[Bugfix] use_fast failing to be propagated to Qwen2-VL image processor
#17838
Conversation
…ocessor Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
DarkLight1337
added
the
ready
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
…sor (vllm-project#17838) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: 汪志鹏 <wangzhipeng628@gmail.com>
…sor (vllm-project#17838) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>
…sor (vllm-project#17838) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
…sor (vllm-project#17838) Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: Yuqi Zhang <yuqizhang@google.com>
3 participants
As discussed in https://vllm-dev.slack.com/archives/C07QCGVDNUF/p1746687303802249?thread_ts=1745944078.067259&cid=C07QCGVDNUF
I decided not to add regression tests because the default behavior of
use_fast=None(which occurs whenuse_fastfails to be propagated) will change in transformers v4.52