Getting a 404 on the build the vllm use case example

Hi,
Who do I alert to an issue w a build example?
While attempting to follow the directions to build out a single instance w vllm and then using the Qwen 3.8 model… the sample script is missing. I get a 404 issue.

When looking at the other script it looks like its an example for a different model. (nemotron)
Here’s the example I am following:

Here’s the broken link…

https://raw.githubusercontent.com/NVIDIA/dgx-spark-playbooks/refs/heads/main/nvidia/playbook-vllm/assets/sync-vllm-single-spark.sh

Hi @msegel,

Thanks for flagging this, and this is the right place to report it.

sync-vllm-single-spark.sh was retired on Oct 8. One generic script was replaced by a separate script for each model and memory size. That’s why your link returns a 404 and the remaining scripts look like they’re for other models. The playbook page has been updated since then. If you still see the old link, do a hard refresh (or reopen the page) and go to Step 5.

For Qwen 3.8 on a single DGX Spark, choose the script that matches your Spark’s memory:

Your Spark Model Script
64GB Qwen3.8 27B (NVFP4) sync-vllm-spark-64gb-qwen38-27b.sh
128GB Qwen3.8 27B (NVFP4) sync-vllm-spark-128gb-qwen38-27b.sh
128GB Qwen3.8 Flash-Next sync-vllm-spark-128gb-qwen38-flash-next.sh

Some things to check before you run it:

  • Container image and model weights: these are also different for each script now. Use the docker pull and hf download lines from the table in Step 4 for the row that matches your script, not the ones from an older copy of the page.

  • Qwen3.8 27B: this version also needs the speculative-decoding draft model:

    hf download z-lab/Qwen3.8-27B-DFlash2
    
  • Logs: the container name now includes the model, for example docker logs --tail 50 --follow vllm-spark-64gb-qwen38-27b.

If the page you’re viewing still points at sync-vllm-single-spark.sh after a refresh, send me the exact URL. I’ll get it fixed. Let us know how it goes!