* 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.
one MoE layer · 64 experts, 6+2 fire per token
each expert ≈ 3M params
resident: 11 MoE layers × 64 experts × 3M ≈ 2.4B
active/token: 11 × 6 × 3M ≈ 227M experts
+ 914M always-on dense (attention, embeddings, shared stack) = 1.1B active
Optimize
3B params, ~7 GB at bfloat16 — fits on one Spark as-is.
Make it fit
Get the weights (and headroom for KV) inside the 128 GB pool — the hard gate.
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 ~119.6 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.
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.
Edit or steer it
One-shot weight math — no gradients, no corpus.