GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
ngqtrung
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
ngqtrung
Model Tree
Qwen/Qwen3-VL-8B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Accuracy in percent, evaluated with lmms-eval at 100 frames per video under one setting for every
row. "In-domain" is the 1,000-question held-out Video-HopChain split.
| Model | Video-MME | Perception-Comp | Video-MMMU | Video-Holmes | VCRBench | MMR-V | LongVideo-Reason | VRBench | mean | in-domain |
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-VL-8B-Instruct | 64.0 | 28.1 | 63.5 | 40.7 | 32.0 | 42.7 | 72.6 | 74.7 | 52.3 | 13.4 |
| + standard RL | 65.8 | 34.3 | 63.0 | 47.4 | 35.9 | 43.4 | 76.1 | 77.6 | 55.4 | 13.4 |
| + standard RL + SWE | 67.9 | 34.4 | 62.5 | 48.5 | 35.1 | 46.4 | 78.8 | 79.5 | 56.6 | 16.4 |
| + standard RL + Video-HopChain | 68.6 | 36.2 | 64.7 | 48.6 | 42.1 | 45.9 | 77.8 | 79.4 | 57.9 | 21.2 |
| this model | 69.2 | 37.4 | 67.5 | 48.6 | 44.9 | 46.6 | 78.9 | 81.0 | 59.3 | 23.2 |
VCRBench is reported on its multiple-choice subset.
python
from transformers import AutoProcessor, Qwen3VLForConditionalGenerationmodel = Qwen3VLForConditionalGeneration.from_pretrained("ngqtrung/video-hopchain-8b", dtype="auto", device_map="auto")processor = AutoProcessor.from_pretrained("ngqtrung/video-hopchain-8b")
The weights are BF16. The model was trained to reason inside <think>...</think> and to put the final
answer in <answer>\boxed{...}</answer>, so use the same system prompt as training. It is in the
dataset rows and in the repository.
Second-Wave Exploration addresses advantage collapse in GRPO. With 8 rollouts per question, SWE draws
the first 4 normally. If those 4 are all correct or all incorrect, the group carries no reward
variance and no gradient, so SWE draws the last 4 with the policy's top token masked wherever its
probability exceeds tau = 0.95, inside the reasoning span only. The masked positions are dropped
from the loss while all 8 rollouts enter the group advantage, so the method spends no extra rollouts.
| Base model | Qwen3-VL-8B-Instruct |
| Stage 1 | GRPO on a 105,993-row general video QA mixture, 24 frames |
| Stage 2 | GRPO + SWE on Video-HopChain, 22,550 rows, 140 frames |
| Rollouts per question | 8, first wave 4 |
| Mask threshold | 0.95, reasoning span only |
| Reward | exact match on the integer sum, plus a format term |
| Hardware | 4 nodes of 8 H100 80GB, asynchronous verl trainer |
Appendix C of the paper gives every hyperparameter.
Evaluated at one model scale only. The dataset is synthetic and verified against captions rather than against the frames, so a caption error can reach a label. The paper reports both limitations, together with the exploration settings that we did not ablate.
bibtex
@article{videohopchain2026,title = {Video-HopChain: Multi-Hop Questions and Second-Wave Exploration for Video Reasoning Models},author = {Nguyen, Quang Trung and Dong, Yuhao and Sun, Shuo and Liu, Shuai and Tian, Shulin and Yap, Kim-Hui and Liu, Ziwei},journal = {arXiv preprint},year = {2026}}