[Roadmap] SparseML Roadmap Q2 2024

[mgoin]

Short-term Work

Issues or PRs that the NM team are planning to tackle for this quarter:

• Dependencies

  • Forkless Transformers Forkless transformers [4.34.1] #2199
  • Upgrade Transformers to latest version Update transformer version to <4.40 #2204
  • Upgrade PyTorch to latest version upgrade torch support to 2.2 #2247 Allow torch 2.3 and remove torch ceiling version restriction #2259
  • Upgrade Pydantic to v2 Update pydantic support to v2 #2248

• SparseGPT one-shot pruning for Transformers

  • General model support AutoModel Add AutoModel support for SparseAutoModel #2003

• Saving compressed models

  • Export sparse models w/ sparsity_config to Transformers SparseML Compression Pt 3: SparseAutoModel interface & inferring params #2190
  • Export weight-quantized models w/ quant_config to Transformers
  • Export W8A8-quantized models w/ quant_config to Transformers

Long-term Tracked Feature Requests/Issues