gemma-3-12b-it-ultra-uncensored-heretic
llmfan46/gemma-3-12b-it-ultra-uncensored-hereticgemma-3-12b-it-ultra-uncensored-heretic at Q4_K_M is exactly 7,300,780,096 bytes (6.80 GiB / 7.30 GB) — an effective 4.792 bits per weight, not the nominal 4. Its KV cache at 32K is 2.47 GiB, not the 12.00 GiB a flat formula predicts.
Shipped quantizations
| Quant | Size● | Exact bytes● | Effective bpw● | Tensors● | Publisher |
|---|---|---|---|---|---|
| Q2_K | 4.44 GiB | 4,768,223,296 | 3.130 | — | mradermacher |
| IQ3_XXS | 4.46 GiB | 4,784,901,632 | 3.141 | — | llmfan46 |
| IQ3_XS | 4.85 GiB | 5,206,166,912 | 3.417 | — | llmfan46 |
| IQ3_S | 5.08 GiB | 5,458,316,672 | 3.583 | — | llmfan46 |
| Q3_K_S | 5.08 GiB | 5,458,317,376 | 3.583 | — | mradermacher |
| IQ3_M | 5.27 GiB | 5,655,723,392 | 3.712 | — | llmfan46 |
| Q3_K_M | 5.60 GiB | 6,008,819,776 | 3.944 | — | mradermacher |
| Q3_K_L | 6.04 GiB | 6,480,187,456 | 4.254 | — | mradermacher |
| IQ4_XS | 6.10 GiB | 6,550,965,632 | 4.300 | — | llmfan46 |
| IQ4_XS | 6.15 GiB | 6,606,262,336 | 4.337 | — | mradermacher |
| IQ4_NL | 6.41 GiB | 6,887,165,312 | 4.521 | — | llmfan46 |
| Q4_K_S | 6.46 GiB | 6,935,334,976 | 4.553 | — | mradermacher |
| Q4_K_M | 6.80 GiB | 7,300,780,096 | 4.792 | — | mradermacher |
| Q5_K_S | 7.67 GiB | 8,231,963,776 | 5.404 | — | llmfan46 |
| Q5_K_S | 7.67 GiB | 8,231,964,736 | 5.404 | — | mradermacher |
| Q5_K_M | 7.87 GiB | 8,445,037,696 | 5.543 | — | llmfan46 |
| Q5_K_M | 7.87 GiB | 8,445,038,656 | 5.543 | — | mradermacher |
| Q6_K | 9.00 GiB | 9,660,812,416 | 6.341 | — | llmfan46 |
| Q6_K | 9.00 GiB | 9,660,813,376 | 6.341 | — | mradermacher |
| Q8_0 | 11.65 GiB | 12,510,213,376 | 8.212 | — | llmfan46 |
| Q8_0 | 11.65 GiB | 12,510,214,336 | 8.212 | — | mradermacher |
KV cache by context
| Context | KV cache (f16)● | Flat formula | Overstated by | Full / windowed / recurrent |
|---|---|---|---|---|
| 4,096 | 0.72 GiB | 1.50 GiB | 2.09× | 8 / 40 / 0 |
| 8,192 | 0.97 GiB | 3.00 GiB | 3.10× | 8 / 40 / 0 |
| 16,384 | 1.47 GiB | 6.00 GiB | 4.09× | 8 / 40 / 0 |
| 32,768 | 2.47 GiB | 12.00 GiB | 4.86× | 8 / 40 / 0 |
| 65,536 | 4.47 GiB | 24.00 GiB | 5.37× | 8 / 40 / 0 |
| 131,072 | 8.47 GiB | 48.00 GiB | 5.67× | 8 / 40 / 0 |
40 of 48 layers cache only a 1,024-token window rather than the full context, on a period of 6. Figures assume the default configuration; --swa-full disables the saving entirely.
Compare with
Will it run on your card?
Why other calculators give a different number
A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 6.38 GiB. The real file is 6.80 GiB, because a quantization is a mixture and some tensors are always kept at higher precision. The larger discrepancy is the cache: a flat formula gives 12.00 GiB at 32K context where the real figure is 2.47 GiB, because most of this model's layers cache a fixed window rather than the whole context.
Architecture
Questions people ask
- How much VRAM does gemma-3-12b-it-ultra-uncensored-heretic need?
- Q4_K_M is exactly 7,300,780,096 bytes (6.80 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
- How large is gemma-3-12b-it-ultra-uncensored-heretic's KV cache?
- 2.47 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
- Which quantization of gemma-3-12b-it-ultra-uncensored-heretic should I use?
- Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.