Llama 3.1 8B Instruct
Meta
The most widely supported open model there is. If a tool, adapter or tutorial exists, it was written for this one first.
Llama 3.1 8B Instruct does not fit on CPU only — 64 GB workstation.
CPU-only inference. Expect a few tokens per second at best.
- Unmatched ecosystem support
- 128k context
- Very stable fine-tuning behaviour
- Benchmarks now behind newer 7–9B models
- Community licence with an acceptable-use policy
llama3.1:8bRunning Llama 3.1 8B Instruct on CPU only — 64 GB workstation
| Quantization | Quality | Weights | KV cache | Total | ~tok/s | Fit |
|---|---|---|---|---|---|---|
| F16 | lossless | 15 GB | 1.0 GB | 16 GB | 6 | Not recommended |
| Q8_0 | near-lossless | 7.9 GB | 1.0 GB | 9.5 GB | 11 | Not recommended |
| Q6_K | near-lossless | 6.1 GB | 1.0 GB | 7.7 GB | 14 | Not recommended |
| Q5_K_M | high | 5.3 GB | 1.0 GB | 6.9 GB | 16 | Not recommended |
| Q4_K_M | balanced | 4.5 GB | 1.0 GB | 6.1 GB | 19 | Not recommended |
| Q3_K_M | degraded | 3.6 GB | 1.0 GB | 5.2 GB | 24 | Not recommended |
KV cache is sized at 8,192 tokens. Longer contexts cost proportionally more — the recommendation above reserves room for a working context.
Local fine-tuning on this machine
- Base weights
- 4.2 GB
- Optimizer
- 625 MB
- Activations
- 900 MB
- Peak
- 7.2 GB
- Local fine-tuning needs a GPU or Apple Silicon. CPU training is not practical.
- Base weights
- 15 GB
- Optimizer
- 625 MB
- Activations
- 900 MB
- Peak
- 18 GB
- Local fine-tuning needs a GPU or Apple Silicon. CPU training is not practical.
Where this model runs
VRAM 32 GB · F16 · 16 GB
VRAM 24 GB · Q8_0 · 9.5 GB
VRAM 16 GB · Q6_K · 7.7 GB
VRAM 16 GB · Q6_K · 7.7 GB
VRAM 24 GB · Q8_0 · 9.5 GB
VRAM 12 GB · Q6_K · 7.7 GB
VRAM 12 GB · Q6_K · 7.7 GB
VRAM 16 GB · Q6_K · 7.7 GB
VRAM 48 GB · F16 · 16 GB
VRAM 80 GB · F16 · 16 GB
Unified 128 GB · F16 · 16 GB
Unified 48 GB · F16 · 16 GB
Unified 24 GB · Q8_0 · 9.5 GB
Unified 192 GB · F16 · 16 GB
Unified 16 GB · Q4_K_M · 6.1 GB
VRAM 24 GB · Q8_0 · 9.5 GB
VRAM 16 GB · Q6_K · 7.7 GB
VRAM 0 MB
VRAM 0 MB
69.4%
Llama 3.1 model card
72.6%
Llama 3.1 model card
Reported by the model's author. ModelLM has not run these benchmarks and does not treat them as verified.
Qwen2.5 7B Instruct
7.6B · Apache 2.0
The default starting point for local work on 8–12 GB cards. Strong instruction following and reliable tool-call formatting for its size.
Qwen2.5 Coder 7B Instruct
7.6B · Apache 2.0
The practical local copilot. Supports fill-in-the-middle, so it works as an inline completion model rather than only a chat assistant.
Qwen3 8B
8.2B · Apache 2.0
Switchable thinking mode: the same weights answer directly or reason step by step depending on the prompt. Long context for its size.
Llama 3.2 3B Instruct
3.2B · Llama 3.2 Community License
Small enough for a laptop CPU or a 6 GB card, and still coherent. A sensible target for edge deployment.
Catalogue figures come from each model’s published card. ModelLM has not independently measured them.