Model comparison
NuExtract3 vs Qwythos-9B-v2
At Q4_K_M, NuExtract3 is the smaller download — 2,708,803,456 bytes against 6,047,707,680.
From the file· summed bytes, KV per layer
Side by side
| NuExtract3 | Qwythos-9B-v2 | |
|---|---|---|
| Parameters | 4.5B | 9.7B |
| Architecture | qwen35 | qwen35 |
| Layers | 32 | 32 |
| Native context | 262,144 | 1,048,576 |
| Mixture of experts | no | no |
| Quantizations published | 6 | 43 |
| Smallest quantization | 1.78 GiB | 3.64 GiB |
| Q4_K_M | 2.52 GiB | 5.63 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | NuExtract3 | Qwythos-9B-v2 | Ratio |
|---|---|---|---|
| 4,096 | 0.13 GiB | 0.13 GiB | 1.00× |
| 8,192 | 0.25 GiB | 0.25 GiB | 1.00× |
| 16,384 | 0.50 GiB | 0.50 GiB | 1.00× |
| 32,768 | 1.00 GiB | 1.00 GiB | 1.00× |
| 65,536 | 2.00 GiB | 2.00 GiB | 1.00× |
| 131,072 | 4.00 GiB | 4.00 GiB | 1.00× |