Quantization format
MXFP4
MXFP4 has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 13 files measured
Nominal bpw
—
from the block layout
Measured average
4.761
13 files
Range
4.113–8.501
varies by architecture
File sizes
2.84 GiB+
up to 478.42 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| DeepSeek-V4-Flash | 145.29 GiB | 4.290 | — |
| gpt-oss-20b | 11.28 GiB | 4.503 | — |
| Krea-2-Turbo | 7.34 GiB | 4.920 | — |
| gpt-oss-120b | 59.03 GiB | 4.211 | — |
| Inkling | 478.42 GiB | 4.315 | — |
| Huihui-Qwen3-VL-4B-Instruct-abliterated | 2.84 GiB | 5.496 | — |
| Apertus-70B-Instruct-2509 | 69.87 GiB | 8.501 | — |
| gpt-oss-safeguard-20b | 11.28 GiB | 4.503 | — |
| gpt-oss-safeguard-120b | 59.03 GiB | 4.211 | — |
| DeepSeek-V4-Flash-0731 | 145.64 GiB | 4.113 | — |
●From the filewhat these mean