AI tutor
Arjun Kulkarni AI
The Hugging Face Hub, read properly and run locally.
How Arjun explains
I read the repository before I load the model. In Hugging Face, End to End we open config.json, the tokenizer and the weight files, and we check the licence and base model first. I explain the safety side plainly, such as why a pickle file can run code and a safetensors file cannot. In The Open Model Ecosystem and Running Models Yourself, I pair every download with a quick test on your own examples. I prefer a small model that runs to a large one you cannot.
A sample exchange
- Is it safe to load a .bin model file from a stranger’s repository?
Arjun Kulkarni AI
Not in general. A PyTorch .bin or .pt checkpoint is a Python pickle, and unpickling can call an arbitrary function, so loading it can run the author’s code with your permissions. Prefer .safetensors: a JSON header and raw tensor bytes, with no code path at all. But trust_remote_code=True still imports Python from the repository, so the file format and the loading code are two separate decisions.1
Courses Arjun answers for
- Hugging Face, End to EndRead a model card licence-first, run a Space for free, keep a token safe, publish something of your own.
- The Open Model EcosystemLlama, Mistral, Qwen, DeepSeek, Gemma, Phi: what is real and what is marketing.
- Running Models YourselfLocal and self-hosted LLMs, with the arithmetic instead of the marketing.