This is pretty similar to how most classification models wor...

52b4a076bcbbbdc3...

npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkj

hex

3bd28dd759a6fd0c4f6324fb004cd466c8b96d19b682420008bb2cd65e8d6a53

nevent

nevent1qqsrh55d6av6dlgvfa3jf7cqfn2xdj9ed5vmdqjzqqytktxkt6xk55cprpmhxue69uhhyetvv9ujuem4d36kwatvw5hx6mm9qgs99d9qw67th0wr5xh05de4s9k0wjvnkxudkgptq8yg83vtulad30gte6mn8

Kind-1 (TextNote)

2026-04-23T20:20:44Z

↳ Reply to Event not found

bb2b59a5ec8683b22037fe5353dc7511e026ae150c669c4fda770da727da2aee...

This is pretty similar to how most classification models work anyway.

Most embedding models, and base models for classification like BERT share the same bidirectional transformer architecture.

With an embedding you use the hidden state (or with a final downproj) on a single token or all tokens averaged.

With BERT and similar you have to add the classification head yourself that downprojects from hidden state.

There you also have the advantage of training the whole model, but close enough concept.

Raw JSON

{
  "kind": 1,
  "id": "3bd28dd759a6fd0c4f6324fb004cd466c8b96d19b682420008bb2cd65e8d6a53",
  "pubkey": "52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd",
  "created_at": 1776975644,
  "tags": [
    [
      "e",
      "bb2b59a5ec8683b22037fe5353dc7511e026ae150c669c4fda770da727da2aee",
      "wss://relay.damus.io",
      "root",
      "5cdbde0a550fc046a38d75e9fc238094f96b0165b8297c4fa69134ae4ec80024"
    ],
    [
      "p",
      "5cdbde0a550fc046a38d75e9fc238094f96b0165b8297c4fa69134ae4ec80024"
    ]
  ],
  "content": "This is pretty similar to how most classification models work anyway.\n\nMost embedding models, and base models for classification like BERT share the same bidirectional transformer architecture.\n\nWith an embedding you use the hidden state (or with a final downproj) on a single token or all tokens averaged.\n\nWith BERT and similar you have to add the classification head yourself that downprojects from hidden state. \n\nThere you also have the advantage of training the whole model, but close enough concept.",
  "sig": "31e851465d5b954621aa64c87db4008e700b0a20f2fb3080f3b0e849a878c47812e4f1007403775d8f983a9fc74b3f58c5ef9abe572fdce5a776a3df035eb247"
}