April 2026 (Month: 2026-04) — Sglang stability and memory safety improvements. Implemented a targeted memory management fix to cap effective_mamba_size to the maximum allowable size based on mamba_size and pre_alloc_size, reducing overflow risk and ensuring predictable memory usage across configurations. This change enhances reliability for deployments with varying memory budgets and contributes to safer production behavior across ping1jing2/sglang.

Concise monthly summary for 2026-01 focused on kvcache-ai/sglang. Delivered a stability-focused bug fix to request logging during prefill warmup, addressing a TypeError in RequestLogger when input_ids is a list of lists. Ensured logging does not crash and improved robustness during prefill node warmup, enabling reliable operation at log-requests-level >= 2 and enhancing observability for the ML inference pipeline.