A 7B hybrid attention-recurrent model outperforms its pure-transformer counterpart on pretraining metrics and scales more efficiently, supported by a proof that hybrids are strictly more expressive than either transformers or linear RNNs.
• Rather than using the ad-hoc piecewise learning rate schedule from Olmo 3 7B (Olmo Team, 2025), we use a standard cosine decay to 10% of the maximum learning rate
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Olmo Hybrid: From Theory to Practice and Back
A 7B hybrid attention-recurrent model outperforms its pure-transformer counterpart on pretraining metrics and scales more efficiently, supported by a proof that hybrids are strictly more expressive than either transformers or linear RNNs.