Quantization format

NVFP4

NVFP4 has a nominal rate of — bits per weight. We have no measured files carrying this label yet.

From the file· 30 files measured
Nominal bpw
—
from the block layout
Measured average
6.849
30 files
Range
4.463–10.637
varies by architecture
File sizes
2.21 GiB+
up to 62.52 GiB

Real files

one per model, most downloaded first
ModelSize●Effective bpw●vs nominal●
Ornith-1.0-9B6.17 GiB5.765—
Krea-2-Turbo7.65 GiB5.123—
Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-Preserved18.30 GiB5.747—
diffusiongemma-26B-A4B-it13.45 GiB4.473—
Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF1621.80 GiB5.333—
Qwen3.6-27B-NVFP418.51 GiB8.754—
Krea-2-Raw9.72 GiB6.513—
Qwen3.6-27B-uncensored-heretic-v217.51 GiB5.499—
Qwen3.6-27B-NVFP421.58 GiB8.731—
gemma-4-26B-A4B-it-qat-q4_0-unquantized-uncensored-heretic16.46 GiB5.480—
gemma-4-12B-it-qat-q4_0-unquantized-uncensored-heretic7.60 GiB5.461—
Ministral-3-14B-Instruct-25127.25 GiB4.463—
Qwen3.6-35B-A3B-NVFP421.10 GiB9.699—
gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic17.99 GiB4.941—
Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP421.32 GiB8.737—
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic17.99 GiB4.941—
gemma-4-12B-it-uncensored-heretic7.60 GiB5.461—
Leanstral-1.5-119B-A6B62.52 GiB4.514—
Qwen3.6-35B-A3B-NVFP4-Fast18.81 GiB7.408—
Nemotron-3-Puzzle-75B-A9B44.97 GiB4.934—
Qwopus3.6-27B-Coder-MTP14.64 GiB——
gemma-4-31B-it-NVFP420.61 GiB8.911—
Qwen3.6-35B-REAP-Pruned-ratio-0.512.48 GiB5.644—
Gemma-4-26B-A4B-NVFP416.46 GiB9.829—
Mage-Flow2.21 GiB4.613—
●From the filewhat these mean

Other quantizations

Q2_KQ3_K_SQ3_K_MQ3_K_LQ4_0Q4_K_SQ4_K_MQ5_K_SQ5_K_MQ6_KQ8_0IQ4_XSIQ3_XXSIQ2_XXSF16BF16F32