# Efficient, VRAM-Constrained xLM Inference on Clients

arXiv: https://arxiv.org/abs/2604.26334 · HTML: https://mlsanity.com/paper/2604.26334/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 18 of its 34 references that are in the map.
- First posted on arXiv 29 Apr 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)
- Mixture-of-experts / [Efficient MoE training and inference systems](https://mlsanity.com/moe/efficient-moe-training-and-inference-systems/)

## Builds on (most cited first, 18 of 18)

- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding](https://mlsanity.com/paper/2006.16668/) (2020)
- [KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization](https://mlsanity.com/paper/2401.18079/) (2024)
- [FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU](https://mlsanity.com/paper/2303.06865/) (2023)
- [Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve](https://mlsanity.com/paper/2403.02310/) (2024)
- [DeepSpeed Inference: Enabling Efficient Inference of Transformer Models at Unprecedented Scale](https://mlsanity.com/paper/2207.00032/) (2022)
- [PowerInfer: Fast Large Language Model Serving with a Consumer-grade GPU](https://mlsanity.com/paper/2312.12456/) (2023)
- [Compressing Context to Enhance Inference Efficiency of Large Language Models](https://mlsanity.com/paper/2310.06201/) (2023)
- [PowerInfer-2: Fast Large Language Model Inference on a Smartphone](https://mlsanity.com/paper/2406.06282/) (2024)
- [FastVLM: Efficient Vision Encoding for Vision Language Models](https://mlsanity.com/paper/2412.13303/) (2024)
- [MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs](https://mlsanity.com/paper/2411.11217/) (2024)
- [EdgeMoE: Empowering Sparse Large Language Models on Mobile Devices](https://mlsanity.com/paper/2308.14352/) (2023)
- [LLM Platform Security: Applying a Systematic Evaluation Framework to OpenAI's ChatGPT Plugins](https://mlsanity.com/paper/2309.10254/) (2023)
- [ABQ-LLM: Arbitrary-Bit Quantized Inference Acceleration for Large Language Models](https://mlsanity.com/paper/2408.08554/) (2024)
- [STI: Turbocharge NLP Inference at the Edge via Elastic Pipelining](https://mlsanity.com/paper/2207.05022/) (2022)

## Abstract (arXiv)

To usher in the next round of client AI innovation, there is an urgent need to enable efficient, lossless inference of high-accuracy large language models (LLMs) and vision language models (VLMs), jointly referred to as xLMs, on client systems. To address this, we present pipelined sharding, a novel, benchmark-profile-guided CPU-GPU hybrid scheduling technique to achieve efficient, VRAM-constrained inference for both dense and mixture-of-experts (MoE) LLMs. Using a combination of model sharding at the sub-layer level, CPU offloading, pipelined copy-compute, and prioritized tensor placement in VRAM, it optimizes both time-to-first-token (TTFT) and tokens per second (TPS) metrics, while flexibly adapting to system and inference conditions. For efficient, high-accuracy VLM inference, we combine pipelined sharding with a llama.cpp implementation of three well-understood prior ideas (jointly called VLMOpt), namely, vision tensor CPU offloading, flash attention, and vision and language model VRAM overlap avoidance. These enhancements are targeted at improving client xLM inference in future releases of two important NVIDIA products - the In-Game Inferencing software development kit (IGI SDK) and the Cosmos-Reason1 (CR1) physical AI reasoning VLM. Highlights from our rigorous evaluation spanning multiple models and client systems include: for interactive use, TTFT improves by up to 6.7x and TPS by up to 30x for LLMs, and CR1 inference's VRAM demand is down by 10x, while in batched mode, throughput improves by up to 8.2x, all compared to their respective aggressive baselines. This paper is accepted at the 9th MLSys Conference (Industry Track), 2026. Code and artifact available at: https://github.com/deepshnv/pipeshard-mlsys26-ae