implementation detail · filed under inference & serving
In-flight block-wise FP8 weight quantization
Re-quantizes weights to FP8 during inference weight synchronization, computing per-block scales on the fly without calibration.
Also called in-flight block-wise weight quantization.
- source
- 1
- model
- 1
- labs adopt it
- 0
- strongest
- not used
How sources treat it
One count per evidence span, weakest treatment to strongest.
not used 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
we quantized the inference-audit weights to FP8 with an in-flight block-wise copying: weights were re-quantized on the inference replica during each weight sync, with per-block scales computed on the fly without any calibration data.
not usedinference servingin LAGUNA M.1/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