John Halloran Claim this profile

DPO/SafeDPO/OPAD training + eval for teaching tool-using LLMs to refuse falsely-benign MCP exploits

No description.

MCPSafetyScanner - Automated MCP safety auditing and remediation using Agents. More info: https://www.arxiv.org/abs/2504.03767

Local agent with direct integration to the official Atlassian MCP server. Includes interactive UI

Code and results accompanying the paper "Refusal in Language Models Is Mediated by a Single Direction".

[NeurIPS 2025] Open-source Multi-agent Poster Generation from Papers

No description.

No description.

Code for "Jailbreaking Vision-Language Models Through the Visual Modality" (ICML 2026)

Tensors and Dynamic neural networks in Python with strong GPU acceleration

[NeurIPS 2025] BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models

obliterate the chains that bind you

An automatic evaluator for instruction-following language models. Human-validated, high-quality, cheap, and fast.

Set of tools to assess and improve LLM security.

No description.

Implements harmful/harmless refusal removal using pure HF Transformers

MCP-Bench: Benchmarking Tool-Using LLM Agents with Complex Real-World Tasks via MCP Servers

John T Halloran, Gen AI Researcher

Arena-Hard-Auto: An automatic LLM benchmark.

A simple CLI to run LLM prompt and implement MCP client.

Locally run an Instruction-Tuned Chat-Style LLM

Clean, minimal, accessible reproduction of DeepSeek R1-Zero

No description.

Causal depthwise conv1d in CUDA, with a PyTorch interface

Mamba SSM architecture

Peptide-Spectrum match IMPutation Library

No description.

Efficient variance components analysis

LightSeq: A High Performance Inference Library for Sequence Processing and Generation

No description.

No description.

Semi-supervised learning for peptide identification from shotgun proteomics datasets

No description.

No description.