howtospark
models / Qwen/Qwen3.6-27B

Qwen/Qwen3.6-27B
28B paramsbfloat16GQA 24:4262K ctxmultimodalapache-2.0
Sparks
~72 GB
~58 GB
~44 GB
~30 GB
~25 GB
110 usable
Spec decode
Context33K / 262KKV
On the Sparks — 1× · 1-bit GGUF (dynamic) · 33K ctx · FP16 KV~359K ctx fits usable
spark
85 GB free
110 usable
25* / 110 GB
22% of usable
Dense 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
64 layerslinear attention×48full attention×16
in
out
all 64 layers fire per token — 27.8B active
Optimize
Verdict

28B params, ~56 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 ~4.9 tok/s ceiling; a trained EAGLE-style draft multiplies tokens per weight-read.

Tame the KV cache

At 262144 tokens the KV cache alone is ~69 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.