Model comparison
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 18,366,980,192. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 11.6× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 30.0B | 8.0B |
| Architecture | llama | gemma4 |
| Layers | 47 | 42 |
| Native context | 131,072 | 131,072 |
| Mixture of experts | yes, 8 experts | no |
| Quantizations published | 34 | 36 |
| Smallest quantization | 6.07 GiB | 3.30 GiB |
| Q4_K_M | 17.11 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.73 GiB | 0.12 GiB | 5.97× |
| 8,192 | 1.47 GiB | 0.18 GiB | 8.26× |
| 16,384 | 2.94 GiB | 0.29 GiB | 10.23× |
| 32,768 | 5.88 GiB | 0.51 GiB | 11.61× |
| 65,536 | 11.75 GiB | 0.94 GiB | 12.46× |
| 131,072 | 23.50 GiB | 1.82 GiB | 12.92× |