0.3.5

The OpenCompass team is thrilled to announce the release of OpenCompress v0.3.5!

🌟 Highlights

• 🚀 Introduction of two new datasets: CMO&AIME, expanding our evaluation capabilities.
• 📖 Several updates to our documentation, ensuring clearer guidance for all users.
• ⚙ Several enhancements and refactoring efforts to make our codebase more robust and maintainable.

🚀 New Features

• 🆕 Added support for the CMO&AIME datasets, broadening the scope of models we can evaluate. (#1610)
• 🆕 Introduced the CompassArenaSubjectiveBench, a new benchmark for subjective evaluations. (#1645)
• 🆕 Added configurations for the lmdeploy DeepSeek model, enhancing compatibility with cutting-edge technologies. (#1656)

📖 Documentation

• 📚 Updated the documentation to reflect the latest changes and improvements, making it easier than ever to navigate and understand. (#1655)

🐛 Bug Fixes

• 🔧 Fixed issues with the ruler_16k_gen component, ensuring more accurate and reliable results. (#1643)
• 🔧 Resolved an error in the get_loglikelihood function when using lmdeploy as the accelerator. (#1659)
• 🔧 Addressed problems with automatic downloads for certain datasets, streamlining the user experience. (#1652)

⚙ Enhancements and Refactors

• 💪 Enhanced the summarizer configurations for models, improving the efficiency and effectiveness of summarization tasks. (#1600)
• 💪 Added new model configurations, keeping up with the latest advancements in machine learning. (#1653)
• 💪 Updated the WildBench maximum sequence length, allowing for better handling of longer input sequences. (#1648)
• 💪 Updated the Needlebench OSS path, ensuring smoother data access and processing. (#1651)
• 💪 Improved the mmmlu_lite dataloader, optimizing data loading processes. (#1658)

🎉 Welcome New Contributors

• 👏 A warm welcome to @jnanliu, who has made their first contribution by adding the CMO&AIME datasets! (#1610)

For a complete overview of all changes, please refer to the full changelog: 0.3.4...0.3.5

0.3.4

The OpenCompass team is thrilled to announce the release of OpenCompass v0.3.4!

🎉 OpenCompass v0.3.4 brings major enhancements including new benchmarks, improved documentation, and numerous bug fixes.
🌈 Notable features include support for new datasets and the integration of lmdeploy pipeline API.

🔧 Support for New Datasets:

• Addition of GaoKaoMath Dataset for Evaluation.
• Support for MMMLU & MMMLU-lite Benchmark.
• Integration of Judgerbench and reorganization of subeval.
• Support for LiveCodeBench.

📝 Output Format Enhancements:

• Support for printing and saving results as markdown format tables.

🔧 Pipeline and Integration Improvements:

• Integration of lmdeploy pipeline API.
• Update of TurboMindModel through integration of lmdeploy pipeline API.
• Removal of prefix bos_token from messages when using lmdeploy as the accelerator.

🛠️ Miscellaneous Enhancements:

• Updates to the common summarizer regex extraction.
• Internal humaneval postprocess addition and updates.

📖 Documentation Updates

🐛 Bug Fixes

🎉 Welcome New Contributors
👋 New Contributors Joined the Team:

@BobTsang1995 - Contributed support for MMMLU & MMMLU-lite Benchmark.
@noemotiovon - Provided NPU support fixes.
@changlan - Fixed RULER datasets.
@BIGWangYuDong - Added support for printing and saving results as markdown format tables.
Thank you to all contributors who have made this release possible. For a complete list of changes, please see the full changelog linked below.

Full Changelog: 0.3.3...0.3.4

0.3.3

🌟 OpenCompass v0.3.3 Release Log
The OpenCompass team is thrilled to announce the release of OpenCompass v0.3.3!

🚀 New Features

• 🔧 Added support for the SciCode summarizer configuration.
• 🛠 Introduced support for internal Followbench.
• 🔧 Updated models and configurations for MathBench & WikiBench under FullBench.
• 🛠 Enhanced support for OpenAI O1 models and Qwen2.5 Instruct.
• 🔧 Included a postprocess function for custom models.
• 🛠 Added InternTrain feature for broader model training scenarios.

📖 Documentation

• 📚 Updated the README with the latest information on how to use OpenCompass effectively.

🐛 Bug Fixes

• 🔧 Fixed issues with the link-check workflow and wildbench.
• 🛠 Resolved errors in partitioning and corrected typos throughout the codebase.
• 🔧 Addressed compatibility issues with lmdeploy interface type changes.
• 🛠 Fixed the followbench dataset configuration and token settings.

⚙ Enhancements and Refactors

• 🛠 Enhanced support for verbose output in OpenAI API interactions.
• 🔧 Updated maximum output length configurations for multiple models.
• 🛠 Improved handling of the "begin section" in meta_template for better parsing.
• 🔧 Added a common summarizer for qabench and expanded test coverage for various models.

🎉 Welcome New Contributors
👋 We'd like to extend a warm welcome to our new contributors who have made their first contributions to OpenCompass:

• @x54-729 introduced InternTrain.
• @chuanyangjin helped correct typos.
• @cuauty added support for reasoning from BaiLing LLM.

Thank you to all our contributors for making this release possible!

Full Changelog: 0.3.2.post1...0.3.3

0.3.2.post1

What's Changed

• [Fix]Init import fix by @MaiziXiao in #1500
• [Bump] Bump version to 0.3.2.post1 by @MaiziXiao in #1502

Full Changelog: 0.3.2...0.3.2.post1

0.3.2

The OpenCompass team is thrilled to announce the release of OpenCompass v0.3.2!

🚀 New Features

• 🛠 Added extra_body support for OpenAISDK and introduced proxy URL support when connecting to OpenAI's API.
• 🗂 Included auto-download functionality for Mmlu-pro, Needlebench, Longbench and other datasets.
• 🤝 Integrated support for the Rendu API.
• 🧪 Added a model postprocess function.

📖 Documentation

• 📜 Updated the README file for better clarity and guidance.

🐛 Bug Fixes

• 🛠 Fixed CLI evaluation for multiple models.
• 🛠 Updated requirements to resolve dependency issues.
• 🛠 Corrected configurations for the Llama model series.
• 🛠 Addressed bad cases and added environment information to improve testing.

⚙ Enhancements and Refactors

• 🛠 Made OPENAI_API_BASE compatible with OpenAI's default environment settings.
• 🛠 Optimized SciCode for improved performance.
• 🛠 Added an api_key attribute to TurboMindAPIModel.
• 🛠 Implemented fixes and improvements to the CI test environment, including baselines for vllm.

🎉 Welcome New Contributors

• 👋 @cpa2001 contributed with the addition of icl_sliding_k_retriever.py and updates to __init__.py.
• 👋 @gyin94 made the OPENAI_API_BASE compatible with OpenAI's default environment.
• 👋 @chengyingshe added an attribute api_key into TurboMindAPIModel.
• 👋 @yanzeyu supported the integration of Rendu API.

Full Changelog: 0.3.1...0.3.2

OpenCompass v0.3.1

The OpenCompass team is thrilled to announce the release of OpenCompass v0.3.1!

---

🌟 Highlights

• 🚀 Support pip installation, update Readme and evaluation demo
• 🐛 Fixed various dataset loading issues.
• ⚙️ Enhanced auto-download features for datasets.

---

🚀 New Features

• 🆕 Introduced support for Ruler datasets.
• 🆕 Enhanced model compatibility.
• 🆕 Improved dataset handling, support auto-download for various datasets

---

📖 Documentation

• 📚 Updated README to reflect the latest changes.
• 📚 Improved documentation for dataset loading procedures.

---

🐛 Bug Fixes

• 🐞 Resolved modelscope dataset load issues.
• 🐞 Corrected evaluation scores for the Lawbench dataset.
• 🐞 Fixed dataset bugs for CommonsenseQA and Longbench.

---

⚙ Enhancements and Refactors

• 🔧 Retained first and last halves of prompts to avoid max_seq_len issues.
• 🔧 Updated Compassbench to v1.3.
• 🔧 Switched to Python runner for single GPU operations.

---

🎉 Welcome New Contributors

• 🙌 @Yunnglin for fixing modelscope dataset load problem.
• 🙌 @changyeyu for addressing max_seq_len issues with prompt handling.
• 🙌 @seetimee for updates to openai_api.py.
• 🙌 @HariSeldon0 for adding the scicode dataset.

---

What's Changed

• [Fix] Fix modelscope dataset load problem by @Yunnglin in #1406
• [Fix] the issue where scores are negative in the Lawbench dataset evaluation(#1402) by @yaoyingyy in #1403
• [Doc] Update README by @tonysy in #1404
• Retain first and last halves of prompts to avoid max_seq_len issues by @changyeyu in #1373
• [UPDATE] Compassbench v1.3 by @MaiziXiao in #1396
• [Fix] longbench dataset load fix by @MaiziXiao in #1422
• [Fix] Sub summarizer order fix by @bittersweet1999 in #1426
• [Update] Support auto-download of FOFO/MT-Bench-101 by @tonysy in #1423
• [Bug] Commonsenseqa dataset fix by @MaiziXiao in #1425
• [Feature] Add abbr for rolebench dataset by @xu-song in #1431
• [Feature] Add Ruler datasets by @MaiziXiao in #1310
• [Fix] Fix openai api tiktoken bug for api server by @liushz in #1433
• Update openai_api.py by @seetimee in #1438
• [Feature] Add model support for 'huggingface_above_v4_33' when using '-a' by @liushz in #1430
• Add scicode by @HariSeldon0 in #1417
• [Doc] Update Readme by @MaiziXiao in #1439
• [Fix] Update option postprocess & mathbench language summarizer by @liushz in #1413
• [ci] add commond testcase into daily testcase by @zhulinJulia24 in #1447
• [Feature] Switch to python runner for single GPU by @xu-song in #1308
• [Fix] Update SciCode and Gemma model by @tonysy in #1449
• [Bump] Bump version to 0.3.1 by @tonysy in #1450

Full Changelog: 0.3.0...0.3.1

---

Thank you for your continued support and contributions to OpenCompass!

OpenCompass v0.3.0

The OpenCompass team is thrilled to announce the release of OpenCompass v0.3.0! This release brings a variety of new features, enhancements, and bug fixes to improve your experience.

🌟 Highlights

1. Support for OpenAI ChatCompletion
2. Updated Model Support List
3. Support Dataset Automatic Download
4. Support pip install opencompass

🚀 New Features

1. Support for CompassBench Checklist Evaluation
   • PR #1339 by @bittersweet1999
2. Adding support for Doubao API
   • PR #1218 by @LeavittLang
3. Support for ModelScope Datasets
   • PR #1289 by @wangxingjun778

📖 Documentation

1. Update NeedleBench Docs
   • PR #1330 by @DseidLi
2. Update Documentation
   • PR #1318 by @Leymore

🐛 Bug Fixes

1. Fix Typing and Typo
   • PR #1331 by @xu-song
2. Fix Lint Issues
   • PR #1334 by @DseidLi
3. Fix Summary Error in subjective.py
   • PR #1363 by @WenjinW

⚙ Enhancements and Refactors

1. Upgrade Default Math pred_postprocessor
   • PR #1340 by @xu-song
2. Fix Path and Folder Updates
   • PR #1344 by @tonysy
3. Update Get Data Path for LCBench and HumanEval
   • PR #1375 by @tonysy

🔗 Full Change Logs

• [Fix] Change abbr for arenahard dataset by @bittersweet1999 in #1302
• [Fix] Force register by @Leymore in #1311
• [Fix] add bc for alignbench summarizer by @bittersweet1999 in #1306
• [Fix] update Faq by @bittersweet1999 in #1313
• [Fix] Fix rouge evaluator of rolebench_zh by @xu-song in #1322
• [Doc] Update NeedleBench Docs by @DseidLi in #1330
• [Fix] Fix typing and typo by @xu-song in #1331
• [Fix] Fix lint by @DseidLi in #1334
• [Feature] support compassbench Checklist evaluation by @bittersweet1999 in #1339
• Add compassbench wiki&math part by @liushz in #1342
• Compassbench v1_3 subjective evaluation by @MaiziXiao in #1341
• [Fix] Update path and folder by @tonysy in #1344
• Upgrade default math pred_postprocessor by @xu-song in #1340
• commit inference ppl datasets by @Quehry in #1315
• CompassBench subjective summarizer added by @MaiziXiao in #1349
• Fix MathBench Generation Config by @liushz in #1351
• [Update] Update model support list by @bittersweet1999 in #1353
• [Update] update Subeval demo config by @bittersweet1999 in #1358
• [Fix] Fix the summary error in subjective.py by @WenjinW in #1363
• [Fix] Support HF models deployed with an OpenAI-compatible API. by @heya5 in #1352
• update docs by @Leymore in #1318
• [Feature] Make NeedleBench available on HF by @DseidLi in #1364
• 【bug fix】: Remove extra ampersands. by @baymax591 in #1365
• [Fix] minor update wildbench by @kleinzcy in #1335
• Adding support for Doubao API by @LeavittLang in #1218
• [Fix] origin_prompt should be None in llm-compression task by @mqy004 in #1225
• Calm dataset by @pengbo807 in #1287
• Add en and zh groups to longbench summarizer; Fix longbench overall score by @xu-song in #1216
• [Revert] "Calm dataset (#1287)" by @bittersweet1999 in #1366
• Charm by @jxd0712 in #1230
• Support ModelScope datasets by @wangxingjun778 in #1289
• [Feature] Update pip install by @tonysy in #1324
• add support for hf_pulse_7b by @QXY716 in #1255
• [Fix] Update get_data_path for LCBench and HumanEval by @tonysy in #1375
• [Bug] Fix bug in turbomind by @tonysy in #1377
• [Fix] Fix version mismatch of CIBench by @kleinzcy in #1380
• [Fix] Fix InternLM2.5-7B-Chat-1M config by @DseidLi in #1383
• [Feature] Support import configs/models/summarizers from whl by @tonysy in #1376
• Calm dataset by @pengbo807 in #1385
• [Feature] Support OpenAI ChatCompletion by @tonysy in #1389
• [Fix] Fix slurm env by @tonysy in #1392
• [Fix] Fix CaLM import by @tonysy in #1395
• [Bump] Bump version for v0.3.0 by @tonysy in #1398

🎉 Welcome New Contributors

• @MaiziXiao made their first contribution in #1341
• @Quehry made their first contribution in #1315
• @WenjinW made their first contribution in #1363
• @heya5 made their first contribution in #1352
• @LeavittLang made their first contribution in #1218
• @pengbo807 made their first contribution in #1287
• @wangxingjun778 made their first contribution in #1289
• @QXY716 made their first contribution in #1255

Full Changelog: 0.2.6...0.3.0

OpenCompass v0.2.6

The OpenCompass team is thrilled to announce the release of OpenCompass v0.2.6!

🌟 Highlights

• No noteworthy highlights.

🚀 New Features

1. #1215 #1224 #1266 Add Datasets MT-Bench-101, Fofo, wildbench
2. #1286 Add Models InternLM2.5-7B

📖 Documentation

1. #1252 Add doc for accelerator function
2. #1263 Update quick start guide

🐛 Bug Fixes

1. #1221 Resolve release version installation and import issues
2. #1228 Fix pip version issues
3. #1282 Update MathBench summarizer & fix cot setting

⚙ Enhancements and Refactors

1. #1284 Reorganize subjective eval

🎉 Welcome New Contributors

• @mqy004, @sefira, @Zor-X-L and @baymax591 made their first contributions. Welcome to the OpenCompass community!

🔗 Full Change Logs

• [Fix] fix summarizer by @bittersweet1999 in #1217
• 解决release版本安装后不能导入opencompass.cli.main的问题 by @mqy004 in #1221
• MT-Bench-101 by @sefira in #1215
• [Feature] add dataset Fofo by @bittersweet1999 in #1224
• [Fix] fix pip version by @bittersweet1999 in #1228
• add ",<2.0.0" to "numpy>=1.23.4" in requirements/runtime.txt, as pand… by @Zor-X-L in #1267
• Support wildbench by @kleinzcy in #1266
• Add doc for accelerator function by @liushz in #1252
• flash attn installation in daily testcase by @zhulinJulia24 in #1272
• Update mtbench101.py by @sefira in #1276
• [Sync] Sync with internal codes 2024.06.28 by @Leymore in #1279
• Update MathBench summarizer & fix cot setting by @liushz in #1282
• npu适配 by @baymax591 in #1250
• [ci] update daily testcase by @zhulinJulia24 in #1285
• [Feature] Add InternLM2.5 by @tonysy in #1286
• [Feat] Update owners for issues by @tonysy in #1293
• [Refactor] Reorganize subjective eval by @bittersweet1999 in #1284
• [Doc] quick start swap tabs by @Leymore in #1263

Full Changelog: 0.2.5...0.2.6

OpenCompass v0.2.5

The OpenCompass team is thrilled to announce the release of OpenCompass v0.2.5!

🌟 Highlights

• Simplify the huggingface / vllm / lmdeploy model wrapper. meta_template is no longer needed to be hand-crafted in model configs
• Introduce evaluation results README in ~20 dataset config folders.

🚀 New Features

1. #1065 Add LLaMA-3 Series Configs
2. #1048 Add TheoremQA with 5-shot
3. #1094 Support Math evaluation via judgemodel
4. #1080 Add gpqa prompt from simple_evals, openai
5. #1074 Add mmlu prompt from simple_evals, openai
6. #1123 Add Qwen1.5 MoE 7b and Mixtral 8x22b model configs

📖 Documentation

1. #1053 Update readme
2. #1102 Update NeedleInAHaystack Docs
3. #1110 Update README.md
4. #1205 Remove --no-batch-padding and Use --hf-num-gpus

🐛 Bug Fixes

1. #1036 Update setup.py install_requires
2. #1051 Fixed the issue caused
3. #1043 fix multiround
4. #1070 Fix sequential runner
5. #1079 Fix Llama-3 meta template

⚙ Enhancements and Refactors

1. #1163 enable HuggingFacewithChatTemplate with --accelerator via cli
2. #1104 fix prompt template
3. #1109 Update performance of common benchmarks

🎉 Welcome New Contributors

• @liuwei130, @IcyFeather233, @VVVenus1212, @binary-husky, @dmitrysarov, @eltociear, @acylam, @lfy79001, @JuhaoLiang1997, @yaoyingyy, and @jxd0712 made their first contributions. Welcome to the OpenCompass community!

🔗 Full Change Logs

• [Fix] Update setup.py install_requires by @Leymore in #1036
• add ChemBench by @liuwei130 in #1032
• [Fix] logger.error -> logger.debug in OpenAI by @Leymore in #1050
• [Sync] Bump version to 0.2.4 by @Leymore in #1052
• [Doc] Update readme by @tonysy in #1053
• [fix]Fixed the issue caused by the repeated loading of VLLM model dur… by @IcyFeather233 in #1051
• [Sync] Sync with internal code 2024.04.19 by @Leymore in #1064
• [Fix] fix multiround by @bittersweet1999 in #1043
• [Feature] Add LLaMA-3 Series Configs by @Leymore in #1065
• [Feature] Add TheoremQA with 5-shot by @Leymore in #1048
• [Fix] Fix sequential runner by @Leymore in #1070
• Add lmdeploy tis python backend model by @ispobock in #1014
• Fix Llama-3 meta template by @liushz in #1079
• Add humaneval prompt from simple_evals, openai by @jingmingzhuo in #1076
• [Feature] Support Math evaluation via judgemodel by @bittersweet1999 in #1094
• [Feature] support arenahard evaluation by @bittersweet1999 in #1096
• Update CIBench by @kleinzcy in #1089
• [Feature] Add gpqa prompt from simple_evals, openai by @Francis-llgg in #1080
• [Deperecate] Remove multi-modal related stuff by @kennymckormick in #1072
• add vllm get_ppl by @VVVenus1212 in #1003
• fix: python path bug by @binary-husky in #1063
• fix output typing, change mutable list to immutable tuple by @dmitrysarov in #989
• [Doc] Update NeedleInAHaystack Docs by @DseidLi in #1102
• [Feature] add support for Flames datasets by @Yggdrasill7D6 in #1093
• adapt to lmdeploy v0.4.0 by @lvhan028 in #1073
• [Fix] fix prompt template by @bittersweet1999 in #1104
• [Fix] Fix Math Evaluation with Judge Model Evaluator & Add README by @liushz in #1103
• [Update] Update performance of common benchmarks by @tonysy in #1109
• [Fix] fix cmb dataset by @bittersweet1999 in #1106
• [Docs] Update README.md by @eltociear in #1110
• [Feature] Adding support for LLM Compression Evaluation by @acylam in #1108
• [Fix] remove redundant pre-commit check by @Leymore in #891
• fix LightllmApi workers bug by @helloyongyang in #1113
• [Feature] Add mmlu prompt from simple_evals, openai by @Leymore in #1074
• [Feature] update drop dataset from openai simple eval by @kleinzcy in #1092
• add mgsm datasets by @Yggdrasill7D6 in #1081
• [Fix] Fix AGIEval chinese sets by @xu-song in #972
• S3Eval Dataset by @lfy79001 in #916
• [Feature] Add AceGPT-MMLUArabic benchmark by @JuhaoLiang1997 in #1099
• [Fix] fix links by @bittersweet1999 in #1120
• [Fix] Fix NeedleBench Summarizer Typo by @DseidLi in #1125
• [Feature] Add Qwen1.5 MoE 7b and Mixtral 8x22b model configs by @acylam in #1123
• [Sync] Update accelerator by @Leymore in #1122
• [Fix] fix alpacaeval while add caching path by @bittersweet1999 in #1139
• [Fix] fix multiround by @bittersweet1999 in #1146
• [Fix] Fix Needlebench Summarizer by @DseidLi in #1143
• [Feature] Add huggingface apply_chat_template by @Leymore in #1098
• [Feat] Support dataset_suffix check for mixed configs by @xu-song in #973
• [Format] Add some config lints by @Leymore in #892
• [Sync] Sync with internal codes 2024.05.14 by @Leymore in #1156
• [Fix] fix arenahard summarizer by @bittersweet1999 in #1154
• [Fix] use ProcessPoolExecutor during mbpp eval by @Leymore in #1159
• [Fix] Update stop_words in huggingface_above_v4_33 by @Leymore in #1160
• Update accelerator by @liushz in #1152
• [Feat] enable HuggingFacewithChatTemplate with --accelerator via cli by @Leymore in #1163
• update test workflow by @zhulinJulia24 in #1167
• [Sync] Sync with internal codes 2024.05.17 by @Leymore in #1171
• add dependency in daily test workflow by @zhulinJulia24 in #1173
• [Sync] Sync with internal codes 2024.05.21.1 by @Leymore in #1175
• Update MathBench by @liushz in #1176
• [Fix] fix template by @bittersweet1999 in #1178
• Fix a bug in drop_gen.py by @kleinzcy in #1191
• [Fix] temporary files using tempfile by @yaoyingyy in #1186
• [Fix] add support for lmdeploy api judge by @bittersweet1999 in #1193
• [Fix] fix length by @bittersweet1999 in #1180
• support CHARM (https://github.com/opendatalab/CHARM) reasoning tasks by @jxd0712 in #1190
• [Feat] Update charm summary by @Leymore in #1194
• Update accelerator by @liushz in #1195
• [Sync] S...

OpenCompass v0.2.5.rc1

[Feature] Add lmdeploy tis python backend model (#1014)

* add lmdeploy tis python backend model

* fix pr check

* update