Add MXFP8/NVFP4 quantization, quantized model init, collator state, a…#1572

Merged

savitha-eng merged 10 commits into

mainfrom

savitha/lingua-7b-fp8-clean-pr

May 19, 2026

Commits on May 12, 2026

Add MXFP8/NVFP4 quantization, quantized model init, collator state, and 70B support

savitha-eng
and
claude
committed
Rename use_fp32_master_weights_fused to use_fp32_master_weights and remove dead code

savitha-eng
and
claude
committed

Commits on May 13, 2026

Revert TokenPackingDataset state_dict/load_state_dict
savitha-eng
committed
Clean up hydra configs: rename 7b to 8b, remove experiment configs, restore pytest
savitha-eng
committed

Commits on May 14, 2026