Flux2 nvfp4 quantization correctness on Blackwell (B200) (#23625)

This commit is contained in:
Johnsonms
2026-05-02 09:57:35 +08:00
committed by GitHub
parent bfccc8e504
commit 4c2ed9a254
8 changed files with 128 additions and 42 deletions
+7 -13
View File
@@ -275,19 +275,13 @@ jobs:
artifact-suffix: component-accuracy
multimodal-gen-test-1-b200:
# TODO: Temporarily disabled (`if: false`) — deterministically failing
# with `RobertaProcessing.__new__() got an unexpected keyword argument
# 'cls'` (tokenizer/transformers compatibility) and blocking CI.
# Restore the original `if:` condition (commented below) once the
# underlying fix lands.
if: false
# if: |
# (inputs.target_stage == 'multimodal-gen-test-1-b200') ||
# (
# !inputs.target_stage &&
# ((github.event_name == 'schedule' || inputs.test_parallel_dispatch == 'true') || (inputs.caller_needs_failure != 'true' && !cancelled())) &&
# inputs.multimodal_gen == 'true'
# )
if: |
(inputs.target_stage == 'multimodal-gen-test-1-b200') ||
(
!inputs.target_stage &&
((github.event_name == 'schedule' || inputs.test_parallel_dispatch == 'true') || (inputs.caller_needs_failure != 'true' && !cancelled())) &&
inputs.multimodal_gen == 'true'
)
runs-on: ${{ inputs.b200_runner }}
timeout-minutes: 240
steps: