Jazzilla on Nostr: nprofile1q…qzw4s nprofile1q…dsn3x You need a ton of RAM to do any building with ...
nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpq8suxpradq6zcl5r5nfhrkzd7lm3lt3fe0ndg5dcmu4gs07779whqwqzw4s (nprofile…zw4s) nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqlmdfpvdmnstf3ktacxfr4l2ser8hswxfcv5yla8an8de3tjedw7qadsn3x (nprofile…sn3x)
You need a ton of RAM to do any building with LLMs. I mean actually training and refining models for specific tasks. It's about double what it takes to make inferences.
That's why you need a unified memory architecture (not separate vram). It quickly gets ridiculous to even prototype.
That's a real threat to traditional PC Linux architectures, and leaves Apple with a monopoly.
I would like this to NOT continue, but most developers seem fine to just be sheepish consumers.
You need a ton of RAM to do any building with LLMs. I mean actually training and refining models for specific tasks. It's about double what it takes to make inferences.
That's why you need a unified memory architecture (not separate vram). It quickly gets ridiculous to even prototype.
That's a real threat to traditional PC Linux architectures, and leaves Apple with a monopoly.
I would like this to NOT continue, but most developers seem fine to just be sheepish consumers.