// RUN — FITS

Can the NVIDIA DGX Spark run AWD-LSTM + MoS + Partial Shuffled?

Yes — here's how fast and up to what context, with real measured numbers.

AWD-LSTM + MoS + Partial Shuffled on the NVIDIA DGX Spark

QuantFits?Max contextSpeedBeyond context
FP16 yes 1734k 2847-3393 t/s ~104.3-124.3 (slow)
Q8 yes 1734k 5694-6786 t/s ~208.6-248.6 (slow)
Q4 yes 1735k 10167.9-12117.9 t/s ~372.4-443.9 (slow)

Estimate (memory-bound). Beyond "max context" the KV cache spills to SSD swap and speed drops to the "beyond" figure.

→ All models on the NVIDIA DGX Spark → AWD-LSTM + MoS + Partial Shuffled specs → Other hardware

FAQ

Can the NVIDIA DGX Spark run AWD-LSTM + MoS + Partial Shuffled?▶

Yes. At Q4 it generates ~10167.9-12117.9 tok/s and fits up to 1735k context; beyond that it spills to SSD swap and slows to ~372.4-443.9 tok/s.

How much context fits?▶

At Q4, up to about 1735k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.