VideoJSON LoRA for Qwen-3.5-Flash

LoRA adapter of Qwen/Qwen3.5-35B-A3B fine-tuned on 64 VideoJSON trajectories (352 training examples) from MediaStorm-Livestream, from VideoJSON: A Minimal Harness Beats Purpose-Built Memory for Ultra-Long Video Understanding (paper, Section 6).

The adapter weights (adapter_model.safetensors) are not uploaded yet.

adapter_config.json has base_model_name_or_path set to the local path /model used in training; load the adapter on top of Qwen/Qwen3.5-35B-A3B.

Downloads last month
11
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for weathon/videojson_release_lora

Adapter
(48)
this model