I had a chance to play with lukealonso/MiniMax-M2.7-NVFP4 and wanted to share my very first results (Dual Node Setup - 2x Asus Ascent GX10). I let vLLM calculate context and was able to get 196608 – not too terrible. Benchmarks: | model | test | t/s | peak t/s | …

The 2 Sparks don’t shut down, they appear frozen. On one window where I was looking at VLLM server, after the initial ctrl-c to stop the server, if I hit ctrl-c and come back 20 minutes later I see VLLM’s message it received a SIGTERM signal. I have to manually power them down. (Hold the power butt…

[image] Marsy: I was trying to get Nvidia/MiniMax-M2.5-NVFP4 to run using vllm-node on two sparks. It would freeze mid-startup and become unresponsive. I got it to run once using vllm-node-tf5 like the OP but when I shut vLLM down it froze again. On a second run it froze mid-startup when I incre…

Minimax M2.7 awq recipe couldn’t run on my spark cluster: It stucked at this point: 2026-05-14 01:26:28,555 INFO usage_lib.py:448 -- Usage stats collection is disabled. 2026-05-14 01:26:28,557 INFO scripts.py:940 -- Local node IP: 169.254.54.241 2026-05-14 01:26:31,783 SUCC scripts.py:979 -- -----…

nvidia version better than AWQ that is all I ever ran is there a better recipe and model could you help me ?

It’s just odd, since I normally run with .92 and I believe 2 Sparks are enough for M2.5 NVFP4 at 128K context. I mean, it doesn’t explain the symptoms. I’m running TF5 and 0.80, will update when it’s down loading, llama-benchy, etc… :) Update: [image] Does look like it’s a memory issue. I will …

Please check my post with recipe: [image] Best 2026 model for agentic work on a 2-node Spark cluster? DGX Spark / GB10 here is my recipe: VLLM_SPARK_EXTRA_DOCKER_ARGS=“-v $HOME/models:/models” ./launch-cluster.sh --no-ray exec vllm serve /models/MiniMax-M2…

Thanks but no luck… I also tried nvidia/minimax-m2.7-nvfp4, it stucked on the same place.

Hey, I can see a few issues in your log: Fix 1: Download the model locally first Don’t load it directly from HuggingFace — download it to disk on both nodes first: bash cd ~/spark-vllm-docker ./hf-download.sh cyankiwi/MiniMax-M2.7-AWQ-4bit -c --copy-parallel Fix 2: Use this recipe instead of lau…

For the out of memory hangs, it appeared that there was too much swap configured and what was happening was the system was thrashing heavily and so appeared to be hung. I ended up just removing the swap configuration from /etc/fstab and deleting /swap.img. Now if Ia process exceeds the amount of m…

MiniMax M2.7 NFVP4 Recipe & Benchmarks

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

karol.spark May 14, 2026, 11:04am 102

Please check my post with recipe:

Topic		Replies	Views
MiniMax M2.5 released (not available on HuggingFace as of now) -- is DGX Spark ready? DGX Spark / GB10	92	6799	April 12, 2026
MiniMax M3 : NVFP4 for Quad DGX Spark DGX Spark / GB10 agentic-ai , deepseek	116	8060	June 25, 2026
MiniMax M3 NVFP4 and NVFP4 REAP 50 for 4x & 2x DGX Sparks DGX Spark / GB10 Projects	53	4006	July 2, 2026
MiniMax 2.5 REAP - NVFP4 on single DGX Spark DGX Spark / GB10	25	3346	April 1, 2026
MiniMax-M3-W4A16-GPTQ 2xGB10 Deployment (36 t/s) (fp8, nvfp4, KVarN, EAGLE-3) DGX Spark / GB10 agentic-ai	24	2505	July 22, 2026
Minimax3 on 2 nodes decode ~10.7 tok/s, 4bits DGX Spark / GB10 llama	26	1489	June 21, 2026
Can someone with 2 Sparks benchmark NVFP4 MiniMax M2.1 quant? DGX Spark / GB10	24	1565	January 15, 2026
Help: Running NVFP4 model on 2x DGX Spark with vLLM + Ray (multi-node) DGX Spark / GB10 mistral-large	18	2798	December 25, 2025
MiniMax-M3 (428B MoE + vision) at ~14–15 tok/s on 2× DGX Spark — EAGLE3 speculative decoding is the unlock DGX Spark / GB10 Projects llama , deepseek	2	648	July 4, 2026
RedHatAI/Qwen3.5-122B-A10B-NVFP4 seems to be the best option for a single Spark DGX Spark / GB10 Projects llm	75	6878	May 4, 2026

MiniMax M2.7 NFVP4 Recipe & Benchmarks

Related topics