minus-squareAcceptable_Can5509@alien.topBtoLocalLLaMA•40x or more speedup by selecting important neuronslinkfedilinkEnglisharrow-up1·1 year agoBasically gpt 4 turbo linkfedilink
minus-squareAcceptable_Can5509@alien.topBtoLocalLLaMA•bellman-7b - a Swedish llama2 finetunelinkfedilinkEnglisharrow-up1·1 year agoCan you share the colab so others can look at how it was done? linkfedilink
minus-squareAcceptable_Can5509@alien.topBtoLocalLLaMA•Clearing up confusion: GPT 3.5-Turbo may not be 20b after alllinkfedilinkEnglisharrow-up1·1 year agoProbably heavily quantized and uses a smaller gpt-3 model. linkfedilink
minus-squareAcceptable_Can5509@alien.topBtoLocalLLaMA•I am going to buy H100s. There are too many options.linkfedilinkEnglisharrow-up1·1 year agoWait, whos money is it? Can’t you just rent as well? linkfedilink
Acceptable_Can5509@alien.topB to LocalLLaMAEnglish · 1 year agoLlama-2 7b Unquantized Transformers using 26.8GB of Vram.plus-squaremessage-squaremessage-square0fedilinkarrow-up11arrow-down10
arrow-up11arrow-down1message-squareLlama-2 7b Unquantized Transformers using 26.8GB of Vram.plus-squareAcceptable_Can5509@alien.topB to LocalLLaMAEnglish · 1 year agomessage-square0fedilink
Basically gpt 4 turbo