New fine-tuning of language models: Match meaning, not tokens
306
14.6
Next
35 days – 2569:05Popular
23 days – 9851:59
1 day – 7 5770:19Published on 14 May 2026, 17:06
Language models are usually trained to predict the next word, but that does not always lead to the best overall answers. We introduce energy-based fine-tuning, a new method that trains models to produce better full responses, leading to stronger results without the need for complex reward models or verifiers.
Project: energy-based-fine-tuning.githu...
Paper: arxiv.org/abs/2603.12248
GitHub: github.com/sjelassi/ebft_openr...
This session aired on May 14, 2026, at Microsoft Research Forum, Season 2 Episode 4.
Register for the series to hear about new releases: microsoft.com/en-us/research/e...
Explore all previous episodes: aka.ms/researchforumYTplaylist
Project: energy-based-fine-tuning.githu...
Paper: arxiv.org/abs/2603.12248
GitHub: github.com/sjelassi/ebft_openr...
This session aired on May 14, 2026, at Microsoft Research Forum, Season 2 Episode 4.
Register for the series to hear about new releases: microsoft.com/en-us/research/e...
Explore all previous episodes: aka.ms/researchforumYTplaylist
Fresh videos
9 days – 21 0643:39:12
15 days – 15 1861:41
15 days – 335 5399:03
33 days – 123 4612:26Random videos
82 days – 13 31749:07
18.11.24 – 3 87524:48
11.08.23 – 2 7390:24
23.02.22 – 26 7934:01
04.05.18 – 23 5580:20
6 days – 9 3380:39
1 day – 601:27
5 days – 3 6371:34:54
6 days – 111 85828:15
6 days – 78 38318:00
7 days – 31 3089:03
8 days – 1 2650:11
2 hours – 5 85017:04
1 day – 5150:16