Change8

b10604

📦 llama-cppView on GitHub →
6 features🐛 1 fixes🔧 2 symbols

Summary

This release introduces support for Deepseek 4 with '-sm tensor' and includes various improvements such as coarser head split granularity and enhanced model saving capabilities. It also addresses a bug in dspark.

✨ New Features

  • Added support for Deepseek 4 with '-sm tensor' option.
  • Introduced coarser granularity for head splits.
  • Added model saving for DSV4.
  • Allowed dflash to return on specific devices.
  • Added comment about dsv4 seq_rm.
  • Added shared expert delayed allreduce.

🐛 Bug Fixes

  • Fixed dspark.

Affected Symbols