Free LLM Server Memory, Measured: A 20-Run Context-Bleed Probe

Описание и применение

Проверка памяти бесплатного LLM‑сервера: 20‑разовый тест выявил отсутствие сохранения контекста, что позволяет безопасно использовать сервер как статeless‑прокси.

Задача

Determine whether a free LLM server retains context between independent requests, to inform secure integration design.

Результат

The 20‑run probe consistently showed isolated responses, indicating the free server does not retain context between calls. This informed a stateless integration strategy.

Рабочий процесс

• Send a request asking the server to remember a unique marker.
• Send a second request immediately afterward asking for that marker, with no history or reference to the first.
• Check if the marker appears in the second response.
• Repeat the pair 20 times with fresh markers and a short delay.
• Record and analyze the pattern of leaks.
• Decide on trust posture based on the pattern.
• Apply safe usage rules: treat server as stateless, include all context in each request, avoid secrets in prompts, and verify outputs with assertions.

Доказательства

The twenty runs produce one of three patterns, and each pattern implies a different trust posture: All runs isolated → Stateless endpoint.

Источник

The harness measures one endpoint, over one network path, on one day, and it says nothing about output quality or speed.

Источник