2× DGX Spark
62.9tok/sPeak
1 request, JSON prompt Source
Intelligence25with thinking
Local AI · Recipe · 2× DGX Spark
MiMo-V2.5 310B NVFP4 with vLLM on 2× DGX Spark (160k context, with MTP, tonyd2wild)
MiMo-V2.5 310B NVFP4 with vLLM on 2× DGX Spark (160k context, with MTP, tonyd2wild): 38.3 tok/s according to github.com (dataset as of Nov 3, 2026).
by tonyd2wild
Engine
- Engine
- vLLM 0.21.1rc1.dev85+gd87ee1893
- Quantization
- NVFP4 experts + MXFP8 dense, fp8_e4m3 KV
- Model family
- MiMo-V2.5
- Context
- 163,840
- Parameters
- 310B MoE / ~15B active (lt. README)
- Creator
- tonyd2wild
- GitHub stars
- 9
- Repo updated
- Jul 10, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
Decode · 1 request
What you need
- Hardware
- 2 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightslukealonso/MiMo-V2.5-NVFP4 MIT
- Engine
- vLLM 0.21.1rc1.dev85+gd87ee1893
- Context
- 163,840 tokens
Notes
What matters before you rebuild it.
- Experimental
- Custom kernel required
- Superseded
Sources
Related recipes
3× DGX Spark
38.8tok/sPeak
1 request, best value reported by source Source
Intelligence25with thinking
- Superseded
3× DGX Spark
38.2tok/sPeak
1 request, mixed prompt set Source
Intelligence25with thinking
- Experimental
- Custom kernel required
2× DGX Spark
37.8tok/sPeak
1 request, JSON prompt Source
Intelligence25with thinking
- Custom kernel required
2× DGX Spark
31.9tok/sPeak
1 request, best value reported by source Source
Intelligence25with thinking
- Custom kernel required
2× DGX Spark
30tok/sPeak
1 request, best value reported by source Source
Intelligence25with thinking
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.