Nf4.rar Site

: An information-theoretically optimal data type for normally distributed weights. It uses 16 quantization levels based on the quantiles of a standard normal distribution.

: A feature to handle memory spikes during training by offloading to CPU RAM. 🔬 Key Technical Details NF4.rar

: A process that quantizes the quantization constants themselves to save additional memory. 🔬 Key Technical Details : A process that

: Compresses 16-bit weights to 4 bits, effectively reducing VRAM usage by ~75% (e.g., a 65B parameter model can be loaded with ~35GB instead of ~130GB). The paper explains why NF4 is superior to

💡 : If you are looking for the software/machine learning paper, search for "QLoRA" or "4-bit NormalFloat" on arXiv .

The paper explains why NF4 is superior to standard 4-bit integers (Int4) or floating-point (Float4) formats:

: RNF4 mediates the degradation of the PML-RARα fusion protein.