Can I run Arete-27B on an RTX 3060 12 GB?
Yes, with a GGUF version
The published weights are too big, but the IQ2_M file from bartowski/NousResearch_Arete-27B-GGUF (10.3 GB) fits, with room for ≈ 5K tokens of context. That file is under 3 bits per weight, so expect noticeably lower quality.
Full analysis of Arete-27B
Everything that runs on RTX 3060 12 GB
Get the IQ2_M file on Hugging Face
ollama run hf.co/bartowski/NousResearch_Arete-27B-GGUF:IQ2_MFiles and context on RTX 3060 12 GB
Each file of Arete-27B (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.
| File | Size | On RTX 3060 12 GB | Context (16-bit) | Context (8-bit) |
|---|---|---|---|---|
| As published | 54.7 GB | Doesn't fit | — | — |
| BF16 | 53.8 GB | Doesn't fit | — | — |
| Q8_0 | 28.7 GB | Doesn't fit | — | — |
| Q6_K_L | 24.7 GB | Doesn't fit | — | — |
| Q6_K | 23.6 GB | Doesn't fit | — | — |
| Q6_K_S | 22.6 GB | Doesn't fit | — | — |
| Q5_K_M | 20.7 GB | Doesn't fit | — | — |
| Q5_K_S | 19.3 GB | Doesn't fit | — | — |
| Q4_K_L | 18.6 GB | Doesn't fit | — | — |
| Q4_1 | 17.6 GB | Doesn't fit | — | — |
| IQ4_NL | 17.2 GB | Doesn't fit | — | — |
| Q4_K_M | 17.2 GB | Doesn't fit | — | — |
| Q4_K_S | 16.1 GB | Doesn't fit | — | — |
| Q4_0 | 16.1 GB | Doesn't fit | — | — |
| IQ4_XS | 15.2 GB | Doesn't fit | — | — |
| IQ3_M | 14.6 GB | Doesn't fit | — | — |
| Q3_K_L | 13.9 GB | Doesn't fit | — | — |
| Q3_K_M | 13.2 GB | Doesn't fit | — | — |
| IQ3_XS | 12.6 GB | Doesn't fit | — | — |
| Q3_K_S | 12.5 GB | Doesn't fit | — | — |
| IQ3_XXS | 12.1 GB | Doesn't fit | — | — |
| Q2_K | 10.6 GB | Doesn't fit | 968 | 2K |
| IQ2_M | 10.3 GB | Tight | 5K | 11K |
| IQ2_S | 9.45 GB | Tight | 18K | 36K |
| IQ2_XS | 8.85 GB | Tight | 27K | 54K |
| IQ2_XXS | 8.64 GB | Tight | 30K | 60K |