Yesterday it was 48% with 57k RLHF pairs. Today it's 50% with zero. This is notable because the previous paper tried using less data and found it was necessary:
"Training data from math train 7.5k to Open-Reasoner-
Zero 57k, we observe a consistent increase in both training reward and response length for training and evaluation set, indicating that data scale plays a crucial role in training performance."
Leading to my conclusion that for zero pairs, the previous record was close to 0%. Maybe this isn't strictly true, but I expect it to be more predictive than seeing a 2% change
ynniv /
npub12a…x6fdh
2025-05-12 20:30:20
in reply to nevent1q…cwuf
Author Public Key
npub12akj8hpakgzk6gygf9rzlm343nulpue3pgkx8jmvyeayh86cfrus4x6fdhPublished at
2025-05-12 20:30:20Event JSON
{
"id": "d95c381db5fff384336216eb9b12c4fa09a08d67e69ea760ec4cdf1d35750d91",
"pubkey": "576d23dc3db2056d208849462fee358cf9f0f3310a2c63cb6c267a4b9f5848f9",
"created_at": 1747081820,
"kind": 1,
"tags": [
[
"e",
"23182d6a78752d9428cd77052fe666b28ac3d14b06497823d2781411c0ae5996",
"wss://feeds.nostr.band/bens",
"root"
],
[
"e",
"48dcc3c499f926ecb9a402ab923be2fad70ce3cc23d70351f1497038c183df83",
"wss://feeds.nostr.band/bens",
"reply"
],
[
"p",
"576d23dc3db2056d208849462fee358cf9f0f3310a2c63cb6c267a4b9f5848f9",
"",
"mention"
],
[
"p",
"82341f882b6eabcd2ba7f1ef90aad961cf074af15b9ef44a09f9d2a8fbfbe6a2",
"",
"mention"
],
[
"p",
"da18e9860040f3bf493876fc16b1a912ae5a6f6fa8d5159c3de2b8233a0d9851"
],
[
"r",
"wss://nostr.wine/"
],
[
"r",
"wss://nos.lol/"
],
[
"r",
"wss://nostr.land/"
],
[
"r",
"wss://relay.getalby.com/v1"
],
[
"r",
"wss://relay.mostr.pub/"
],
[
"r",
"wss://relay.snort.social/"
],
[
"r",
"wss://relay.damus.io/"
],
[
"r",
"wss://theforest.nostr1.com/"
],
[
"r",
"wss://relay.primal.net/"
]
],
"content": "Yesterday it was 48% with 57k RLHF pairs. Today it's 50% with zero. This is notable because the previous paper tried using less data and found it was necessary:\n\n\"Training data from math train 7.5k to Open-Reasoner-\nZero 57k, we observe a consistent increase in both training reward and response length for training and evaluation set, indicating that data scale plays a crucial role in training performance.\"\n\nLeading to my conclusion that for zero pairs, the previous record was close to 0%. Maybe this isn't strictly true, but I expect it to be more predictive than seeing a 2% change",
"sig": "8cc0d780ed979d904a44879728bd24eb8510d4710c0e7a36fd01c28ba02fc21f6819de213b96a5e0a9d99adb284d438d1aab930ab2e531088003dd1ba52e057a"
}