scw · HackerTrans

22 дня назадdiscuss

As a kid, my first machine was a 486 DX2 66. My dad mandated we buy on value, and I scoured the MicroTimes looking for the best possible deal, finally settling on a machine from a small independent shop off of South Bascom Ave in San Jose. The office smelt like burnt coffee and exhaust from burning-in PCs, but the machine was awesome: a PCI video card, and as we later found out, a built-in SCSI adapter which meant we needed to throw in another $240 for an Adaptec SCSI controller when we later upgraded to Pentium.

scw

в прошлом месяцеdiscuss

The original post [1] now includes an update:

  UPDATE! Within a day of this blowing up on Hacker News, AMD reached back 
  out to me and said they would be looking into the matter after all.

[1] https://mrbruh.com/amd2/

scw

3 месяца назадdiscuss

They're unlikely to find much unless they instead try “dispositive”.

scw

4 месяца назадdiscuss

TurboQuant has a specific benefit by compressing the KV cache at a negligible cost to quality. That mainly means that the context lengths can go up in models for the same amount of memory, however the KV cache only accounts for something like 20% of the overall model size, and this will not dramatically decrease memory demands in the way that some of the more sensationalist reporting has stated.

scw

в прошлом годуdiscuss

Still active, but many fewer resources than in the past. Many backends like CUDA for Windows have been dropped and others pushed off to partners with varying levels of support. TensorFlow 2.19 is going to release soon without Python 3.13 support, it's hard not to imagine that resource constraints are at play.

scw

2 года назадdiscuss

The Microsoft stake finally allowed him to let loose and buy a keyboard with a working shift key.

scw

2 года назадdiscuss

Exciting concept! Note that the LLM corrected version does drop a full paragraph from the output at the bottom of the second page (starting with an asterisk and "My views regarding inflationary possibilities". I'm not sure if there is a simple way to mitigate this risk but would be nice to fall back on uncorrected text if the LLM can't produce valid results for some region of the document.

comments