LoRA adapter for Qwen2.5-VL-7B-Instruct, trained with the evidence margin alone at α = 0.5, with no placebo margin, no bound and uniform operator weights (ablation).
Released with the paper Accuracy Is Not Evidence: Measuring and Training Evidence Dependence in Video Temporal Grounding (ICLR 2027 submission, under review). Code, data identifiers and evaluation scripts are in iclropen/evidence-audit-code.
It appears in Tables 2, 6, 7, 13 and 16 and in Figures 3 and 5 of the paper.
Load it with transformers 5.x (tested 5.12.1, peft 0.19.1 or later).
Adapters saved under transformers 4.51 and 5 use different module names. Loading one under the other version raises no error but leaves the LoRA weights unused, so use the version given here. tools/check_lora_keys.py in the code repository checks that an adapter actually loaded.
# pip install "transformers>=5.12" "peft>=0.19.1" qwen-vl-utils
import torch
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from peft import PeftModel
base = "Qwen/Qwen2.5-VL-7B-Instruct"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(base, torch_dtype=torch.bfloat16, device_map="auto")
processor = AutoProcessor.from_pretrained(base)
model = PeftModel.from_pretrained(model, "iclropen/evidence-audit-qwen7b-evidence-margin-only-a05-s0")
The evaluation and audit scripts of the code repository take --model <base> --adapter <this repository or folder>.
An ablation of the objective in Section 4 of the paper. It keeps the answer likelihood, weighted by α, and only the evidence margin: the annotated answer must be more likely on the clean video than on the video with the event edited.
| Setting | Value |
|---|---|
| Operator weights | uniform |
| α | 0.5 |
| β | 0.1 |
| Placebo bound | none |
| Training pairs | 2,400 from the Charades-STA training split |
| Optimizer steps | 300 |
| Gradient accumulation | 8 |
| Learning rate | 5e-6 |
| Seed | 0 |
| LoRA | r=16, alpha=32, dropout 0.05, on q/k/v/o and gate/up/down projections |
| Edit operators | swap, mask, blur, tshift, shuffle (two per example) |
| Video input | 2 fps, at most 768 frames |
| Prompt template | timer1_default |
| Trainer | code/27_papo_train_qcdro_v3seln_r5.py |
Anonymous. Accuracy Is Not Evidence: Measuring and Training Evidence Dependence in Video Temporal Grounding. ICLR 2027 submission, under review.
Qwen/Qwen2.5-VL-7B-Instruct