feat(quantization): support compressed-tensors W8A8 and mixed modules - #582

Open

sam0336 wants to merge 8 commits into

InfiniTensor:InfiniLM-v0.2.9cfrom

sam0336:feat/compressed-tensors-w8a8

Commits on Sep 18, 2026

feat(quantization): parse compressed-tensors configs
sam0336
committed
be55e01
feat(quantization): add module target matching
sam0336
committed
d6bc912
feat(quantization): resolve compressed-tensors config groups
sam0336
committed
9260ff0
feat(quantization): select compressed-tensors schemes per module
sam0336
committed
d436587
feat(models): resolve Qwen3 attention quantization by module
sam0336
committed
112095c
fix(quantization): shard W8A8 output parameters
sam0336
committed
79016fd
fix(quantization): normalize W8A8 scales to float32
sam0336
committed
ccabee5
test(quantization): cover real Qwen3 W8A8 config
sam0336
committed
29ddfb9