7 ms·

This isn't quantized, right? Just a smaller context?

• DSingularity 2mo ago
  Its 256k context window. Quantization is orthogonal. We cant really tell directly so it could be quantized.
• gpugreg 2mo ago
  The model is already natively MXFP4-quantized during training, so there is no quality loss.