whichllm

Verdict: whichllm is worth trying if your main problem is not “can this model fit?” but “which local model is actually the best trade-off for my machine?” It is not an inference runtime, chat app, or fully automatic optimizer. A better analogy is a command-line local-LLM selection advisor that combines hardware detection, public benchmark signals, and practical ranking rules.

Its practical value is that it checks GPU, CPU, and memory, then ranks candidate models by compatibility, estimated speed, and multiple benchmark sources instead of only showing the largest model that fits in VRAM. The available posts repeatedly highlight that distinction, and some examples mention use-case-based recommendations and simulated GPU selection. For “proof of attention,” the X reposts and media-style shares show strong interest. For “proof of usefulness,” the stronger evidence comes from walkthrough-style Zhihu/Qiita articles with sample outputs, though the sample size is still limited.

On cost and difficulty, the evidence only supports that this is an open-source CLI aimed at people already comfortable with local LLM concepts like VRAM, quantization formats, and Hugging Face model naming.