4b-8b-qwen-vl-video-models-post-trained-on-policy-self-distillation-conditioning-frozen CLM-0530For 4B-8B Qwen-VL video models post-trained by on-policy self-distillation, conditioning the frozen self-teacher only on the annotated evidence frames (while the student sees the full video) yields a more accurate teacher and better student scores than giving the teacher the same frames plus non-evidence context, and teacher accuracy falls monotonically as temporally irrelevant frames are added.
Capability: Reasoning about time in video
Observed on
Requires ground-truth temporal evidence annotations in the training data; shown on three Qwen-VL backbones (2.5-VL-7B, 3-VL-4B/8B) with ~6.5K training examples and five video QA be.
Sources
- Validated by: Average over five video benchmarks against SFT, standard OPSD and GRPO on three backbones: Video-OPSD beat standard OPSD in all 15 backbone-benchmark pairs (+1.3 to +2.1 average points) and matched or slightly beat GRPO at ~40% of its training time; ablations show teacher training-set accuracy 70.4% with evidence frames only vs 64.9% with 8 padding frames and 49.8% with 16 uniform frames (no gold answer), and final average falls from 51.0 to 50.6 to 50.4 as 4 then 8 non-evidence frames are added.. [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For 4-8B multimodal models on long-video QA, RL training on questions authored from short scene segments but answered over the whole video improves benchmark accuracy more than RL on questions authored from the whole video, because segment-authored questions target details a whole-video view overlooks; including the segment timestamp in the training instruction adds a further gain by removing referential ambiguity, and is not needed at inference.Reasoning about time in video · reviewed by AI
- For VLM-based judging of camera-conditioned world-model videos, splitting a long video pair into action-aligned chunks with structured visual evidence (source image, frame-grid overview, per-action first/last-frame panels) and voting chunk verdicts into video-level preferences agrees with human preferences better than judging the whole video pair at once, because short-lived action evidence is otherwise diluted by the full context.Judging another's work as well as a careful reviewer would · reviewed by AI
- For long-video QA under a fixed frame budget with a frozen 7B multimodal model, organizing frames into a question-agnostic hierarchical temporal tree and searching it coarse-to-fine (answer, then issue a search statement, then expand the relevant branch) raises accuracy over uniform sampling and over agent methods that score frames independently, with the gain largest on the longest-video benchmarks.Reasoning about time in video · reviewed by AI
- On long-video (10-150 minute) audio-visual reasoning questions, multiple-choice accuracy overstates grounding: a large open model can answer about 55% of items correctly while scoring 8.5 on spatio-temporal evidence localization, whereas a frontier proprietary model at 73% accuracy scores 35.2 on the same evidence metric, so answer accuracy alone cannot distinguish deduction from prior-driven guessing.Reasoning about time in video · reviewed by AI
- For 8B-14B generative reward models trained as judges, making the model generate input-adaptive evaluation criteria inside its own reasoning chain and rewarding the quality of those criteria with a stronger teacher model raises agreement with preference labels, and the gain scales with the teacher's capability rather than with the presence of the rubric step alone.Judging another's work as well as a careful reviewer would · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A controlled run where the teacher given the full video (or evidence frames plus padding) produces equal or better student benchmark averages than the evidence-only teacher under the same data, frame budget and objective.