Can I run LFM2.5-24B-A2B on an RTX 5090?
Yes, with a GGUF version
The published weights are too big, but the Q8_0 file from unsloth/LFM2.5-24B-A2B-GGUF (25.5 GB) fits, with room for ≈ 128K tokens of context.
ollama run hf.co/unsloth/LFM2.5-24B-A2B-GGUF:Q8_0Files and context on RTX 5090
Each file of LFM2.5-24B-A2B (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 5090 | Context (16-bit) | Context (8-bit) |
|---|---|---|---|---|
| As published | 47.9 GB | Doesn't fit | — | — |
| BF16 | 48.0 GB | Doesn't fit | — | — |
| Q8_0 | 25.5 GB | Tight | 128K | 128K |
| UD-Q6_K_XL | 21.9 GB | Fits | 128K | 128K |
| UD-Q5_K_XL | 17.9 GB | Fits | 128K | 128K |
| UD-Q4_K_XL | 14.9 GB | Fits | 128K | 128K |
| UD-Q3_K_XL | 10.9 GB | Fits | 128K | 128K |
| UD-IQ3_XXS | 8.96 GB | Fits | 128K | 128K |
| UD-Q2_K_XL | 8.04 GB | Fits | 128K | 128K |
| UD-IQ2_M | 7.54 GB | Fits | 128K | 128K |
| UD-IQ2_XXS | 7.47 GB | Fits | 128K | 128K |
| BF16 | 48.0 GB | Doesn't fit | — | — |
| F16 | 48.0 GB | Doesn't fit | — | — |
| Q8_0 | 25.5 GB | Tight | 128K | 128K |
| Q6_K | 19.7 GB | Fits | 128K | 128K |
| Q5_K_M | 17.0 GB | Fits | 128K | 128K |
| Q4_K_M | 14.5 GB | Fits | 128K | 128K |
| Q4_0 | 13.6 GB | Fits | 128K | 128K |