2× DGX Spark
26tok/sEveryday
1 request, prose prompt, no speculative decoding Source
Intelligenceno independent value
Local AI · Recipe · 1× DGX Spark
Ling-3.0-flash-VL 124B NVFP4 with vLLM on 1× DGX Spark
Ling-3.0-flash-VL 124B NVFP4 with vLLM on 1× DGX Spark: 43.5 tok/s according to github.com (dataset as of Nov 3, 2026).
by r0b0tlab
Engine
- Engine
- vLLM vllm-ling-v3 @ e2e5751
- Quantization
- NVFP4 + FP8 mixed (ModelOpt 0.48) + FP8 KV
- Model family
- Ling-3.0-flash-VL
- Context
- 131,072
- Parameters
- 124B total / 5.1B active MoE
- Creator
- r0b0tlab
- GitHub stars
- 3
- Repo updated
- Sep 9, 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
Prefill by context
Time to first token
Shorter is better.
What you need
- Hardware
- 1 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightsr0b0tlab/Ling-3.0-flash-VL-NVFP4-MP MIT
- BaseinclusionAI/Ling-3.0-flash-VL MIT
- Engine
- vLLM vllm-ling-v3 @ e2e5751
- Context
- 131,072 tokens
Notes
What matters before you rebuild it.
- Custom kernel required
Sources
Related recipes
1× DGX Spark
71.8tok/sEveryday
1 request, prose prompt (sampled, thinking off), with LiLiCorr (per recipe) Source
Intelligenceno independent value
- ≤ 3 bit: quantization may cost quality
- Custom kernel required
- License unclear
1× DGX Spark
60.3tok/sEveryday
1 request, prose prompt (essay), context 2k, no drafter (per recipe) Source
Intelligenceno independent value
- Custom kernel required
1× DGX Spark
59.4tok/sEveryday
1 request, chat prompt (20 prompts, thinking off), no drafting (default profile) Source
Intelligenceno independent value
- Custom kernel required
1× DGX Spark
58.3tok/sEveryday
1 request, prose prompt, with DFlash2 (per recipe) Source
Intelligence34with thinkingno thinking 20
- Custom kernel required
1× DGX Spark
55.6tok/sEveryday
1 request, prose prompt, with MTP (per recipe) Source
Intelligence40with thinking
- Custom kernel required
- License unclear
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.