"author:nodegrove" matching MCP servers:

Lets AI assistants check whether an open-weight LLM fits a given GPU, reporting the memory split of weights, KV cache and overhead, a single-stream speed ceiling, the longest context that fits, and every change that would make a failing setup work. It also lists what fits on a card, covers any Hugging Face repo read from its config.json, and answers with the assumptions behind each figure.