HologramHologram

Start

Quickstart

Find a model, download one file, and prove the bytes, in three requests and no key.

Markdown

You want a file from a model and you want to know it is the right file. Three requests do it. Nothing to install, nothing to sign up for.

Every command on this page runs as written. hexgrad/Kokoro-82M is a real model in the index; swap in any other <owner>/<name> the search returns.

1. Find a model

curl -s "https://gethologram.ai/api/models?search=kokoro&limit=2"

[
  {
    "modelId": "hexgrad/Kokoro-82M",
    "pipeline_tag": "text-to-speech",
    "downloads": 11804278,
    "hologram": {
      "manifest": "blake3:2d184ec4d22e1ccab14342f3020c2838788e42186877669ccb1c7d56749c7a4a",
      "files": 72,
      "weight_bytes": 361048692,
      "sources": ["Hugging Face", "ModelScope", "P2P", "IPFS"]
    }
  },
  …
]

Truncated to the fields that matter: each row is Hugging Face's list shape plus a hologram block that says how many files the model has, how big its weights are, and where the bytes live today.

2. List its files

curl -s https://gethologram.ai/api/models/hexgrad/Kokoro-82M/tree/main

[
  { "type": "file", "oid": "9b4d7a54809bf22127d19d936d9e749249e1a655e2e0a1df8a6546f77a819658", "size": 534, "path": "EVAL.md" },
  { "type": "file", "oid": "91dcabced89db6f109b8786642f50402d3ee87450e8189589b6f85520e7f4d78", "size": 6348, "path": "README.md" },
  …
]

oid is the SHA-256 the file's bytes must have. Keep it: it is the only hash you will trust.

3. Fetch the file

curl -sL https://gethologram.ai/hexgrad/Kokoro-82M/resolve/main/EVAL.md -o EVAL.md

Without -L you see what the hub actually does: a 302 to a source that was up a moment ago. The weights never pass through the hub.

HTTP/1.1 302 Found
Location: https://huggingface.co/hexgrad/Kokoro-82M/resolve/f3ff3571791e39611d31c381e3a41a3af07b4987/EVAL.md
X-Hub-Source: huggingface.co
Etag: "9b4d7a54809bf22127d19d936d9e749249e1a655e2e0a1df8a6546f77a819658"

4. Prove it

echo "9b4d7a54809bf22127d19d936d9e749249e1a655e2e0a1df8a6546f77a819658  EVAL.md" | sha256sum -c

EVAL.md: OK

The hash on the left came from step 2, the index. The bytes came from step 3, a source. They agree, so the file is what the index says it is, and the source never had a chance to say otherwise. That is the whole product. The Verification page has the rule in full.

The same thing with one variable

Every tool built on huggingface_hub reads from the hub once you set one environment variable. Same commands, same cache.

export HF_ENDPOINT=https://gethologram.ai
hf download hexgrad/Kokoro-82M EVAL.md

transformers, diffusers, sentence-transformers, vLLM and SGLang all go through that variable; llama.cpp reads MODEL_ENDPOINT instead. The Hugging Face tools page lists each one.

Next