Add Btoks MMEB-V2 results
Submit Btoks results for MMEB-V2.
Model: Btoks
Experiment: sieve_qwen3vl_2b_v1_expanded_train_v2_weight0.2_0513
Checkpoint: checkpoint-4500
Technical report: https://arxiv.org/abs/2604.11095
We submit the MMEB-V2 results for Btoks.
Model: Btoks
Experiment: sieve_qwen3vl_2b_v1_expanded_train_v2_weight0.2_0513
Checkpoint: checkpoint-4500
Technical report / arXiv: https://arxiv.org/abs/2604.11095
Local validation with the leaderboard scripts gives:
Overall 68.29, Image-Overall 71.55, Video-Overall 49.12, Visdoc-Overall 77.77.
Additional model detail: Btoks is fine-tuned from Qwen/Qwen3-VL-2B-Instruct. I updated scores/Btoks.json to include this as metadata ( and ). The leaderboard scores are unchanged.
Additional model detail, clarified: Btoks is fine-tuned from Qwen/Qwen3-VL-2B-Instruct. The submission JSON now records model_backbone = Qwen3-VL-2B-Instruct and base_model = Qwen/Qwen3-VL-2B-Instruct. The leaderboard scores are unchanged.