Skip to content

Quant compat0 - #1581

Open
mi804 wants to merge 12 commits into
modelscope:mainfrom
mi804:quant-compat0
Open

Quant compat0#1581
mi804 wants to merge 12 commits into
modelscope:mainfrom
mi804:quant-compat0

Conversation

@mi804

@mi804 mi804 commented Aug 11, 2026

Copy link
Copy Markdown
Collaborator
  1. Fix torchao backend with low vram management.
  2. Fix ideogram-4with diskoffload
  3. Add lowvram inference and training support for ideogram-4
  4. Fix mixed config quant.
  5. Fix Zimage diskoffload

mi804 added 12 commits August 10, 2026 15:56
…on fix

Adds the QuantBackend contract (quantized_linear_classes, checkpoint_key_patterns,
resolve_checkpoint_keys, check_backend_contract, build_quantized_shell) and wires it
through QuantizeConfig/MixedQuantizeConfig and VRAM management; refactors torchao and
ideogram4_fp8 onto it. Fixes torchao unflatten crashing on root-level (dotless) tensors.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant