_fullOpenCodejierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_v2nopy_full_oc_3465.jsonl455 MB3,465
v2nopy_fullOpenHands SDKjierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_v2nopy_full_oh_sdk_3266.jsonl620 MB3,266
v2nopy_fullTerminus-2jierun/home/ywxzml3j/ywxzml3juser57/LLaMA-Factory/data/chaofan_jierun_traj/jierun_glm5_v2nopy_full_t2_3675.jsonl411 MB3,675

统计方法说明

平均轮数 / Token / Tool Calls

平均轮数:每条轨迹的 messages 数组长度的平均值。

平均 Token:使用 tiktoken cl100k_base tokenizer 对所有 message 的 content + reasoning_content 精确编码计数的平均值。

平均 Tool Calls:assistant 消息中 tool_calls 数组长度之和的平均值。对 Terminus-2 脚手架,统计 assistant 消息 JSON content 中 commands 数组的长度。

质量评分

composite_score(0-1)由五个维度加权:efficiency(效率)、style(风格)、tool_mastery(工具掌握)、completion(完成度)、precision(精确度)。

仅部分文件包含 _score 字段,无分数的文件显示 "—"。