Method init

tools/convert/quant/quant.py:129–131 · view source on GitHub ↗

(self, weight)

Source from the content-addressed store, hash-verified

127	w = w.cuda().to(torch.bfloat16)
128	w_q, scales = scaled_mxfp4_quant(w)
129	w_q, scales = w_q.to(device), scales.to(device)
130	return w_q, scales, self.extra
131
132
133	@CONVERT_WEIGHT_REGISTER("mxfp6")
134	class QuantWeightMxFP6(QuantTemplate):

nothing calls this directly

__init__Method · 0.45

no test coverage detected