openvino icon indicating copy to clipboard operation
openvino copied to clipboard

Support for INT4 decompression patterns from NNCF

Open alexsu52 opened this issue 1 year ago • 3 comments

Details:

  • [torch.compile] Added translation decompression subgraph to u4 in translate_stack_fx
  • [torch.compile] Added capture NNCF INT4 decompression patterns in make_partitions
  • [pytorch frontend] Added ov::frontend::pytorch::pass::U4ConvertReshape transformation to support NNCF INT4 decompression patterns
  • Updated u4_compression_stack transformation to support Torch FX flow.

Tickets:

  • ref: 153918

alexsu52 avatar Oct 15 '24 07:10 alexsu52

@mvafin, @cavusmustafa please take a look at.

alexsu52 avatar Oct 18 '24 07:10 alexsu52

build_jenkins

mvafin avatar Oct 18 '24 09:10 mvafin

Please add test

Tests were added.

alexsu52 avatar Oct 20 '24 14:10 alexsu52