Qwen/Qwen3.5-4B
Qwen/Qwen3.5-4B
Parameters
3.5B
Context
256K
Downloads
7.5M
Likes
811
Architecture
qwen3_5
32 layers · 4 KV heads · head 256 · hidden 2560
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 14.4 GB RAM 32.0 · Disk 50.6 | 30.6 GB RAM 45.8 · Disk 76.6 | 127.9 GB RAM 191.8 · Disk 115.6 |
| FP88bit | 9.7 GB RAM 32.0 · Disk 44.8 | 25.6 GB RAM 38.4 · Disk 70.8 | 122.7 GB RAM 184.1 · Disk 109.8 |
| INT88bit | 9.7 GB RAM 32.0 · Disk 44.8 | 25.6 GB RAM 38.4 · Disk 70.8 | 122.7 GB RAM 184.1 · Disk 109.8 |
| AWQ4bit | 7.4 GB RAM 32.0 · Disk 42.0 | 23.3 GB RAM 34.9 · Disk 68.0 | 120.3 GB RAM 180.4 · Disk 107.0 |
| GPTQ4bit | 7.4 GB RAM 32.0 · Disk 42.1 | 23.3 GB RAM 34.9 · Disk 68.1 | 120.3 GB RAM 180.5 · Disk 107.1 |
| GGUF4bit | 7.5 GB RAM 32.0 · Disk 42.1 | 23.4 GB RAM 35.0 · Disk 68.1 | 120.4 GB RAM 180.5 · Disk 107.1 |
GPU Recommendations
1× RTX 5090
32 GB VRAM
≈ 928.5 tok/s
Effective production throughput ≈ 649.9 tok/s
score 0.6916
2× Tesla V100
32 GB VRAM
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.6405
8× H200 SXM
1128 GB VRAM
≈ 7958.5 tok/s
Effective production throughput ≈ 5571 tok/s
score 0.2197
1× RTX 3090
24 GB VRAM
≈ 485 tok/s
Effective production throughput ≈ 339.5 tok/s
score 0.6424
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
RTX 3090
24 GB VRAM
$45.4
/mo · vast
RTX 4090
24 GB VRAM
$98.1
/mo · vast
RTX 4090D
24 GB VRAM
$117
/mo · vast
L4
24 GB VRAM
$148
/mo · vast
A10
24 GB VRAM
$176
/mo · vast