# xMIx: High-Performance Serving-Time Platform for Mechanistic Interpretability Apps

arXiv: https://arxiv.org/abs/2607.22595 · HTML: https://mlsanity.com/paper/2607.22595/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security.
- Builds on 23 of its 33 references that are in the map.
- First posted on arXiv 11 Jun 2026.

## Placed in

- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)
- Mechanistic interpretability / [Circuits and mechanistic analysis](https://mlsanity.com/mechinterp/circuits-and-mechanistic-analysis/)

## Builds on (most cited first, 23 of 23)

- [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)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations](https://mlsanity.com/paper/2410.02707/) (2024)
- [Programming Refusal with Conditional Activation Steering](https://mlsanity.com/paper/2409.05907/) (2024)
- [EasyEdit: An Easy-to-use Knowledge Editing Framework for Large Language Models](https://mlsanity.com/paper/2308.07269/) (2023)
- [SEAL: Steerable Reasoning Calibration of Large Language Models for Free](https://mlsanity.com/paper/2504.07986/) (2025)
- [Semantics-Adaptive Activation Intervention for LLMs via Dynamic Steering Vectors](https://mlsanity.com/paper/2410.12299/) (2024)
- [Steering Without Side Effects: Improving Post-Deployment Control of Language Models](https://mlsanity.com/paper/2406.15518/) (2024)
- [Steering MoE LLMs via Expert (De)Activation](https://mlsanity.com/paper/2509.09660/) (2025)
- [Bridging the Language Gaps in Large Language Models with Inference-Time Cross-Lingual Intervention](https://mlsanity.com/paper/2410.12462/) (2024)
- [Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training](https://mlsanity.com/paper/2505.14681/) (2025)
- [Toxicity Detection for Free](https://mlsanity.com/paper/2405.18822/) (2024)
- [SAKE: Steering Activations for Knowledge Editing](https://mlsanity.com/paper/2503.01751/) (2025)
- [EasySteer: A Unified Framework for High-Performance and Extensible LLM Steering](https://mlsanity.com/paper/2509.25175/) (2025)
- [Universal Jailbreak Suffixes Are Strong Attention Hijackers](https://mlsanity.com/paper/2506.12880/) (2025)
- [EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Models](https://mlsanity.com/paper/2504.15133/) (2025)
- [Interpretability Can Be Actionable](https://mlsanity.com/paper/2605.11161/) (2026)
- [Jailbreaking Leaves a Trace: Understanding and Detecting Jailbreak Attacks from Internal Representations of Large Language Models](https://mlsanity.com/paper/2602.11495/) (2026)

## Abstract (arXiv)

Mechanistic interpretability (MI) has emerged as a powerful approach for analyzing and intervening in inference computations, with a growing number of applications such as jailbreak attempt detection, truthfulness evaluation, and hallucination detection. Unfortunately, MI deployment in production model-serving systems is currently not practical, as most existing MI frameworks introduce prohibitively high runtime overheads. The fundamental problem is that MI functions do not compose cleanly with served models: they fragment deployment, often force draining requests and rebuilding serving state, and conflict with critical performance optimizations such as continuous batching and CUDA-graph execution, essential for production deployments. We present xMIx, a serving-native framework for deploying MI applications in production inference serving environments. xMIx enables attaching MI functions to a predefined set of locations in the model runtime, interposing on activations within the layers and residual streams. xMIx supports conditional invocation of MI functions depending on the outputs in preceding model layers. Multiple MI applications can be deployed in a single model instance. xMIx compiles them all into the serving path but activates them dynamically at runtime only when necessary, with negligible performance cost, and without requiring a separate model instance or alternative execution stack. We integrate xMIx with the vLLM serving system and evaluate it across three major models and seven diverse MI applications. xMIx achieves performance comparable to native vLLM execution, incurring a slowdown of 1.3% mean inter-token latency (ITL), 1.2% for tail P99 ITL, 2.6% for mean time to first token (TTFT), and 1.6% for mean total token throughput (TTT).