Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

Benchmark Model Rank Results
code-generation-on-swe-bench-verifiedMacaron-V1-Venti (Claude Code)#3Accuracy: 85.6
code-generation-on-terminal-bench-2-1Macaron-V1-Venti (Claude Code)#6Accuracy: 87.6
optical-character-recognition-ocr-on-ocrbenchMacaron-V1-Tall#9Score: 896
code-generation-on-deepsweMacaron-V1-Venti (Claude Code)#19Pass@1: 0.584
multimodal-reasoning-on-mmmuMacaron-V1-Tall#56Accuracy: 61.22