Chenjie LuoandClaude Opus 5.5 3947a86045 Report packer input errors in the original order ...

check_scaled_pack_inputs checked that the block scales were on CUDA before
check_pack_inputs looked at the input and grid, so a call with several bad
arguments named the scales first. The pybind wrappers it replaced checked
the input, grid and scales devices in that order, then everything else.

check_pack_inputs now takes the scales optionally and checks their device
right after the grid's, which restores that order. Over 5 formats and 8
combinations of bad arguments, every first error now matches main's.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Signed-off-by: Chenjie Luo <chenjiel@nvidia.com>

2026-10-01 06:36:28 +00:00

..

deploy

fix(llm_eval): repair test_qwen3_eval_fp8 end-to-end (#1650)

2026-06-08 18:11:51 +00:00

onnx

[5565357] Fix SDXL NVFP4 export and performance (#2336)

2026-09-18 17:22:28 +00:00

recipe

[OMNIML-5570] 2/2 Compose GEMM and KV-cache AutoQuant workflows (#2273)

2026-09-29 00:15:40 +00:00

torch

Report packer input errors in the original order

2026-10-01 06:36:28 +00:00

__init__.py

Add experimental support for transformers>=5.0 + min torch 2.8 (#975)

2026-04-09 09:59:37 +05:30