1 article
Mistralâs latest open-source speech model squeezes into 128MB of RAMâsmall enough for a but untested in noisy subway tunnels.