2nd GPU for PFlash?

[tomByrer]

I happen to have both RTX3080 (10GB) & 3090.
You mentioned that a drafter on 2nd GPU would prevent loading & unloading of the draft model.
How exactly can I do this?
& could other small LLMs be used, like Qwen Coder 1.5B or Qwen3.5 4b?

[davide221]

2nd GPU works in test_dflash today via --target-gpu 0 --draft-gpu 1 (or envs DFLASH_TARGET_GPU / DFLASH_DRAFT_GPU).
Draft model can easily be changed with qwen coder 1.5b or qwen3.5 4b, the target model need to be in the list of supported models by the https://huggingface.co/z-lab/