Hi everyone 👋 Over the past weeks, I’ve been working with @eugr on adding structured, reproducible “benchmark recipes” to our community Docker runtime for LLMs . @eugr also added new export formats to Llama-Benchy . Those are foundations for a better knowledge sharing experience, but we still needed …

Added all those suggestions to the website. Thanks

@giraudremi92 Thanks for posting this recipe to Spark Arena — I was just about to do the same, so you definitely saved me some time! Hard to believe how much progress has been made in the six weeks since I wrote this post. Great to see the performance metrics lining up as expected. Huge thanks to …

Thanks @griffith.mark For those in the day0 trend : Here is the recipe : (needs to update in the container transformers version) I just added a mod for run.sh to upload transformers I will publish it with the benchmark results when its finished recipe_version: “1” name: Qwen3.5-35B-A3B descript…

Hi Rémi, Thanks for sharing it here as well. Whenever you have time. Please open a PR on GitHub - eugr/spark-vllm-docker: Docker configuration for running VLLM on dual DGX Sparks . This pip install --upgrade transformers mod should be unnecessary whenever we have our versioned images.

PR is done. QWEN 3.5 models : create new mod for update transformers & qwen 3.5 recipe (#64) main ← giraudremi92:main opened 07:10PM - 24 Feb 26 UTC (UTC) […

This shouldn’t be needed. Just build the image with --tf5 (or --pre-tf which is an older name for the flag). You can add this to your recipe: # Container image to use container: vllm-node-tf5 # Build arguments for build-and-copy.sh # tf5 = transformers 5.0 (required for GLM-4.7) build_args: - --…

Hmmm… When was the last time you rebuilt your vLLM container? With today build (no transformers v5, just the bundled version): ./build-and-copy.sh -t vllm-node-20260224 --rebuild-vllm ./launch-cluster.sh -t vllm-node-20260224 --solo exec vllm serve Qwen/Qwen3.5-35B-A3B --gpu-memory-utilization 0.7…

Thanks @eugr i was indeed on a previus build & version. I will do iit after my becnhmark

model test t/s peak t/s ttfr (ms) est_ppt (ms) e2e_ttft (ms) Qwen/Qwen3.5-35B-A3B pp2048 3253.80 ± 370.45 648.44 ± 79.06 638.73 ± 79.06 648.61 ± 79.00 Qwen/Qwen3.5-35B-A3B tg32 30.81 ± 0.04 31.00 ± 0.00 llama-benchy (0.3.1) date: 2026-02-24 11:22:06 | latency mode: api (probably c…

Introducing the Spark Arena

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10

raphael.amorim February 24, 2026, 6:54pm 65

Working on integrating this on Spark Arena. @christopher_owen did you push a PR?

Topic		Replies	Views
New tool: llama-benchy - llama-bench style benchmarking for ANY LLM backend (vLLM, SGLang, llama.cpp, etc.) DGX Spark / GB10 Projects llama	17	2618	April 21, 2026
Fastest Qwen 3.5 122B Int4 recipe on DGX Spark tested and published on Spark-Arena DGX Spark / GB10 llama	59	2671	June 3, 2026
HOW-TO: Run Qwen3-Coder-Next on Spark DGX Spark / GB10 llama	92	10044	March 24, 2026
(sparkrun) Qwen3.5 GGUF Benchmarks over llama.cpp RPC DGX Spark / GB10 Projects llama	3	741	March 11, 2026
Qwen/Qwen3.6-35B-A3B (and FP8) has landed DGX Spark / GB10 agentic-ai	308	26307	June 9, 2026
DGX Spark + Qwen3-Next-80B: Proven Performance, But Missing Clear Path to NIM, TensorRT-LLM & Web UIs DGX Spark / GB10 cuda , nim , llama	16	4609	March 6, 2026
Qwen/Qwen3.5-122B-A10B - Alibaba/Qwen thought about us... :-D DGX Spark / GB10	340	16654	March 24, 2026
Dgx spark benchmark performance DGX Spark / GB10	16	2185	December 21, 2025
Does Qwen3.5-35B-A3B on GB10 leave a lot of performance on the table? DGX Spark / GB10 agentic-ai	40	5919	March 16, 2026
RedHatAI/Qwen3.5-122B-A10B-NVFP4 seems to be the best option for a single Spark DGX Spark / GB10 Projects llm	75	6191	May 4, 2026

Introducing the Spark Arena

Related topics