software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Added experimental MXFP4 optimizations

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-01-29 11:56:17 -08:00
parent b58ba7b19a
commit 53a8b45bcb
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@@ -144,11 +144,11 @@ Don't do it every time you rebuild, because it will slow down compilation times.
For periodic maintenance, I recommend using a filter: `docker builder prune --filter until=72h`
### 2026-01-26
### 2026-01-29
Added an experimental build option, optimized for DGX Spark and gpt-oss models by [Christopher Owen](https://github.com/christopherowen/spark-vllm-mxfp4-docker/blob/main/Dockerfile).
It is currently the fastest way to run GPT-OSS on DGX Spark, achieving 60 t/s on a single Spark. It's currently not working in a cluster configuration.
It is currently the fastest way to run GPT-OSS on DGX Spark, achieving 60 t/s on a single Spark and 75 t/s on dual Sparks.
To use this build, first build the container with `--exp-mxfp4` flag. I recommend using a separate label as it is currently not recommended to use this build for models other than gpt-oss:
@@ -183,7 +183,7 @@ Then, to run on a single Spark:
--max-num-batched-tokens 8192"
```
On a Dual Spark cluster (**CURRENTLY NOT WORKING**):
On a Dual Spark cluster:
```bash
./launch-cluster.sh -t vllm-node-mxfp4 exec vllm serve \