🚀 RELEASES & LAUNCHES
- ExLlamaV3 adds CPU offload and support for GLM-5.3 and Qwen3.8 Flash
ExLlamaV3 added memory-saving offload features and support for new Flash models.- The updates target NVIDIA GPU users running large language models locally.
- ExLlamaV3 added CPU offload for MoE experts and n-gram disk offload for Qwen3.8-Flash-Next.
- Support was added for GLM-5.3-Flash and Qwen3.8-Flash-Next, alongside self-calibrated optimization and other performance improvements.
- The changes let builders run larger EXL3-quantized models on consumer hardware with reduced GPU-memory requirements.