Estimate VRAM and parameter footprint for Safetensors-based Hugging Face models.
npx skills add https://github.com/alvarobartt/hf-mem --skill hf-mem
Please help me install this Agent Skill. Skill: hf-mem Source: https://github.com/alvarobartt/hf-mem/tree/main Command: npx skills add https://github.com/alvarobartt/hf-mem --skill hf-mem
hf-mem provides estimates of the VRAM and parameter footprint required to load Safetensors-based Hugging Face models, helping you plan deployment and capacity.
Use the hf-mem CLI to query a model, e.g.: uvx hf-mem --model-id <org/model-name> uvx hf-mem --model-id <org/model-name> --experimental
High-intent search queries and answers about installing and using this skill.
Estimating VRAM requirements for Hugging Face models involves fetching Safetensors metadata from model.safetensors or model_index.json to aggregate per-component and total memory usage across Transformers, Diffusers, and Sentence Transformers architectures.
You can calculate KV cache memory for causal and multi-modal models by enabling the experimental flag, which uses max_model_len, batch_size, and kv_cache_dtype to estimate the additional cache size required during inference.
VRAM estimation supports gated and private repositories by authenticating Safetensors metadata fetch requests with the HF_TOKEN environment variable to access restricted Hugging Face Hub models.
Checking if a Diffusers model fits available GPU memory requires aggregating Safetensors metadata from model_index.json to calculate total parameter footprint and per-component memory usage for deployment capacity planning.
Getting JSON output for Hugging Face model memory estimation involves running the CLI command, which returns per-component memory details and optional KV cache metrics in structured JSON format alongside a readable table view.
Tyler-R-KendrickEstimate VRAM and memory requirements for Safetensors and GGUF models from the Hugging Face Hub.
lucifertrjDiagnose Qdrant capacity and guide horizontal scaling decisions.
CHENyiru3Optimize distributed deep learning training with DeepSpeed using ZeRO and mixed precision.
huggingfaceEstimate inference memory for Safetensors and GGUF models from Hugging Face Hub.
