enmingzhangzz/Qwen2.5-VL-7B-OPSD-VisionZip-r010-Fbottom20abs-delta04-10240

Qwen2.5-VL-7B-OPSD-VisionZip-r010-Fbottom20abs-delta04-10240

Final PEFT/LoRA adapter from an OPSD experiment.

Variant

• —Objective: OPSD + bottom-|F| token reweighting
• —Base model: Qwen/Qwen2.5-VL-7B-Instruct
• —Training samples: 10240
• —Dataset tag: openmmreasoner_llava_cot_exact_prefix10240_decontam_v1_seed42
• —Balanced base-outcome sampling: false
• —Pruning: visionzip
• —Vision-token retention ratios: [0.1]
• —EMA teacher decay: 0.9999
• —Global batch size: 32 (4 GPUs x micro-batch 8 x accumulation 1)
• —LoRA: r=16, alpha=32
• —Image pixels: 846720

• —F reweighting mode: token_projection_fraction_bottom_abs_reweight
• —F intervention gap: 0.01 (absolute retention ratio)
• —Bottom-|F| fraction: 0.2
• —Bottom-token weight delta: 0.4
• —Mean-normalized token weights: True

Files

adapter_model.safetensors and adapter_config.json are the final adapter at step 10240. Audit and reproducibility metadata are under training/.

Adapter SHA256: 218a7cdb3358d9a30ddbd564e5b8fe9b0aadb08dda4a280e7a53016e6649d240

Load this adapter on top of Qwen/Qwen2.5-VL-7B-Instruct with PEFT. The VisionZip runtime patch used by the OPSD repository is still required for pruned inference.