Complex Reasoning on LLaVA Bench (val)
2.1875PerplexityMARS
Evaluation results
| Method | Links | |
|---|---|---|
| MARSModel=LLaVA-OV-7B2026.02 | 2.1875 | |
| MARSModel=LLaVA-OV-7B2026.02 | 2.1875 | |
| LoRAModel=LLaVA-OV-7B, lr_llm=1e-52026.02 | 2.2317 | |
| LoRAModel=LLaVA-OV-7B, r_llm=16, r_ve=best2026.02 | 2.2317 | |
| Full-rank TuningModel=LLaVA-OV-7B2026.02 | 2.2693 | |
| LoRAModel=LLaVA-OV-7B, lr_llm=1e-62026.02 | 2.295 | |
| LoRAModel=LLaVA-OV-7B, r_llm=8, r_ve=best2026.02 | 2.3181 | |
| LoRAModel=LLaVA-OV-7B, lr_llm=1e-72026.02 | 2.4346 | |
| LoRAModel=LLaVA-OV-7B, r_llm=32, r_ve=best2026.02 | 2.442 | |
| GeoLoRAModel=LLaVA-OV-7B2026.02 | 2.4888 | |
| AdaLoRAModel=LLaVA-OV-7B2026.02 | 2.5189 | |
| MARSModel=LLaVA-OV-0.5B2026.02 | 2.7188 | |
| MARSModel=LLaVA-OV-0.5B2026.02 | 2.7188 | |
| Full-rank TuningModel=LLaVA-OV-0.5B2026.02 | 2.7209 | |
| LoRAModel=LLaVA-OV-0.5B, r_llm=32, r_ve=best2026.02 | 2.7331 | |
| LoRAModel=LLaVA-OV-0.5B, lr_llm=1e-52026.02 | 2.7336 | |
| LoRAModel=LLaVA-OV-0.5B, r_llm=16, r_ve=best2026.02 | 2.7336 | |
| LoRAModel=LLaVA-OV-0.5B, r_llm=8, r_ve=best2026.02 | 2.7582 | |
| LoRAModel=LLaVA-OV-0.5B, lr_llm=1e-62026.02 | 2.771 | |
| LoRAModel=LLaVA-OV-0.5B, lr_llm=1e-72026.02 | 2.8472 | |
| GeoLoRAModel=LLaVA-OV-0.5B2026.02 | 2.8801 | |
| AdaLoRAModel=LLaVA-OV-0.5B2026.02 | 2.8973 | |
| MARSModel=Qwen2.5-VL-7B2026.02 | 3.3616 | |
| MARSModel=Qwen2.5-VL-7B2026.02 | 3.3879 | |
| LoRAModel=Qwen2.5-VL-7B, r_llm=8, r_ve=best2026.02 | 3.5011 | |
| LoRAModel=Qwen2.5-VL-7B, lr_llm=1e-52026.02 | 3.5032 | |
| LoRAModel=Qwen2.5-VL-7B, r_llm=16, r_ve=best2026.02 | 3.5032 | |
| MARSModel=Qwen2.5-VL-3B2026.02 | 3.5825 | |
| LoRAModel=Qwen2.5-VL-7B, lr_llm=1e-62026.02 | 3.5908 | |
| LoRAModel=Qwen2.5-VL-3B, r_llm=8, r_ve=best2026.02 | 3.5912 | |
| Full-rank TuningModel=Qwen2.5-VL-7B2026.02 | 3.5917 | |
| MARSModel=Qwen2.5-VL-3B2026.02 | 3.5925 | |
| LoRAModel=Qwen2.5-VL-7B, r_llm=32, r_ve=best2026.02 | 3.6098 | |
| LoRAModel=Qwen2.5-VL-3B, lr_llm=1e-52026.02 | 3.6156 | |
| LoRAModel=Qwen2.5-VL-3B, r_llm=16, r_ve=best2026.02 | 3.6156 | |
| GeoLoRAModel=Qwen2.5-VL-7B2026.02 | 3.6378 | |
| AdaLoRAModel=Qwen2.5-VL-7B2026.02 | 3.6394 | |
| Full-rank TuningModel=Qwen2.5-VL-3B2026.02 | 3.6528 | |
| GeoLoRAModel=Qwen2.5-VL-3B2026.02 | 3.6676 | |
| AdaLoRAModel=Qwen2.5-VL-3B2026.02 | 3.6679 | |
| LoRAModel=Qwen2.5-VL-3B, lr_llm=1e-62026.02 | 3.7415 | |
| LoRAModel=Qwen2.5-VL-3B, r_llm=32, r_ve=best2026.02 | 3.7811 | |
| LoRAModel=Qwen2.5-VL-7B, lr_llm=1e-72026.02 | 3.8716 | |
| LoRAModel=Qwen2.5-VL-3B, lr_llm=1e-72026.02 | 4.1557 |