Record 026 · Models · 2024
50%24.26.L
Llama 3
Meta
“Llama 3 8B/70B became the new default GGUF on every homelab.”
Llama 3 (and 3.1/3.2/3.3) reset local baselines: 8B that felt like last year's 70B for a lot of chores, 70B that fit in 48GB quantized.
The second-wave weights the current local stack assumes.
- Llama 3 8B/70B
- 128K context later 3.1
- Edge 3.2 1B/3B
The 8B that replaced the 7B.
