memory-leak

Ollama released v0.32.1 on July 16, 2024, addressing a critical MLX model cache memory leak that increased memory usage across requests. The update also improves Gemma 4 tool calling reliability and adds authentication prompts for agent web search features.

Mistral AI's engineering team documented their investigation of a memory leak in vLLM that caused 400MB/minute memory growth during disaggregated serving with Mistral Medium 3.1. The leak, which only appeared with specific conditions including graph compilation and NIXL-based KV cache transfer, was eventually traced to mmap allocations outside the traditional heap that standard profiling tools couldn't detect.