Model comparison
Llama-3.1-Minitron-4B-Width-Base vs Qwen3-8B
At Q4_K_M, Llama-3.1-Minitron-4B-Width-Base is the smaller download — 2,778,282,976 bytes against 5,027,783,488.
From the file· summed bytes, KV per layer
Side by side
| Llama-3.1-Minitron-4B-Width-Base | Qwen3-8B | |
|---|---|---|
| Parameters | 4.5B | 8.2B |
| Architecture | llama | qwen3 |
| Layers | 32 | 36 |
| Native context | 131,072 | 40,960 |
| Mixture of experts | no | no |
| Quantizations published | 35 | 51 |
| Smallest quantization | 1.71 GiB | 2.12 GiB |
| Q4_K_M | 2.59 GiB | 4.68 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Llama-3.1-Minitron-4B-Width-Base | Qwen3-8B | Ratio |
|---|---|---|---|
| 4,096 | 0.50 GiB | 0.56 GiB | 1.13× |
| 8,192 | 1.00 GiB | 1.13 GiB | 1.13× |
| 16,384 | 2.00 GiB | 2.25 GiB | 1.13× |
| 32,768 | 4.00 GiB | 4.50 GiB | 1.13× |
| 65,536 | 8.00 GiB | 9.00 GiB | 1.13× |
| 131,072 | 16.00 GiB | 18.00 GiB | 1.13× |