Phi-4
microsoft/Phi-4
Microsoft Phi-4, compact 14B with strong reasoning.
Parameters
14.7B
Context
16K
Downloads
520K
Likes
2K
Architecture
phi3
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 39.6 GB RAM 59.4 · Disk 81.7 | 56.9 GB RAM 85.4 · Disk 107.7 | 155.4 GB RAM 233.1 · Disk 146.7 |
| FP88bit | 22.3 GB RAM 33.4 · Disk 60.4 | 38.8 GB RAM 58.2 · Disk 86.4 | 136.5 GB RAM 204.7 · Disk 125.4 |
| INT88bit | 22.3 GB RAM 33.4 · Disk 60.4 | 38.8 GB RAM 58.2 · Disk 86.4 | 136.5 GB RAM 204.7 · Disk 125.4 |
| AWQ4bit | 13.9 GB RAM 32.0 · Disk 50.1 | 30.1 GB RAM 45.1 · Disk 76.1 | 127.4 GB RAM 191.1 · Disk 115.1 |
| GPTQ4bit | 14.0 GB RAM 32.0 · Disk 50.2 | 30.2 GB RAM 45.3 · Disk 76.2 | 127.5 GB RAM 191.3 · Disk 115.2 |
| GGUF4bit | 14.3 GB RAM 32.0 · Disk 50.5 | 30.4 GB RAM 45.7 · Disk 76.5 | 127.8 GB RAM 191.6 · Disk 115.5 |
GPU Recommendations
1× H100 SXM
80 GB VRAM
≈ 471.8 tok/s
Effective production throughput ≈ 330.3 tok/s
score 0.585
2× RTX 3090
48 GB VRAM
≈ 210.9 tok/s
Effective production throughput ≈ 147.6 tok/s
score 0.6216
8× H200 SXM
1128 GB VRAM
≈ 2163.4 tok/s
Effective production throughput ≈ 1514.4 tok/s
score 0.2197
1× RTX A6000
48 GB VRAM
≈ 108.2 tok/s
Effective production throughput ≈ 75.7 tok/s
score 0.6571
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
RTX A6000
48 GB VRAM
$265
/mo · vast
RTX 5090D
32 GB VRAM
$294
/mo · vast
RTX 5090
32 GB VRAM
$325
/mo · vast
L40S
48 GB VRAM
$341
/mo · vast
A100 PCIE
80 GB VRAM
$390
/mo · vast