// RUN — FITS

Can the Intel Gaudi 3 run AWD-LSTM-DRILL + dynamic evaluation† (WT2)?

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

AWD-LSTM-DRILL + dynamic evaluation† (WT2) on the Intel Gaudi 3

QuantFits?Max contextSpeedBeyond context
FP16 yes 1773k 31558.8-39176.5 t/s ~341.2-423.5 (slow)
Q8 yes 1773k 63117.6-78352.9 t/s ~682.4-847.1 (slow)
Q4 yes 1774k 112710.1-139916 t/s ~1218.5-1512.6 (slow)

Estimate (memory-bound). Beyond "max context" the KV cache spills to system RAM offload (depends on your system RAM) and speed drops to the "beyond" figure.

FAQ

Can the Intel Gaudi 3 run AWD-LSTM-DRILL + dynamic evaluation† (WT2)?

Yes. At Q4 it generates ~112710.1-139916 tok/s and fits up to 1774k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~1218.5-1512.6 tok/s.

How much context fits?

At Q4, up to about 1774k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.