GLM-4.7-flash update transformer==5.0 cannot import name 'AutoModelForVision2Seq' from 'transformers'

[tcxia]

System Info

verl 0.7.0

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

GLM-4.7-flash update transformer==5.0 cannot import name 'AutoModelForVision2Seq' from 'transformers'

Expected behavior

correct bug

[beverly1974-justice0]

try
pip install transformers==4.48

[szhou0202]

i have the same issue.
transformers 5.1.0

[ruanhao566]

transformers 5.2.0 for qwen3.5 same problem

[Kirrito-k423]

中文：这个问题看起来已经在后续版本中修复。#5007 报告的是 verl 0.7.0 在 transformers 5.x 下直接导入 AutoModelForVision2Seq 失败；v0.7.0 的 verl/utils/model.py 和 verl/model_merger/base_model_merger.py 确实包含直接导入路径。后续合入的 PR #5445（https://github.com/verl-project/verl/pull/5445）加入了 get_auto_model_for_vision2seq() 兼容 resolver，在 verl/utils/transformers_compat.py 中优先使用 AutoModelForImageTextToText，并在需要时回退到 AutoModelForVision2Seq。该合并提交已包含在 v0.7.1 及之后版本中。建议升级到 v0.7.1 或更新版本；如果升级后仍复现，请带上完整 traceback、verl 版本、transformers 版本和最小复现脚本重新开启/补充问题。

English: This appears to have been fixed in later releases. #5007 reports that verl 0.7.0 fails when importing AutoModelForVision2Seq with transformers 5.x, and v0.7.0 did directly import that class in verl/utils/model.py and verl/model_merger/base_model_merger.py. PR #5445 (#5445) added the get_auto_model_for_vision2seq() compatibility resolver in verl/utils/transformers_compat.py, preferring AutoModelForImageTextToText and falling back to AutoModelForVision2Seq when needed. That merged commit is included in v0.7.1 and later releases. Please upgrade to v0.7.1 or newer; if the problem still reproduces after upgrading, please provide the full traceback, verl version, transformers version, and a minimal reproduction script.

This response was generated by a scheduled Codex GPT5.5 issue processing task.