information that will lead to the arrest of Hillary Clinton
@gotpurpledrank.bsky.social
Accept no substitutes.
Oh man! The thing from the popular book Don’t Build The Torment Nexus?
October 2, 2025 at 4:01 PM
Oh man! The thing from the popular book Don’t Build The Torment Nexus?
The lowest brainpower model I can get to work with Mira is a 8 Billion Full Precision 16 model.
TL;DR Quantization makes the model run with less computing power by truncating the precision of the tokens.
- FP16: 1.726
- Q8: 1.73
- Q4: 1.7
- Q2: 2.0 (rounded to nearest available value)
TL;DR Quantization makes the model run with less computing power by truncating the precision of the tokens.
- FP16: 1.726
- Q8: 1.73
- Q4: 1.7
- Q2: 2.0 (rounded to nearest available value)
June 12, 2025 at 7:27 AM
The lowest brainpower model I can get to work with Mira is a 8 Billion Full Precision 16 model.
TL;DR Quantization makes the model run with less computing power by truncating the precision of the tokens.
- FP16: 1.726
- Q8: 1.73
- Q4: 1.7
- Q2: 2.0 (rounded to nearest available value)
TL;DR Quantization makes the model run with less computing power by truncating the precision of the tokens.
- FP16: 1.726
- Q8: 1.73
- Q4: 1.7
- Q2: 2.0 (rounded to nearest available value)
This actually happened to me today. I got slop.
But don’t forget: hard fought solutions are durable! You’ll understand *why* your algo is good once it gets there.
But don’t forget: hard fought solutions are durable! You’ll understand *why* your algo is good once it gets there.
June 12, 2025 at 7:07 AM
This actually happened to me today. I got slop.
But don’t forget: hard fought solutions are durable! You’ll understand *why* your algo is good once it gets there.
But don’t forget: hard fought solutions are durable! You’ll understand *why* your algo is good once it gets there.