LFM2

This release fixes a bug where LFM2/LFM2.5 template handlers ignored json_schema for response formatting. It also provides numerous pre-compiled binaries for various operating systems and hardware configurations.

This release fixes missing tensors for LFM2 models and announces a transition for Linux release artifacts from .zip to .tar.gz archives.

v0.18.0 introduces major features like gRPC serving, GPU-less render serving, and significant improvements to KV cache offloading and Elastic Expert Parallelism. Ray is now an optional dependency, and numerous model-specific fixes and kernel optimizations have been integrated.

This release focuses on reducing library bloat and increasing speed through refactored Llama models, megablocks kernel integration, and native distributed training. It also introduces several new model architectures including Ernie 4.5, Voxtral, DeepSeek-V2, and LFM2.