Can I run occamy-1.0 on an RTX 3090?
Yes, with a GGUF version
The published weights are too big, but the Q4_K_M file from Accio-Lab/occamy-1.0-GGUF (21.2 GB) fits, with room for ≈ 18K tokens of context.
ollama run hf.co/Accio-Lab/occamy-1.0-GGUF:Q4_K_MFiles and context on RTX 3090
Each file of occamy-1.0 (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 3090 | Context (16-bit) | Context (8-bit) |
|---|---|---|---|---|
| As published | 70.2 GB | Doesn't fit | — | — |
| Q8_0 | 36.9 GB | Doesn't fit | — | — |
| Q6_K | 28.5 GB | Doesn't fit | — | — |
| Q5_K_M | 24.7 GB | Doesn't fit | — | — |
| Q5_0 | 24.0 GB | Doesn't fit | — | — |
| Q5_K_S | 24.0 GB | Doesn't fit | — | — |
| Q4_K_M | 21.2 GB | Tight | 18K | 35K |
| Q4_K_S | 19.9 GB | Tight | 78K | 157K |
| IQ4_NL | 19.8 GB | Tight | 84K | 167K |
| Q4_0 | 19.7 GB | Tight | 87K | 174K |
| IQ4_XS | 18.7 GB | Tight | 134K | 256K |
| Q3_K_L | 18.1 GB | Tight | 163K | 256K |
| Q3_K_M | 16.8 GB | Fits | 227K | 256K |
| IQ3_M | 15.4 GB | Fits | 256K | 256K |
| Q3_K_S | 15.2 GB | Fits | 256K | 256K |
| IQ3_XS | 14.5 GB | Fits | 256K | 256K |
| Q2_K | 12.9 GB | Fits | 256K | 256K |
| IQ2_M | 11.7 GB | Fits | 256K | 256K |