●Change8

Migrating to llama.cpp b10208

Version b10208 introduces 1 breaking change. This guide details how to update your code.

Released: 7/31/2026

1
Breaking Changes
5
Migration Steps
5
Affected Symbols

⚠️ Check Your Code

If you use any of these symbols, you need to read this guide:

fattn-mkl.cppfattn.cppggml-sycl/fattn.cppggml-sycl/fattn-mkl.cppggml_sycl_get_env()

Breaking Changes

●Issue #1

The `mkl_disable` environment variable has been renamed to `mkl_enable` for clarity. Users previously disabling MKL FA by setting `MKL_FA_DISABLE=1` should now set `GGML_SYCL_ENABLE_MKL_FA=0`.

Migration Steps

  1. 1
    Rename environment variable `MKL_FA_DISABLE` to `GGML_SYCL_ENABLE_MKL_FA`. Set to `0` to disable MKL FA.
  2. 2
    Rename environment variable `MKL_FA_DEBUG` to `GGML_SYCL_MKL_FA_DEBUG`.
  3. 3
    Rename environment variable `MKL_FA_DIAG` to `GGML_SYCL_MKL_FA_DIAG`.
  4. 4
    Replace `fprintf(stderr, ...)` and `fflush(stderr)` with the `GGML_LOG_INFO()` macro.
  5. 5
    Use `ggml_sycl_get_env()` instead of raw `getenv()` for checking environment variables related to MKL FA.

Release Summary

This release introduces oneMKL GEMM flash attention for XMX-accelerated prompt processing on SYCL devices, significantly improving performance. It also includes numerous bug fixes and refinements to MKL FA activation and handling across various KV cache types and configurations.

Need More Details?

View the full release notes and all changes for llama.cpp b10208.

View Full Changelog