# Zebra-Llama: Towards Extremely Efficient Hybrid Models

arXiv: https://arxiv.org/abs/2505.17272 · HTML: https://mlsanity.com/paper/2505.17272/

## Key facts

- Cited by 18 papers in the map (data through 14 Oct 2026).
- Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving.
- Builds on 27 of its 34 references that are in the map.
- First posted on arXiv 22 May 2025.

## Placed in

- Efficient LLM inference and serving / [Serving systems and request scheduling](https://mlsanity.com/inference/serving-systems-and-request-scheduling/)

## Builds on (most cited first, 27 of 27)

- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [PIQA: Reasoning about Physical Commonsense in Natural Language](https://mlsanity.com/paper/1911.11641/) (2019)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answering](https://mlsanity.com/paper/1809.02789/) (2018)
- [Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention](https://mlsanity.com/paper/2006.16236/) (2020)
- [RULER: What's the Real Context Size of Your Long-Context Language Models?](https://mlsanity.com/paper/2404.06654/) (2024)
- [Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality](https://mlsanity.com/paper/2405.21060/) (2024)
- [Qwen2.5-Coder Technical Report](https://mlsanity.com/paper/2409.12186/) (2024)
- [The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale](https://mlsanity.com/paper/2406.17557/) (2024)
- [DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model](https://mlsanity.com/paper/2405.04434/) (2024)
- [RACE: Large-scale ReAding Comprehension Dataset From Examinations](https://mlsanity.com/paper/1704.04683/) (2017)
- [Improved Knowledge Distillation via Teacher Assistant](https://mlsanity.com/paper/1902.03393/) (2019)
- [Jamba: A Hybrid Transformer-Mamba Language Model](https://mlsanity.com/paper/2403.19887/) (2024)
- [An Empirical Study of Mamba-based Language Models](https://mlsanity.com/paper/2406.07887/) (2024)
- [Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling](https://mlsanity.com/paper/2406.07522/) (2024)
- [Compact Language Models via Pruning and Knowledge Distillation](https://mlsanity.com/paper/2407.14679/) (2024)
- [The Mamba in the Llama: Distilling and Accelerating Hybrid Models](https://mlsanity.com/paper/2408.15237/) (2024)
- [Hymba: A Hybrid-head Architecture for Small Language Models](https://mlsanity.com/paper/2411.13676/) (2024)
- [Transformers to SSMs: Distilling Quadratic Knowledge to Subquadratic Models](https://mlsanity.com/paper/2408.10189/) (2024)
- [Annealing Knowledge Distillation](https://mlsanity.com/paper/2104.07163/) (2021)
- [GenQA: Generating Millions of Instructions from a Handful of Prompts](https://mlsanity.com/paper/2406.10323/) (2024)
- [Llamba: Scaling Distilled Recurrent Models for Efficient Language Processing](https://mlsanity.com/paper/2502.14458/) (2025)
- [X-EcoMLA: Upcycling Pre-Trained Attention into MLA for Efficient and Extreme KV Compression](https://mlsanity.com/paper/2503.11132/) (2025)

## Built on by (most cited first, 18 of 18)

- [Distilling to Hybrid Attention Models via KL-Guided Layer Selection](https://mlsanity.com/paper/2512.20569/) (2025)
- [Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts](https://mlsanity.com/paper/2601.22156/) (2026)
- [Hybrid Architectures for Language Models: Systematic Analysis and Design Insights](https://mlsanity.com/paper/2510.04800/) (2025)
- [CARE: Covariance-Aware and Rank-Enhanced Decomposition for Enabling Multi-Head Latent Attention](https://mlsanity.com/paper/2603.17946/) (2026)
- [Distill-then-Replace: Efficient Task-Specific Hybrid Attention Model Construction](https://mlsanity.com/paper/2601.11667/) (2026)
- [Apriel-H1: Towards Efficient Enterprise Reasoning Models](https://mlsanity.com/paper/2511.02651/) (2025)
- [SpikingBrain2.0: Brain-Inspired Foundation Models for Efficient Long-Context and Cross-Platform Inference](https://mlsanity.com/paper/2604.22575/) (2026)
- [Morphing into Hybrid Attention Models](https://mlsanity.com/paper/2606.30562/) (2026)
- [FarSkip-Collective: Unhobbling Blocking Communication in Mixture of Experts Models](https://mlsanity.com/paper/2511.11505/) (2025)
- [MobileLLM-Flash: Latency-Guided On-Device LLM Design for Industry Scale Deployment](https://mlsanity.com/paper/2603.15954/) (2026)
- [Taylor-Calibrate: Principled Initialization for Hybrid Linear Attention Distillation](https://mlsanity.com/paper/2606.16429/) (2026)
- [BlindSight: Harnessing Sparsity for Efficient Vision-Language Models](https://mlsanity.com/paper/2507.09071/) (2025)
- [BOSCH: Black-Box Binary Optimization for Short-Context Attention-Head Selection in LLMs](https://mlsanity.com/paper/2604.05942/) (2026)
- [Long-Context Aware Upcycling: A New Frontier for Hybrid LLM Scaling](https://mlsanity.com/paper/2604.24715/) (2026)
- [DASH: Fast Differentiable Architecture Search for Hybrid Attention in Minutes on a Single GPU](https://mlsanity.com/paper/2605.20936/) (2026)
- [Attention Amnesia in Hybrid LLMs: When CoT Fine-Tuning Breaks Long-Range Recall, and How to Fix It](https://mlsanity.com/paper/2606.11052/) (2026)
- [On-Policy Attention Linearization](https://mlsanity.com/paper/2609.31947/) (2026)
- [Multilinguality in Hybrid Attention LLMs](https://mlsanity.com/paper/2609.35378/) (2026)

## Abstract (arXiv)

With the growing demand for deploying large language models (LLMs) across diverse applications, improving their inference efficiency is crucial for sustainable and democratized access. However, retraining LLMs to meet new user-specific requirements is prohibitively expensive and environmentally unsustainable. In this work, we propose a practical and scalable alternative: composing efficient hybrid language models from existing pre-trained models. Our approach, Zebra-Llama, introduces a family of 1B, 3B, and 8B hybrid models by combining State Space Models (SSMs) and Multi-head Latent Attention (MLA) layers, using a refined initialization and post-training pipeline to efficiently transfer knowledge from pre-trained Transformers. Zebra-Llama achieves Transformer-level accuracy with near-SSM efficiency using only 7-11B training tokens (compared to trillions of tokens required for pre-training) and an 8B teacher. Moreover, Zebra-Llama dramatically reduces KV cache size -down to 3.9%, 2%, and 2.73% of the original for the 1B, 3B, and 8B variants, respectively-while preserving 100%, 100%, and >97% of average zero-shot performance on LM Harness tasks. Compared to models like MambaInLLaMA, X-EcoMLA, Minitron, and Llamba, Zebra-Llama consistently delivers competitive or superior accuracy while using significantly fewer tokens, smaller teachers, and vastly reduced KV cache memory. Notably, Zebra-Llama-8B surpasses Minitron-8B in few-shot accuracy by 7% while using 8x fewer training tokens, over 12x smaller KV cache, and a smaller teacher (8B vs. 15B). It also achieves 2.6x-3.8x higher throughput (tokens/s) than MambaInLlama up to a 32k context length. We will release code and model checkpoints upon acceptance.