| license: apache-2.0 | |
| library_name: transformers | |
| pipeline_tag: video-text-to-text | |
| tags: | |
| - qwen2.5-vl | |
| - video-question-answering | |
| - resadapt | |
| # ResAdapt research checkpoint | |
| This repository preserves the Resadapt checkpoint and its loading files. | |
| - Repository root: Qwen2.5-VL checkpoint. | |
| - `pred/`: trained SmolVLM predictor checkpoint. | |
| - `predictor_smolv3_init/`: initialization weights for the predictor. | |
| Six machine-specific paths in the predictor example code use portable placeholders. Replace those placeholders with your own model and media paths. Weight values and scientific identifiers are unchanged; this publication adds no inference or benchmark result. | |
| [ARTIFACTS.json](ARTIFACTS.json) records the source revision, per-file hashes, and the limited edits. Existing source terms remain applicable. | |