Re:DeepSeek-R1 with Dynamic 1.58-bit Quantization
by whimsicalism on Tuesday January 28, 2025 @10:15AM
(#42853237)
There’s a huge difference both in capabilities and in meaning between “variations of r1” and “r1 distill”. ollama is intentionally misleading people on this but the distills are much much worse
[ Parent
| on HN ]
Re:DeepSeek-R1 with Dynamic 1.58-bit Quantization
by thot_experiment on Tuesday January 28, 2025 @01:18PM
(#42855859)
They're really not? Both subjectively and in benchmarks there is no world in which the delta between the models deserves a "much much".