Gemma-4-31B-it-DFlash
GGUF quantizations of z-lab/gemma-4-31B-it-DFlash.
Converted to BF16 using convert_hf_to_gguf.py, then quantized using llama-quantize from llama.cpp.
Available quants
| Quant | Bits | Size | Notes |
|---|---|---|---|
| Q4_K_M | 4 | 912 MB | Average quality |
| Q5_K | 5 | 1.09 GB | High quality |
| Q6_K | 6 | 1.27 GB | Very high quality |
| Q8_0 | 8 | 1.65 GB | Highest quality, near lossless, Recommended |
| BF16 | 16 | 3.07 GB | Full precision, reference file |
Usage
Use in conjunction with existing Gemma 4 Quants, example config if using llama-server:
[Gemma-4-31B-it-DFlash]
sm = layer
model = /mnt/gguf/Gemma-4-31B-it/Gemma-4-31B-it-Q8_0.gguf
model-draft = /mnt/gguf/Gemma-4-31B-it-DFlash/Gemma-4-31B-it-DFlash-Q8_0.gguf
spec-type = draft-dflash
spec-draft-n-max = 6
(Note: For some reason I cannot get sm = tensor to work, it crashes on launch, pretty sure this is an issue in llama.cpp)
Original model
See the original model card for details on capabilities, benchmarks, and license.