# Nemotron-Labs-3-Puzzle-75B-A9B: Compressing Hybrid MoE LLMs

arXiv: https://arxiv.org/abs/2607.04371 · HTML: https://mlsanity.com/paper/2607.04371/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Efficient MoE training and inference systems, one of 12 directions in Mixture-of-experts.
- Builds on 11 of its 15 references that are in the map.
- First posted on arXiv 5 Jul 2026.

## Placed in

- Mixture-of-experts / [Efficient MoE training and inference systems](https://mlsanity.com/moe/efficient-moe-training-and-inference-systems/)

## Builds on (most cited first, 11 of 11)

- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [Fast Inference from Transformers via Speculative Decoding](https://mlsanity.com/paper/2211.17192/) (2022)
- [ASVD: Activation-aware Singular Value Decomposition for Compressing Large Language Models](https://mlsanity.com/paper/2312.05821/) (2023)
- [Compact Language Models via Pruning and Knowledge Distillation](https://mlsanity.com/paper/2407.14679/) (2024)
- [Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning](https://mlsanity.com/paper/2512.20848/) (2025)
- [NVIDIA Nemotron 3: Efficient and Open Intelligence](https://mlsanity.com/paper/2512.20856/) (2025)
- [Puzzle: Distillation-Based NAS for Inference-Optimized LLMs](https://mlsanity.com/paper/2411.19146/) (2024)
- [LatentMoE: Toward Optimal Accuracy per FLOP and Parameter in Mixture of Experts](https://mlsanity.com/paper/2601.18089/) (2026)
- [SPEED-Bench: A Unified and Diverse Benchmark for Speculative Decoding](https://mlsanity.com/paper/2604.09557/) (2026)
- [Star Elastic: Many-in-One Reasoning LLMs with Efficient Budget Control](https://mlsanity.com/paper/2605.07182/) (2026)
- [Extending Puzzle for Mixture-of-Experts Reasoning Models with Application to GPT-OSS Acceleration](https://mlsanity.com/paper/2602.11937/) (2026)

## Built on by (most cited first, 1 of 1)

- [Low-Discrepancy Dither for Quantized Recurrent State Caches](https://mlsanity.com/paper/2609.39185/) (2026)

## Abstract (arXiv)

We present Nemotron-Labs-3-Puzzle-75B-A9B, a compressed variant of Nemotron-3-Super optimized for interactive deployment. We designed the model to maximize server throughput under high user throughput constraints. In interactive serving workloads on a single 8xB200 node, Puzzle-75B-A9B achieves approximately 2x higher server throughput than Nemotron-3-Super at matched user throughput constraints. In ultra-long-context deployment on a single H100 GPU, the compressed model increases 1M-token concurrency from 1 request to 8 requests. Puzzle-75B-A9B is constructed using a multi-stage pipeline that combines the Iterative Puzzle compression framework with knowledge distillation, reinforcement learning, quantization, and a Multi-Token Prediction head. The compression process jointly optimizes heterogeneous MoE pruning, active parameter budget, and Mamba pruning to improve inference efficiency while preserving model quality. We evaluate Puzzle-75B-A9B on a broad suite of reasoning, coding, multilingual, long-context, and agentic benchmarks. Despite substantial compression, the model retains strong downstream accuracy relative to the parent model across a wide range of tasks. These results demonstrate that large hybrid MoE models can be substantially optimized for deployment efficiency while maintaining strong downstream capability. Our model is publicly available on Hugging Face.