b8746

llama.cpp release b8746 was published on April 10, 2026 (commit 0893f50) and consists of a single change: marking the --split-mode tensor option as experimental in the --help output (PR #21684). This was a follow-up fix to a prior pull request that introduced tensor parallelism support but omitted the experimental label from the help text. The --split-mode tensor feature enables tensor-level splitting across multiple GPUs and is currently only fully functional with the CUDA backend.

b8747
b8744
b8740
b8742
b8741
b8748

Source: local-ai

Related