[Eagle-3 Training] Cloud I use this code for Qwen3-235B draft model training. I can`t find any config about target model`s parallel setting like tp\pp\ep ...

[TheUgliestOfWuli]

[liusong1222]

Thanks for the question.

At the moment, our HF backend does not support configuring target model parallelism (e.g. TP / PP / EP) for Eagle-3 training.

We are actively working on a vLLM-based backend, which will provide proper support for target model parallel settings, including tensor / pipeline / expert parallelism. This is still under development and will be updated in a future release.