implementation detail · filed under inference & serving
Mixed-precision INT4/INT8 layer-wise quantization
Assigns INT4 to the first 30 layers and INT8 to the final 10 layers in the cited setup.
Also called mixed-precision quantization strategy.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
We adopted a mixed-precision quantization strategy: the first 30 layers were quantized to INT4, while the final 10 layers were quantized to INT8 with 1 × 128 group-based weight gl in-smoking.
usedpost trainingin Laguna XS.2Poolside
Filed alongside
Other methods under inference & serving :: inference quantization.
FP8 KV-cache quantizationFP8FP4 KV-cache quantizationNVFP4 quantizationFour-Over-SixMXFP4 weight quantizationPost-training quantizationQuantizationBlock-scaled INT8 quantization with stochastic roundingFP8 E4M3 quantizationGGUFMax-based scalingMSE-based scalingMXFP8 activation quantizationNVFP4 KV-cache quantizationNVFP4 ModelOpt re-quantizationSSM cache quantizationAWQ INT4 weight quantization (W4A16)BF16 inferenceBlock-wise E4M3 FP8 weight quantizationChannel-wise quantizationDynamic activation scalingEmbedding and KV-cache quantizationEmpirical bits-per-element budget selection