Skip to content

Topic

Quantization scaling granularity

The choice of how many scale factors a tensor is given when it is quantized: one for the whole tensor, one per channel, or one per smaller group or block.

Current clusters