Plan for large-size Qwen3 or Qwen3.5 models?

[feichendingxiangyong]

May I ask whether there is a plan to support the large-size Qwen3 or Qwen3.5 models(like Qwen3-235B )? Thanks.

[jianc99]

We are working on this! We will release the DFlash draft models for Qwen3-Coder-Next (80B-A3B), gpt-oss-120b, GLM-4.7-Flash and Qwen3.5-35B-A3B soon (probably this or next week), and then we will focus on the large Qwen3.5 models.

[INQTR]

Any plans for qwen3.5-27b dence model, seems this one super smart for it's size?

[chenxiaodong2002]

We are working on this! We will release the DFlash draft models for Qwen3-Coder-Next (80B-A3B), gpt-oss-120b, GLM-4.7-Flash and Qwen3.5-35B-A3B soon (probably this or next week), and then we will focus on the large Qwen3.5 models.

Hi @jianc99, could you please share if the training code will be released next week? Thanks!

[tha80]

Please prioritize Qwen3-27B over Qwen3-35B-A3B because the 27B variant is way more useful in RL but also way slower, so that speeding it up with be super beneficial.