Creează un cont pentru a te abona la acest blog și a-i urmări articolele pe pagina Lectură.
-
How to Train an LLM to do proofs: Beyond Verifiable Rewards
Code: https://github.com/Tufalabs/LLMProofs While Reinforcement Learning from Verifiable Rewards (RLVR) has demonstrated substantial success in improving large language models (LLM) performance on verifiable tasks, significant challenges remain in extendi…
-
All In on Winning: The dangers of RL without moral reward
In this work, we show that models trained on seemingly benign settings can inadvertently lead to misaligned models, raising significant safety threats for training LLMs.
-
RLSR: Reinforcement Learning from Self Reward
This is an early draft and is not yet ready for publication.
-
AlphaWrite: Inference time compute Scaling for Writing
Sample Story: The Sea Chose for Us and Rue des Cygnes You can try AlphaWrite out here Code Repository: AlphaWrite on GitHub Large Language Models have demonstrated remarkable performance improvements through increased inference-time compute, particularly…
-
LLMs for Engineering: Teaching Models to Design High Powered Rockets
This is an early draft and is not yet ready for publication.
-
LADDER+TTRL
This is an early draft and is not yet ready for publication.
-
Turning the web into RL Questions
This is an early draft and is not yet ready for publication.
-
EAD: Entropy Adpative Decoding
This is an early draft and is not yet ready for publication.
-
MoDEM: Mixture of Domain Expert Models
This is an early draft and is not yet ready for publication.
-
REL: Working out is all you need
This is an early draft and is not yet ready for publication.