Model comparison
Supra-50M-Base vs Supra-Router-51M
At Q4_K_M, Supra-50M-Base is the smaller download — 37,445,408 bytes against 37,446,208.
From the file· summed bytes, KV per layer
Side by side
| Supra-50M-Base | Supra-Router-51M | |
|---|---|---|
| Parameters | 52M | 52M |
| Architecture | llama | llama |
| Layers | 12 | 12 |
| Native context | 1,024 | 5,120 |
| Mixture of experts | no | no |
| Quantizations published | 33 | 18 |
| Smallest quantization | 0.02 GiB | 0.02 GiB |
| Q4_K_M | 0.03 GiB | 0.03 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Supra-50M-Base | Supra-Router-51M | Ratio |
|---|---|---|---|
| 4,096 | 0.05 GiB | 0.05 GiB | 1.00× |
| 8,192 | 0.09 GiB | 0.09 GiB | 1.00× |
| 16,384 | 0.19 GiB | 0.19 GiB | 1.00× |
| 32,768 | 0.38 GiB | 0.38 GiB | 1.00× |
| 65,536 | 0.75 GiB | 0.75 GiB | 1.00× |
| 131,072 | 1.50 GiB | 1.50 GiB | 1.00× |