Skip to content

Releases: flagos-ai/FlagOS-Compressor

FlagOS-Compressor v0.1.0

Choose a tag to compare

@wbavon wbavon released this 28 Sep 09:53
eecf2fa

FlagOS-Compressor v0.1.0

Part of FlagOS 2.2.

Initial release in FlagOS.

5 pull requests and 1 additional commits.

Source: v0.1.0, commit eecf2faed14d.

New Features

  • feat: bootstrap FlagCompressor — safetensors weight-format conversion with BF16 dequant and policy-driven INT4 quantization (#1) by @rdzhu225
  • feat: support fused MoE INT4 packing and compressed-tensors output (#2) by @rdzhu225
  • feat(quantization): add INT8 weight quantization (#3) by @rdzhu225
  • feat(quantization): add dynamic-token W8A8 MoE export (#5) by @rdzhu225

Improvements

  • refactor(cli): log operational summaries (#4) by @rdzhu225

Other

Contributors

Thanks to the contributors to the pull requests included in this release.