{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreifdxqyuo5pxicdzcql42spupgzbjs4qolbcckxyzbh4eclnxz3ti4",
"uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mpeoah2non32"
},
"path": "/t/removing-the-embedding-from-my-embedding-a-byte-transformer-with-a-0-parameter-input-layer-25m-single-rtx-4070/176731#post_6",
"publishedAt": "2026-06-28T18:28:22.000Z",
"site": "https://discuss.huggingface.co",
"tags": [
"New Post"
],
"textContent": "Update: I built two speech models on top of this 0-param substrate — tokenizer-free\n\nSTT + TTS (HoLo-ToLk). Short version: substrate + a spectral lens beats a mel baseline\n\non STT (CER 0.194 vs 0.213, multi-seed); TTS is a tokenizer-free single-speaker\n\nfeasibility demo. Full writeup + combined demo here: [New Post]",
"title": "Removing the embedding from my embedding: a byte transformer with a 0-parameter input layer (25M, single RTX 4070)"
}