update-qwen-implementation
#14
by
jupyterjazz
- opened
Qwen2.5 VL implementation was changed heavily in the last few transformers versions, and since we use adapted implementation and rely on HF internals, this introduced couple of errors during inference. So I'm updating everything - implementation, weights, configs - to work with the new changes. The minimum transformers version required for je-v4 becomes 4.52.0
jupyterjazz
changed pull request status to
open
jupyterjazz
changed pull request status to
merged