Gemma 4 12B
Gemma 4 12B
Architecture & parameters
| Parameters (total) | 12.0B |
|---|---|
| Parameters (active) | 12.0B |
| Architecture | gemma4_unified |
| Layers | 48 |
| KV heads | 8 |
| Native context | 262144 |
Artifacts
| File | Publisher | Format | Quant | Size | Link | Evidence |
|---|---|---|---|---|---|---|
gemma-4-12b-it-Q5_K_M.gguf | Unsloth | gguf | Q5_K_M | 7.84 GiB | source | SOURCE |
gemma-4-12b-it-IQ4_XS.gguf | Unsloth | gguf | IQ4_XS | 5.94 GiB | source | SOURCE |
VRAM (weights-only)
Not a full fit result. This classification considers artifact weights only. It does not include KV cache, runtime buffers, backend allocations, multimodal projectors, concurrency, or safety reserve.
| Component | Value | Evidence |
|---|---|---|
| Artifact weights | 7.84 GiB | SOURCE |
| KV cache | Not modeled | NOT KNOWN YET |
| Runtime buffers | Not modeled | NOT KNOWN YET |
| Safety reserve | Not modeled | NOT KNOWN YET |
| Weights-only memory class | 8gb+ | SOURCE POLICY |
Runtime support
Sources
- Hugging Face: https://huggingface.co/google/gemma-4-12B-it
- Official: https://ai.google.dev/gemma/docs/gemma_4_license
- Official: https://huggingface.co/google/gemma-4-12B-it