// RUN — FITS

Can the Apple M2 Ultra 192GB run AWD-LSTM-DRILL + dynamic evaluation† (PTB)?

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

AWD-LSTM-DRILL + dynamic evaluation† (PTB) on the Apple M2 Ultra 192GB

QuantFits?Max contextSpeedBeyond context
FP16 yes 2613k 11333.3-13666.7 t/s ~141.7-170.8 (slow)
Q8 yes 2613k 22666.7-27333.3 t/s ~283.3-341.7 (slow)
Q4 yes 2614k 40476.2-48809.5 t/s ~506-610.1 (slow)

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

FAQ

Can the Apple M2 Ultra 192GB run AWD-LSTM-DRILL + dynamic evaluation† (PTB)?

Yes. At Q4 it generates ~40476.2-48809.5 tok/s and fits up to 2614k context; beyond that it spills to SSD swap and slows to ~506-610.1 tok/s.

How much context fits?

At Q4, up to about 2614k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.