howtospark
models / qwen3-coder-next

80B params~3.9B activebfloat16MoE 512×10GQA 16:2262K ctxapache-2.0
Sparks
~171 GB
~131 GB
~91 GB
~51 GB
~34 GB
~33 GB
~32 GB
110 usable
REAP dialno pruning · 512 experts
Spec decode
Context33K / 262KKV
On the Sparks — 1× · 2-bit experts + NVFP4 dense · 33K ctx · FP16 KV~829K ctx fits usable
spark
78 GB free
Expert planes19.3* GB2-bit
110 usable
32* / 110 GB
29% of usable
Expert planesDense weightsKV cacheActivations + graphsOS reserve

* segment sizes marked with an asterisk are estimates pending a measured run

Estimated per-node footprint, weights + KV sharded across the Spark (tensor parallel). Overhead ~8 GB/node; KV at FP16.

layer map
48 layersMoE FFN
in
out

one MoE layer · 512 experts, 10 fire per token · showing 256

each expert ≈ 3M params

resident: 48 MoE layers × 512 experts × 3M 77.3B

active/token: 48 × 10 × 3M 1.5B experts

+ 2.4B always-on dense (attention, embeddings, shared stack) = 3.9B active

Recipes (all recipes)

Worked deployments that run this model on DGX Spark — each shows how it fills each node's unified memory.

Optimize
Verdict

80B params, ~159 GB at bfloat16 — doesn't fit natively, but ~80 GB at 8-bit does.

Make it fit

Get the weights (and headroom for KV) inside the 128 GB pool — the hard gate.

Quantize to 8-bit

At 159 GB native it doesn't fit the 110 GB budget; 8-bit weight-only quantization gets it to ~80 GB.

159 GB → ~80 GB

Lossless compression

BF16 weights entropy-code ~30% smaller bit-exact — smaller downloads and less bandwidth per token, no quality question at all.

Make it fast

Once it fits: fewer bytes per token and fewer decode steps against 273 GB/s.

Add a speculative decoder

Decode is bandwidth-bound at ~70 tok/s ceiling; a trained EAGLE-style draft multiplies tokens per weight-read.

Prune experts for speed

Fewer experts ⇒ smaller working set ⇒ better cache/page behavior, even when it already fits.

Tame the KV cache

At 262144 tokens the KV cache alone is ~26 GB at FP16; quantize or evict.

Make it yours

Change what the model does — adapt, edit, or steer it — independent of size.

Fine-tune it

LoRA/QLoRA adapts behavior on one Spark without touching the base weights.

Qwen3-Coder-Next — How to Spark · How To Spark