biondizzle 5faf9916eb fix: UE4M3 activation scales + group_size=16 for NVFP4 mega_moe ...

The mxf4nvf4 MMA instruction shares scale_format_ between SFA and SFB.
For NVFP4 (UE4M3), both activation and weight scales must be UE4M3.

Changes to _stage_deepseek_v4_mega_moe_inputs_kernel:
- GROUP_K=16 (was 32) — NVFP4 scale_vec::4X has group_size=16
- Scale quantization: float → float8_e4m3fn (UE4M3) instead of UE8M0
  exponent extraction (>> 23). Pack 4 UE4M3 bytes per int32.
- FP8 activation quantized against UE4M3 rounded scale

Also updated class docstring (was stale MXFP4 conversion description).

2026-05-11 16:12:36 +00:00

..

deepseek_v4.py

fix: UE4M3 activation scales + group_size=16 for NVFP4 mega_moe

2026-05-11 16:12:36 +00:00

deepseek_v4.py.bak

S11: Fixed substr mapping, stacking, suffix, and o_a_proj - loads weights but attention forward uses FP8 einsum incompatible with NVFP4

2026-05-10 17:45:53 +00:00

deepseek_v4.py.s11

S11: Fixed substr mapping, stacking, suffix, and o_a_proj - loads weights but attention forward uses FP8 einsum incompatible with NVFP4

2026-05-10 17:45:53 +00:00

patch_finegrained_fp8_blackwell.py

Add BF16 upcast script and Blackwell DeepGEMM patch

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

patch_vllm_weights.py

vLLM serving: patched deepseek_v4.py, disabled mega_moe, updated docs

2026-05-10 16:14:17 +00:00

quant_module_patched.py

Add ModelOpt NVFP4 pipeline: patch, run script, README

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