hf_mem estimates the required memory for inference, including model weights and an optional KV cache, for Safetensors and GGUF for models on the Hugging Face Hub using HTTP Range requests i.e., without downloading or loading any weights locally.
uv installed (for uvx)HF_TOKEN from the host's private credential handling for gated/private repositories; avoid putting tokens in command arguments, logs or reports.hf-mem package version and permission to access the requested model metadata. This skill does not bundle the executable.The angle-bracket and square-bracket forms below are usage notation, not shell commands to paste literally. Quote actual model IDs and file paths. Inspect the selected package's --help, choose a reviewed version (for example with uvx --from "hf-mem==REVIEWED_VERSION" hf-mem ...) and use a private cache where metadata is sensitive. Do not print credentials.
Run with --model-id pointing to the Hugging Face Hub repository which will check that it either contains Safetensors (via model.safetensors, model.safetensors.index.json if sharded, or model_index.json for Diffusers) or GGUF model weights within.
uvx hf-mem --model-id <model-id> --json-output
If the repository contains GGUF model weights in multiple precisions / quantizations, the estimations will be on a per-file basis, whereas for inference you won't load all of those but rather only a single precision. This being said, for GGUF you might as well need to provide --gguf-file to target the specific file (or path if sharded) you want to run.
uvx hf-mem --model-id <model-id> --gguf-file <file-or-path> --json-output
Additionally, hf-mem comes with an --experimental flag that will also calculate the KV cache memory requirements too, useful for large-language models, meaning it applies to LLMs (...ForCausalLM), VLMs (...ForConditionalGeneration), and GGUF models.
As per the context window, it will be read from the default or overridden with --max-model-len a la vLLM. And, same goes for the KV cache precision, which will default to the model precision unless manually set via --kv-cache-dtype a la vLLM too.
For Safetensors use as:
uvx hf-mem --model-id <model-id> --experimental [--max-model-len N] [--batch-size N] [--kv-cache-dtype auto|bfloat16|fp8|fp8_ds_mla|fp8_e4m3|fp8_e5m2|fp8_inc] --json-output
And, for GGUF use as:
uvx hf-mem --model-id <model-id> --gguf-file <file-or-path> --experimental [--max-model-len N] [--batch-size N] [--kv-cache-dtype auto|F32|F16|Q4_0|Q4_1|Q5_0|Q5_1|Q8_0|Q8_1|Q2_K|Q3_K|Q4_K|Q5_K|Q6_K|Q8_K|IQ2_XXS|IQ2_XS|IQ3_XXS|IQ1_S|IQ4_NL|IQ3_S|IQ2_S|IQ4_XS|I8|I16|I32|I64|F64|IQ1_M|BF16|TQ1_0|TQ2_0|MXFP4] --json-output
For Transformers with Safetensors weights:
uvx hf-mem --model-id MiniMaxAI/MiniMax-M2 --json-output
For Diffusers with Safetensors weights:
uvx hf-mem --model-id Qwen/Qwen-Image --json-output
For Sentence Transformers with Safetensors weights:
uvx hf-mem --model-id google/embeddinggemma-300m --json-output
With --experimental to include the KV cache estimation for LLMs and VLMs:
uvx hf-mem --model-id mistralai/Mistral-7B-v0.1 --experimental --json-output
And, for LLMs or VLMs with GGUF weights:
uvx hf-mem --model-id unsloth/Qwen3.5-397B-A17B-GGUF --gguf-file Q4_K_M --experimental --json-output
Report the model revision, selected weight file or quantization, batch size, context length and KV-cache dtype alongside the estimate. Weight/header estimates are not a guarantee that a model fits: runtime workspaces, activations, allocator fragmentation, offloading and concurrent processes add memory. Confirm with the intended inference backend on target hardware before making a purchase or deployment decision. HTTP metadata access still contacts the Hub and can reveal private model identifiers; it does not run inference. Experimental KV estimates depend on supported architectures.