| MMProLong |
| https://github.com/zhaowei-wang-nlp/MMProLong |
|
|
| This repository is a research fork of VeOmni: |
| https://github.com/ByteDance-Seed/VeOmni |
|
|
| VeOmni copyright 2025 ByteDance Ltd. and/or its affiliates. |
| The inherited Apache License 2.0 is provided in LICENSE. Existing copyright, |
| license, and third-party attribution notices in source files are retained. |
| Individual files may carry additional third-party notices; this notice does |
| not replace or relicense them. Model weights and datasets have their own terms. |
|
|
| MMProLong modifications include long-document VQA data synthesis, Qwen-VL |
| data processing and training integration, long-context mixtures, experiment |
| configurations, checkpoint utilities, and release documentation. The veomni |
| Python namespace is retained for compatibility with the upstream framework. |
| This fork is maintained separately from upstream VeOmni. |
|
|
