hf-mem
huggingface800 installs
Size up model memory without downloading weights.
What it does
Estimates the memory needed to run models from the Hugging Face Hub without downloading their weights, optionally including KV cache memory.
When to use it
Use when someone asks about a model's inference memory requirements or whether it fits on a GPU or instance.
How to use it
Give it a Hugging Face model ID; it returns memory estimates using the CLI's JSON output.
What you provide
- a Hugging Face model ID
Uses
Latest review
What you need · 2
Requires uv to be installed so the agent can run uvx.
For gated or private models, authentication uses the HF_TOKEN environment variable or, alternatively, the --hf-token flag.
Access · 2
This skill
Hugging Face Hub
Read
Reads model data from Hugging Face
hf-mem
Execute
Hf-mem through uvx