Qwen/Qwen2.5-VL-3B-Instruct
Qwen/Qwen2.5-VL-3B-Instruct
Parameters
2.6B
Context
125K
Downloads
4.4M
Likes
690
Architecture
qwen2_5_vl
36 layers · 2 KV heads · head 128 · hidden 2048
GPU Requirements
VRAM / RAM / disk estimates by quantization and usage scenario
| Quantization | Minimum | Recommended | Production |
|---|---|---|---|
| FP1616bit | 12.3 GB RAM 32.0 · Disk 49.8 | 14.9 GB RAM 32.0 · Disk 75.8 | 31.1 GB RAM 46.7 · Disk 114.8 |
| FP88bit | 7.9 GB RAM 32.0 · Disk 44.4 | 10.3 GB RAM 32.0 · Disk 70.4 | 26.4 GB RAM 39.5 · Disk 109.4 |
| INT88bit | 7.9 GB RAM 32.0 · Disk 44.4 | 10.3 GB RAM 32.0 · Disk 70.4 | 26.4 GB RAM 39.5 · Disk 109.4 |
| AWQ4bit | 5.8 GB RAM 32.0 · Disk 41.8 | 8.1 GB RAM 32.0 · Disk 67.8 | 24.1 GB RAM 36.1 · Disk 106.8 |
| GPTQ4bit | 5.9 GB RAM 32.0 · Disk 41.8 | 8.1 GB RAM 32.0 · Disk 67.8 | 24.1 GB RAM 36.1 · Disk 106.8 |
| GGUF4bit | 5.9 GB RAM 32.0 · Disk 41.9 | 8.2 GB RAM 32.0 · Disk 67.9 | 24.2 GB RAM 36.2 · Disk 106.9 |
GPU Recommendations
1× RTX 3080 Ti
12 GB VRAM
≈ 509.5 tok/s
Effective production throughput ≈ 356.7 tok/s
score 0.6751
1× Tesla V100
16 GB VRAM
≈ 0 tok/s (N/A — benchmark unavailable)
Effective production throughput ≈ 0 tok/s
score 0.6244
8× RTX 5090
256 GB VRAM
≈ 3203.6 tok/s
Effective production throughput ≈ 2242.5 tok/s
score 0.3522
1× RTX 3080
10 GB VRAM
≈ 424.6 tok/s
Effective production throughput ≈ 297.2 tok/s
score 0.6777
💬 Community — real deployment experience
0 Articles · 0 Benchmarks
GPUs that fit (single card)
Tesla V100
16 GB VRAM
$35.7
/mo · vast
RTX 4070S Ti
16 GB VRAM
$68.5
/mo · vast
RTX 3080
10 GB VRAM
$72.6
/mo · vast
L4
24 GB VRAM
$76.2
/mo · vast
RTX 3080 Ti
12 GB VRAM
$78.8
/mo · vast