Phi-4 (14.7B): VRAM by quant
14.7B params · KV field 204800 · max context field 16384.
| Quant | Bits | Weight GB sketch |
|---|---|---|
| Q4_K_M | 4.85 bpw | ~9.1 GB |
| Q5_K_M | 5.7 bpw | ~10.6 GB |
| Q6_K | 6.6 bpw | ~12.2 GB |
| Q8_0 | 8.5 bpw | ~15.7 GB |
| MXFP4 | 4.25 bpw | ~8.1 GB |
| F16/BF16 | 16.0 bpw | ~29.4 GB |
Add KV for your context length in context VRAM. Overall fit: VRAM fit.
Educational. MoE active params may differ from total params.