Llama
- Llama 2 7B7B
Meta's 2023 dense 7B model. Still shows up in older llama.cpp scoreboards.
- Llama 3.2 3B3B
Small dense Llama. Comfortable on 8 GB at Q8.
- Llama 3 8B8B
Dense 8B Llama 3. A common baseline on 8–16 GB cards.
- Llama 3 70B70B
Dense 70B Llama 3. The Q4 weight estimate is well past a 24 GB card.
- Llama 3.1 8B8B
Dense 8B with a 128k context. Fits on small GPUs at Q4 and Q8.
- Llama 3.1 70B70B
Dense 70B Llama 3.1. Same weight class as Llama 3.3 70B.
- Llama 3.1 405B405B
Dense 405B. Even Q4 is a multi-GPU or high-memory-Mac problem.
- Llama 3.3 70B70B
Dense 70B instruct model. The usual 'will this fit on 24 GB?' question.