@AnonomousWolf@lemmy.world I guess it would be more fairer if we were to mention DeepSeek as being "not bad for the environment". From all LLMs, seems like it's the one who did their homework and tried to optimize things the best they could.
Western LLMs had/have no reason to optimize, because "Moar Nvidia Chips" have been their motto, and Venture Capital corps have been injecting obscene amounts of money into Nvidia chips, so Western LLMs are bad for the environment, all the way from establishing new power-hungry data centers to training and inference...
But DeepSeek needed way less computing and it can run (Qwen-distilled versions) even in a solar-powered Raspberry Pi with some creativity... it can run in most smartphones like if it were another gaming app. Their training also needed less computing, as far as we know.
@AnonomousWolf@lemmy.world I guess it would be more fairer if we were to mention DeepSeek as being "not bad for the environment". From all LLMs, seems like it's the one who did their homework and tried to optimize things the best they could.
Western LLMs had/have no reason to optimize, because "Moar Nvidia Chips" have been their motto, and Venture Capital corps have been injecting obscene amounts of money into Nvidia chips, so Western LLMs are bad for the environment, all the way from establishing new power-hungry data centers to training and inference...
But DeepSeek needed way less computing and it can run (Qwen-distilled versions) even in a solar-powered Raspberry Pi with some creativity... it can run in most smartphones like if it were another gaming app. Their training also needed less computing, as far as we know.