A 1.5B model learns to decide when to use short versus long reasoning via control tokens and a decoupled GRPO objective, reducing thinking-mode usage by 50-90% on math benchmarks with minor accuracy loss.
Llama-nemotron: Efficient reasoning models, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Thinkless: LLM Learns When to Think
A 1.5B model learns to decide when to use short versus long reasoning via control tokens and a decoupled GRPO objective, reducing thinking-mode usage by 50-90% on math benchmarks with minor accuracy loss.