ggml/llama.cpp releases: b11026

TL;DR

The ggml/llama.cpp project released version b11026, which includes a model update to skip gate_up_exps if TENSOR_SKIP is set. This update is crucial for qwen35moe when MTP tensors are fused but not loaded, ensuring compatibility and functionality across different configurations.

Detailed Summary

The ggml/llama.cpp project released a new version (b11026) that includes changes to skip certain model operations if the TENSOR_SKIP flag is set. This update is crucial for supporting the qwen35moe model when specific MTP tensors are fused but not fully loaded. The broader impact involves enhancing flexibility and compatibility in model deployment, particularly on macOS/iOS systems.

Key Points

• • model skips gate_up_exps if TENSOR_SKIP is set
• • required for qwen35moe with MTP tensors fused but not loaded
• • website: <https://llama.app>