Llama 3.2 3B Instruct
Meta · 3.2B parameters · 131,072 token max context · Model card
Memory needed
| Quant | Weights | Total @ 8K ctx |
|---|---|---|
| Q3_K_M | 1.6 GB | 3.3 GB |
| Q4_K_M | 1.9 GB | 3.7 GB |
| Q5_K_M | 2.3 GB | 4.1 GB |
| Q6_K | 2.6 GB | 4.5 GB |
| Q8_0 | 3.4 GB | 5.2 GB |
| F16 | 6.4 GB | 8.3 GB |
Approximate GGUF sizes from bits-per-weight; totals include fp16 KV cache and ~0.8 GB overhead.
Can my hardware run it?
| Device | Verdict | Best quant | Est. tok/s (Q4) |
|---|---|---|---|
| NVIDIA GeForce RTX 5090 | Yes | F16 | ~300 |
| NVIDIA GeForce RTX 4090 | Yes | F16 | ~300 |
| NVIDIA GeForce RTX 3090 | Yes | F16 | ~300 |
| NVIDIA GeForce RTX 3090 Ti | Yes | F16 | ~300 |
| NVIDIA GeForce RTX 5080 | Yes | F16 | ~300 |
| NVIDIA GeForce RTX 5070 Ti | Yes | F16 | ~300 |
| AMD Radeon RX 7900 XTX | Yes | F16 | ~300 |
| NVIDIA GeForce RTX 4080 SUPER | Yes | F16 | ~292 |
| NVIDIA GeForce RTX 3080 10GB | Yes | F16 | ~275 |
| NVIDIA GeForce RTX 4070 Ti SUPER | Yes | F16 | ~261 |
| NVIDIA GeForce RTX 5070 | Yes | F16 | ~251 |
| Apple M5 Max (40-core GPU, 128GB) | Yes | F16 | ~236 |
| Apple M2 Ultra (76-core GPU, 192GB) | Yes | F16 | ~186 |
| Apple M3 Ultra (80-core GPU, 512GB) | Yes | F16 | ~182 |
| NVIDIA GeForce RTX 5060 Ti 16GB | Yes | F16 | ~179 |
| Apple M4 Max (40-core GPU, 128GB) | Yes | F16 | ~164 |
| NVIDIA GeForce RTX 3060 12GB | Yes | F16 | ~149 |
| Apple M5 Pro (20-core GPU, 64GB) | Yes | F16 | ~131 |
| Apple M3 Max (40-core GPU, 64GB) | Yes | F16 | ~131 |
| NVIDIA GeForce RTX 4060 Ti 8GB | Yes | Q8_0 | ~126 |
| Apple M1 Max (32-core GPU, 64GB) | Yes | F16 | ~121 |
| NVIDIA DGX Spark | Yes | F16 | ~113 |
| Apple M4 Pro (20-core GPU, 48GB) | Yes | F16 | ~100.0 |
| Apple M4 (10-core GPU, 24GB) | Yes | F16 | ~47.5 |
Speeds estimated from memory bandwidth (calibrated per device with a measured llama.cpp benchmark where available).