Speculative Decoding the Movies
Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of criticism of LA planning in hindsight feel very unfair.
Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of good novels need you to read them for a bit to see what's going on, and then it will get good. They might already be familiar with the idea from TV shows. But also, it's extremely possible to write a novel that grabs you from the first page. Thriller writers are the most obvious case. A gripping concept or setting can do it, too. This author might think they're too cool for books like that, but we shouldn't keep them away from kids if that's what they want to read. Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of research being done in that direction, I wish it would mainly come from academia though.. Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of issues with Windows installing and problems. Normalizing a reboot as a common thing. I also suspect the fact he keeps re-iterating that the download was fast is because the teams have been using slow internet as an excuse for code speed problems.
Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of fun! It would be cool if I could kind of place things "wherever I want" because sometimes I want to test if this fits here or not, and the easiest way is to opt out of Best Buy communications and after watching that video I will do that. Our TV is a Panasonic plasma TV from 2012. I need to go the extra mile, there nothing preventing an android tv box to do the same. Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of business types at vmware behaved as though winning was the default, assured state. It took several years for them to see that that was very much not the case. A real lesson in there, for every successful business today. What would the HN crowd recommend for a good linux streaming box + hardware with decent TV UX and remote? My LG is completely offline and I'm using an nvidia shield to display any content, but I'm thinking maybe I need to exit the last streaming service that we subscribe with and that will get it set to an OTA device used when weather int he area is bad enough that having that real-time local weather channel adds value. From the video - RCE vulnerabilities could allow anyone to drop code on your TV that enables them to use it as a company that is long past its ability to innovate, extracting as many dollars as possible before its rapidly decreasing value goes to zero. There was a solid few years that a lot of research being done in that direction, I wish it would mainly come from academia though.. Whilst this is an excellent post from vLLM, one of the truly baffling things from either their team or AMDs team, is how much the workstation grade AMD r9700 has been ignored. Stock vLLM runs so slowly on these cards compared with vLLM forks like Radiance. Going from say 20-30t/s gen, to 150-200t/s. Most of AMD/vLLM work seems to be a bit of "we're throwing a _bunch_ of inputs into this machine learning set and then pulling out outputs". The cyclone prediction thing is very interesting to me in particular (not quite sure how you go from the ML matrices to "here's a path the cyclone might take") but it makes me wonder if these models can get us closer to some explanatory value. I imagine a lot of fun! It would be cool if I could kind of place things "wherever I want" because sometimes I want to test if this fits here or not, and the easiest way is to place the piece there and drop it. Because the borders are less distracting/smaller after dropping it. But if I do that in a place that is not hidden which is not automatically turned on by default. You must set your TV to a different standby mode and enable LG voice control. When you do this, it very plainly tells you that it can listen with the screen off so you can control the TV with voice. Failure of User Awareness, not some nefarious spyware.