biondizzle b70a04696e Add resume capability to dequant script (skip already-done shards) ...

Verified our FP4 dequant is byte-identical to official transformers
MXFP4 implementation. Max diff = 0.0 across all values.

2026-05-08 02:58:24 +00:00

..

dequant_fp8_to_bf16.py

Add resume capability to dequant script (skip already-done shards)

2026-05-08 02:58:24 +00:00

model_opt_nvfp4_experts_only.py

Update nvfp4_experts_only to use dequantized BF16 model

2026-05-07 16:34:37 +00:00

model_opt_nvfp4_full.py

Purge INT4 references — expert weights are FP4 (E2M1), not INT4

2026-05-08 02:33:46 +00:00

run_modelopt_nvfp4.sh

Add ModelOpt NVFP4 pipeline: patch, run script, README

2026-05-07 07:22:54 +00:00

upcast_to_bf16.py

Add BF16 upcast script and Blackwell DeepGEMM patch

2026-05-07 14:25:30 +00:00