1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. ๐งต
Introducing ๐๐ฎ๐๐ฅ-๐๐๐ข๐ ๐ก๐ญ๐๐ ๐๐๐ข๐ง๐๐จ๐ซ๐๐๐ฆ๐๐ง๐ญ ๐๐๐๐ซ๐ง๐ข๐ง๐ (๐๐๐๐) โ a new framework forย ๐๐จ๐ ๐ญ๐ซ๐๐ข๐ง๐ข๐ง๐ ๐จ๐ง ๐ฉ๐ซ๐๐๐๐ซ๐๐ง๐๐ ๐๐๐ญ๐, working @AIatMeta
Key ideas:โจโ Transform preference modeling into RLVR and use GRPOโจโ Integrate theย BradleyโTerry model into reinforcement learning via DWRL
๐กImportant to keep the inductive bias from preference modeling!
Introducing ๐๐ฎ๐๐ฅ-๐๐๐ข๐ ๐ก๐ญ๏ฟฝ๏ฟฝ๐ ๐๐๐ข๐ง๐๐จ๐ซ๐๐๐ฆ๐๐ง๐ญ ๐๐๐๐ซ๐ง๐ข๐ง๐ (๐๐๐๐) โ a new framework forย ๐๐จ๐ ๐ญ๐ซ๐๐ข๐ง๐ข๐ง๐ ๐จ๐ง ๐ฉ๐ซ๐๐๐๐ซ๐๐ง๐๐ ๐๐๐ญ๐, working @AIatMeta
Key ideas:โจโ Transform preference modeling into RLVR and use GRPOโจโ Integrate theย BradleyโTerry model into reinforcement learning via DWRL
๐กImportant to keep the inductive bias from preference modeling!
Glad to propose Multi-IF (Multi-Turn and Multilingual Instructions Following)!
Join us in pushing the boundaries of LLM instruction-following capabilities. Let's innovate together!
Excited to share a new benchmark from our team- this eval set, *Multi-IF*, enables benchmarking LLMs for Multi-Turn and Multilingual Instructions Following, which is something that the LLM community been waiting for.
Link to paper: https://t.co/CmrhnCqMUc
Excited to share a new benchmark from our team- this eval set, *Multi-IF*, enables benchmarking LLMs for Multi-Turn and Multilingual Instructions Following, which is something that the LLM community been waiting for.
Link to paper: https://t.co/CmrhnCqMUc
Check out our recent @GoogleAI "HyperPrompt" paper at #ICML2022.
TL;DR: Hypernetwork-learned task prompts outperform prompt tuning, adapters & hyperformer.
pdf: https://t.co/TcH3FcgvAo
Work led by @YunHeHe17 & @HuaixiuZheng at Google Brain.