# Training a personal draft model for Qwen **URL:** \ **Category:** DGX Spark / GB10\ **Created:** [July 30, 2026, 3:58pm UTC](https://forums.developer.nvidia.com/t/training-a-personal-draft-model-for-qwen/378611 "2026-07-30T15:58:18Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![colizu2020](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@colizu2020](https://forums.developer.nvidia.com/u/colizu2020)\ **Post date:** [July 30, 2026, 3:58pm UTC](https://forums.developer.nvidia.com/t/training-a-personal-draft-model-for-qwen/378611/1 "2026-07-30T15:58:18Z")
I currently am using Qwen3.6 35B A3B on vLLM with MTP (num\_speculative\_tokens=3) for everyday coding and tasks. My workflow right now consists of using up to 2-3 subagents in the Claude Code CLI on my DGX Spark. The performance is reasonable (~50 tok/sec sustained decode) but I want to see if I can improve the speculative decoding accuracy any further, and have a plan written up to try using DSpark from the vLLM speculators GitHub repo as a base for training my own draft model. Have any of you tried doing something similar before? Is it worth the effort, or should I wait until the next Qwen release which may or may not support the new speculative methods natively? Thanks! ---
**Author:** ![alexander.kachur](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/alexander.kachur/32/523564_2.png) [@alexander.kachur](https://forums.developer.nvidia.com/u/alexander.kachur)\ **Post date:** [July 30, 2026, 4:52pm UTC](https://forums.developer.nvidia.com/t/training-a-personal-draft-model-for-qwen/378611/2 "2026-07-30T16:52:09Z")
Just try already existing DFlash drafter: [z-lab/Qwen3.6-35B-A3B-DFlash · Hugging Face](https://huggingface.co/z-lab/Qwen3.6-35B-A3B-DFlash) DSpark is only marginally better There’s no free lunch here - training will cost you couple of hundreds with uncertain benefits > [@colizu2020](#): > > Qwen3.6 35B A3B