Model comparison
ArrowMint-Gemma3-4B-YUKI-v0.1 vs Qwen3-8B
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: ArrowMint-Gemma3-4B-YUKI-v0.1's KV cache at 32K is 5.7× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| ArrowMint-Gemma3-4B-YUKI-v0.1 | Qwen3-8B | |
|---|---|---|
| Parameters | 4.3B | 8.2B |
| Architecture | gemma3 | qwen3 |
| Layers | 34 | 36 |
| Native context | 131,072 | 40,960 |
| Mixture of experts | no | no |
| Quantizations published | 24 | 51 |
| Smallest quantization | 1.06 GiB | 2.12 GiB |
| Q4_K_M | — | 4.68 GiB |
| Licence | gemma | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | ArrowMint-Gemma3-4B-YUKI-v0.1 | Qwen3-8B | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.56 GiB | 2.27× |
| 8,192 | 0.33 GiB | 1.13 GiB | 3.45× |
| 16,384 | 0.48 GiB | 2.25 GiB | 4.66× |
| 32,768 | 0.79 GiB | 4.50 GiB | 5.66× |
| 65,536 | 1.42 GiB | 9.00 GiB | 6.34× |
| 131,072 | 2.67 GiB | 18.00 GiB | 6.74× |