Lieber et al.: Jamba hybrid Transformer-Mamba experts

What changed

Interleaved attention and selective state-space layers with sparsely activated experts, studying architecture choices for a long-context language model with reduced inference-memory cost.

Scientific scope

Interleaved attention and selective state-space layers with sparsely activated experts, studying architecture choices for a long-context language model with reduced inference-memory cost. Inherits Transformer, Mamba and MoE components; empirical gains depend on architecture, context and hardware. One original model architecture, not a separate record per checkpoint or later Jamba1.5 release.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits Transformer, Mamba and MoE components; empirical gains depend on architecture, context and hardware. One original model architecture, not a separate record per checkpoint or later Jamba1.5 release.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Jamba: A Hybrid Transformer-Mamba Language Model
• Lieber et al.: Jamba hybrid Transformer-Mamba experts — official company record

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction