1,149 karma · joined August 21, 2016
The cloud-based software for everything else has degraded in quality, tjough. I'll probably upgrade to a lidar-equipped competitor model if this continues to get worse after this bancruptcy.
Miele (at a more premium price point) production is even more concentrated in Germany. https://m.miele.com/en/com/production-sites-2157.htm
(Edit: No replies after 8 hours, but of course they then came in quickly after Europe woke up..)
Less weight on the parts of the body that move most during a run is an obvious benefit of not wearing shoes.
> Even Nvidia itself has tools that do not exclusively rely on CUDA. For example, Triton Inference Server is an open-source tool by Nvidia that simplifies deploying AI models at scale, supporting frameworks like TensorFlow, PyTorch, and ONNX. Triton also provides features like model versioning, multi-model serving, and concurrent model execution to optimize the utilization of GPU and CPU resources.
> Nvidia's TensorRT is a high-performance deep learning inference optimizer and runtime library that accelerates deep learning inference on Nvidia GPUs. [...]
Keller was speaking of OpenAI's Triton (https://openai.com/research/triton), a Python-like language that is compiled to code for Nvidia GPUs, but Tom's Hardware mixed this up with Nvidia's Triton Inference Server, a higher level tool that's really not a replacement for CUDA and not directly related to the Triton language. Easy to confuse these if you are a writer in a hurry.
In countries where the Amazon Prime membership (originally this was for fast free shipping only) is available, Prime Video is included with that: https://en.wikipedia.org/wiki/Amazon_Prime#Availability
Outside of those countries you can typically get Prime Video as a cheaper, separate subscription: https://en.wikipedia.org/wiki/Amazon_Prime_Video#Availabilit...
In Germany, for instance, I don't think you can get Prime Video without the full Amazon Prime subscription. In countries where the shipping benefit is more costly (like the US) they may offer different membership tiers.
Typically I subscribe to these services for just a few months at a time, depending on what's interesting. Harder to do with accounts that are shared by a larger household, I guess.
Pizza producers can game the system somewhat, however, by adding ingredients that are counted positive (like protein or fiber) to partially offset "negative" ingredients such as sugar or saturated fats.
Dairy lobbying has led to cheese as a separate category and that milk is counted as food, rather than drink.
And you need to the inference again and again, not just a single time (like your training).
> Inference will Dominate, not Training
This rings true. While LLMs will be fine-tuned by many, fewer companies will train their own independent foundation models from scratch (which doesn't require a "few GPUs", but hundreds with tight interconnect). The inference cost of running these in applications will dominate in these companies.
> CPUs are Competitive for Inference
I disagree for LLMs. Running the inference still takes a lot of the type of compute that GPUs are optimized for. If you want to respond to your customers' requests with acceptable latency (and achieve some throughput), you will want to use GPUs. For "medium-sized" LLMs you won't need NVLink-level interconnect speeds between your GPUs, though.
The actual cut off for the assistance is typically set to 27.5 km/h, because regulations allow for 10% of tolerance... Then, depending on transmission ratios, it can be quite comfortable to ride at ~30 km/h. The acceleration boost at lower speeds is still super useful.
Of course, VanMoof's and Cowboy's original spiel was that you could easily disable the limit in software.