Skip to content

VenusFactory2 v0.2.0

Choose a tag to compare

@tyang816 tyang816 released this 26 Jul 09:11
· 157 commits to main since this release

Summary

Milestone release after checkpoint hub, one-click installer, and documentation restructure.

Highlights

  • Checkpoint hub: weights hosted on Hugging Face (AI4Protein/VenusFactory2-ckpts); local ckpt/ via scripts/download_ckpts.py / on-demand fetch
  • One-click install: python scripts/setup_quickstart.py (deps + frontend + predict-core weights)
  • Docs: slim README + docs/wiki/ + docs/manual/ (see docs/README.md)

Install

git clone https://github.com/AI4Protein/VenusFactory2.git && cd VenusFactory2
python scripts/setup_quickstart.py
source .venv/bin/activate
python src/webui_v2.py --host 0.0.0.0 --port 7861

Not a pure pip install product package yet — torch/PyG, frontend build, and checkpoints are handled by the installer / hub scripts.

Notes

  • Python ≥3.12, Node.js 25.x for WebUI v2
  • Slow HF access: export HF_ENDPOINT=https://hf-mirror.com

What's Changed

  • Use LoRA finetuning model by @code4luck in #1
  • Eval_tab completed and Multiple bugs fixed. by @shenmaa233 in #2
  • Add and modify LoRA training parameters and procedures by @code4luck in #3
  • Add and motify the lora and qlora training methods by @code4luck in #4
  • update train and eval code for plm-lora and qlora by @code4luck in #6
  • Update ProSST, ProPrime, and requirement.txt by @zlr-zmm in #7
  • merge and update code by @code4luck in #8
  • Update WebUI and Add dataset preview by @shenmaa233 in #5
  • update esm1b and esm1v tokenizer code and model judgement code by @code4luck in #9
  • Correct merge errors from PR integration & Add lora/qlora to eval/predict by @shenmaa233 in #11
  • add DoRA, AdaLoRA, and IA3 by @zlr-zmm in #12
  • update requirements.txt by @zlr-zmm in #13
  • Better tables and monitor figures by @shenmaa233 in #14
  • Update readme inital by @code4luck in #15
  • add PETA model by @zlr-zmm in #16
  • Add preview cmd button for eval and predict by @shenmaa233 in #17
  • Add pics to README by @shenmaa233 in #18
  • Update WebUI including abort and initialization by @shenmaa233 in #19
  • Add the manual tab by @shenmaa233 in #20
  • Update requirements.txt by @zlr-zmm in #22
  • Update requirements.txt by @zlr-zmm in #23
  • Update demo by @shenmaa233 in #24
  • Fix the DeepET_Topt dataset path by @shenmaa233 in #25
  • Add guide notebook by @code4luck in #27
  • add guide notebook by @code4luck in #28
  • update requirements.txt and readme.md by @zlr-zmm in #29
  • update readme.md by @zlr-zmm in #30
  • update readme.md by @zlr-zmm in #31
  • update readme.md by @zlr-zmm in #32
  • Fix some small bugs in eval tab by @shenmaa233 in #33
  • add vplm to requirements by @mingchen-li in #34
  • Update adalora dora ia3 in webui by @shenmaa233 in #35
  • Fix the metric setting bugs for custom datasets by @shenmaa233 in #37
  • Fix bugs in scheduler by @shenmaa233 in #38
  • Fix the problem of continuous training by @shenmaa233 in #39
  • update requirements.txt by @zlr-zmm in #45
  • solving problems when fine-tuning by @zlr-zmm in #46
  • upload demo model weight by @zlr-zmm in #48
  • Added inference scripts using adapter modules based on Ankh, ESM-2, ProtBERT and ProtT5. by @zlr-zmm in #51
  • Refactored the UI and introduced a "Quick Start" module to provide a clearer workflow and direct access to inference tools. by @zlr-zmm in #52
  • Optimised zero-shot prediction with visualisation content by @zlr-zmm in #53
  • Improved the functionality of the Quick Use module by @zlr-zmm in #54
  • Integration of Zero-shot and protein function prediction by @zlr-zmm in #55
  • Create ESM-1b in Zero-Shot by @ShaojunZhai in #56
  • Added protein 3D structure viewing function based on Molecular 3D Viewer by @zlr-zmm in #59
  • Add the by @ShaojunZhai in #61
  • 1、Updated the protein structure display module;2、Modified the Quick Start related functions; 3、Added the Example file to facilitate the use of users by @zlr-zmm in #62
  • Improved the functionality of the Agent and fixed existing bugs. by @zlr-zmm in #63
  • fix bugs by @zlr-zmm in #64
  • Counting function by @ShaojunZhai in #65
  • Update index_tab.py by @ShaojunZhai in #66
  • A brand new interface has been designed for VenusAgent, which is more aesthetically pleasing. In addition, some existing bugs have been fixed by @zlr-zmm in #67
  • Fixed Advance tool's inability to use AI analysis for protein mutatio… by @zlr-zmm in #68
  • Added a progress bar and support for NCBI data downloads. by @zlr-zmm in #70
  • Update UI for AI Analysis and Metal Ion Binding by @zlr-zmm in #71
  • Update index_tab.py by @ShaojunZhai in #72
  • Fixed the issue where AI Expert output was too long, resulting in incomplete display by @zlr-zmm in #73
  • Optimized the functions and improved some of them by @zlr-zmm in #74
  • Index更新 计数功能完善 by @ShaojunZhai in #75
  • Update index_tab.py by @ShaojunZhai in #76
  • Added protein function prediction using ProtBert. by @zlr-zmm in #77
  • Solved the problem of uploading protein structures in the Agent that cannot predict mutations by @zlr-zmm in #78
  • Fix the issue of uploading protein pdb incorrectly on the advance tool by @zlr-zmm in #79
  • Add the function of reading file content and forming a temp_file in quick_tool and advanced_tool by @ShaojunZhai in #80
  • Updated the index interface. Users can now click on 'HyberAI' to be directed to a tutorial on using VenusFactory within the HyberAI platform. by @zlr-zmm in #82
  • Optimized the download layout and updated the README documentation. by @zlr-zmm in #83
  • Update index_tab.py by @ShaojunZhai in #85
  • Update index_tab.py by @ShaojunZhai in #86
  • Added support for fasta and PDB paste input, and readjusted the index interface by @zlr-zmm in #88
  • This update introduces VenusScope, a new tool that generates comprehensive reports from uploaded or pasted sequences, now with added support for analyzing physicochemical features and predicting functional residues. Powered by more advanced AI agents and a Langchain-based memory function, the tool now enables users to ask context-aware, follow-up questions for a more interactive analysis. by @zlr-zmm in #91
  • Patiskey patch 3 by @Patiskey in #90
  • Fix the issue of not being able to create a cache folder correctly by @zlr-zmm in #92
  • Fix the bug when run the residue-function prediction task and beauty the chat tab by @zlr-zmm in #93
  • Move the index button before the training button of the manual button by @Patiskey in #94
  • Refactor Agent with Multi-Agent Support, UniProt/PDB Integration, and Bug Fixes by @zlr-zmm in #95
  • Fix bugs by @zlr-zmm in #96
  • Manual update by @Patiskey in #97
  • Modifying requirements and tutorial by @MrDotXin in #98
  • Refactor core modules and optimize agent performance by @zlr-zmm in #99
  • Update example case for VenusFactoty by @zlr-zmm in #100
  • Renamed the protein function prediction feature for improved accuracy. by @zlr-zmm in #101
  • Renamed the protein function prediction feature for improved accuracy. by @zlr-zmm in #102
  • Updated file permissions for the downloaded zip archive. by @zlr-zmm in #103
  • Optimized user experience and fixed some bugs by @zlr-zmm in #104
  • fix bugs by @zlr-zmm in #105
  • Add VenusREM model, and re-optimize planer prompt by @zlr-zmm in #106
  • Optimized the chat function of the Agent, and sent the results of historical tool calls to the Planner, eliminating the need for multiple executions when calling the tools repeatedly. In addition, protein mutation prediction has added VenusREM based on the foldseek version by @zlr-zmm in #107
  • Optimized context transfer and added structured tool output to the agent. by @zlr-zmm in #108
  • Remove VenusMine for web server by @zlr-zmm in #109
  • 1、Bug Fixes: Resolved existing issues. 2、Refactoring: Optimized code structure/architecture. 3、New Feature: Added literature search capabilities (supports PubMed and arXiv). by @zlr-zmm in #110
  • Increase the selection of multiple AI models by @zlr-zmm in #111
  • Add comprehensive FastAPI support to the project by @zlr-zmm in #112
  • Enhance AI expert analysis with model selection support by @zlr-zmm in #113
  • Merge Dev to Main by @zlr-zmm in #114
  • Add MCP support, standardize file paths, and fix server download issues by @zlr-zmm in #115
  • Add MCP support, standardize file paths, and fix server download issuesDev by @zlr-zmm in #116
  • Add MCP server by @zlr-zmm in #117
  • Fix UI layout issues by @zlr-zmm in #118
  • Update MCP Server by @zlr-zmm in #119
  • Update MCP server by @zlr-zmm in #120
  • Update MCP server by @zlr-zmm in #121
  • Feat/auto install by @MrDotXin in #123
  • Feat/auto install by @MrDotXin in #124
  • fix -index-url by @MrDotXin in #125
  • Dev by @zlr-zmm in #126
  • Improved the user experience of ai_comde-execution_tool by @zlr-zmm in #127
  • Optimized the output of zero shot tools to reduce the issue of output exceeding the token limit due to input proteins being too long by @zlr-zmm in #129
  • Fix/remove pip venv by @MrDotXin in #128
  • Modify the output of zero shot to support outputting file paths by @zlr-zmm in #130
  • Fix existing bug by @zlr-zmm in #131
  • Support using chat for model training and validation by @zlr-zmm in #132
  • Support processing of multi chain structures by @zlr-zmm in #133
  • Fix issues with pasting sequence or PDB content by @zlr-zmm in #134
  • Update tools list to include ESMFold and Web Search by @zlr-zmm in #135
  • Add new tools: FoldSeek Search and Protein Structure Prediction with … by @zlr-zmm in #136
  • Bugfix: FoldSeek tool functionality by @zlr-zmm in #137
  • Add new tools: search_dataset, search_literature, search_web by @zlr-zmm in #138
  • Add agent performance evaluation module by @zlr-zmm in #139
  • feat: add ProteinMPNN for protein de novo design by @zlr-zmm in #140
  • Integrate Human Protein Atlas (HPA) database for multi-dimensional biological analysis by @tqj0802 in #141
  • fix 1 deployment bug for local mode by @Patiskey in #145

New Contributors

Full Changelog: https://github.com/ai4protein/VenusFactory2/commits/v0.2.0