pama

21 ms·

391.
by pama 2y ago
No human would score high on that puzzle if the images were given to them as a series of tokens. Even previous LLMs scored much better than humans if tested in the same way.
392.
by pama 2y ago
Patience. I understand the frustration. I hope this transient login problem gets fixed soon.
393.
Amazon EC2 Trn2 Instances and Trn2 UltraServers for AI/ML now available(aws.amazon.com)
2 points by pama 2y ago|0 comments
394.
Advent of Cyber(tryhackme.com)
2 points by pama 2y ago|0 comments
395.
by pama 2y ago
I thought that Bonus Content #1 and the references down the article were reasonably convincing. It would be great if large companies disclosed such details but it is unlikely.
396.
Verify the Safety of the Rust Standard Library(aws.amazon.com)
2 points by pama 2y ago|0 comments
397.
The Nearest Neighbor Attack: How a Russian Apt Weaponized Nearby Wi-Fi Networks(volexity.com)
4 points by pama 2y ago|0 comments
398.
by pama 2y ago
https://arxiv.org/abs/2408.04498
399.
by pama 2y ago
Agreed on conciseness of the implementation. It is short and clear despite having a Max and Min that share all except one character in 30 lines of code.
400.
by pama 2y ago
All HPC and LLMs tend to get fully optimized to their hardware specs. When you train models with over 405B parameters and process about 2 million tokens per second calculating derivatives on all these parameters every few seconds, you do en
401.
by pama 2y ago
If I knew, I wouldn’t be able to disclose it :-)
402.
by pama 2y ago
Agreed. However also note that if it was only matrix multiplies and no full transformer training, the performance of that Meta cluster would be closer to 16k PFlops/s, still much faster than the El Capitain performance measured on linp
403.
by pama 2y ago
Absolutely. Though the performance of El Capitain is only measured by a linpack benchmark not the actual application.
404.
by pama 2y ago
Noting here that 2700 quadrillion operations per second is less than the estimated sustained throughput of productive bfloat16 compute during the training of the large llama3 models, which IIRC was about 45% of 16,000 quadrillion operations
405.
by pama 2y ago
Well, Cerebras pretty much needs a data center to simply fit the 405B model for inference.
406.
by pama 2y ago
However: “The doctors that could use the software got a median score of 76 percent on making a diagnosis and explaining a reason for it, while the group that used only conventional resources had a median score of 74 percent. Run on its own,
407.
ChatGPT was little help for doctors in diagnosing diseases, study finds(washingtonpost.com)
5 points by pama 2y ago|1 comments
408.
Boltz-1: Democratizing Biomolecular Interaction Modeling(github.com)
1 points by pama 2y ago|0 comments
409.
by pama 2y ago
The author explains what they did: restrict the move options to valid ones when possible (for open models with the ability to enforce grammar during inference) or sample the model for a valid move up to ten times, then pick a random valid m
410.
Visions of A.I. Art from OpenAI's First Artist in Residence(nytimes.com)
2 points by pama 2y ago|0 comments
411.
by pama 2y ago
There are two populations of people reading the NVIDIA specs (and now you switched groups). If NVIDIA ever changes their marketing strategy and the asterisk denotes something else, there might be a third population because I know a lot of
412.
by pama 2y ago
To clarify the title: TFLOP/s is the unit the author goes after, not TFLOP. People in the threads compare CUDA performance on GPUs to WebAssembly performance: please recall that H100 has a theoretical performance of about 1000 TFLOP&#
413.
by pama 2y ago
I’ve automagically recorded my digital piano musings starting about 28 years ago. Initially, I used a simple C code on a Linux server using select to dump from dev MIDI to a timestamped file per session (a variation of an answering machine
414.
by pama 2y ago
An interesting concept, however, the readability of this paper would greatly improve with LaTeX formatting.
415.
by pama 2y ago
Thanks!
416.
by pama 2y ago
Why not simply remove that dependency now?
417.
by pama 2y ago
I work in this area. It took time to know what each protein does in a cell (we still dont really understand most of them), it took time to find all the ways that various cancers use the protein machinery in the cell to their benefit (new wa
418.
Release of LLM-Claude with pdf support(simonwillison.net)
2 points by pama 2y ago|1 comments
419.
by pama 2y ago
Interestingly, python f-strings changed their syntax at version 3.12, so highlighting should depend on the version.
420.
by pama 2y ago
I guess that with a steaming hot potato the higher risk is to inhale rosin in the steam rather than to eat it. I would much rather use duck fat to roast potatoes.

More ›