McGill NLP
McGill NLP is a research group within McGill University and Mila focused on natural language processing, including semantic parsing, question answering, reading comprehension and conversational systems. The group publishes research and open-source models and datasets so researchers and developers can reproduce and build on its work.
At a Glance
- Academic NLP and machine-learning researchers
- Developers building embedding, retrieval and classification systems
- Teams building web-browsing and web-navigation agents
- Organizations evaluating LLM and web-agent safety, reasoning and generalization
- +1 more
AI Tools by McGill NLP
(1)
LLM2Vec
Convert LLMs into Text Encoders
Discussions
No discussions yet
Be the first to start a discussion about McGill NLP
Latest News
LLM2Vec repository announces LLM2Vec-Gen for interpretable generative embeddings
McGill NLP publishes SafeArena, an evaluation framework for the safety of autonomous web agents
McGill NLP publishes AgentRewardBench on evaluating automatic evaluations of web-agent trajectories
McGill NLP publishes DeepSeek-R1 Thoughtology, an analysis of LLM reasoning
Products & Services
An open-source recipe and model family that converts decoder-only large language models into text encoders using bidirectional attention, masked next-token prediction and unsupervised contrastive learning, with optional supervised contrastive fine-tuning. Models and adapters are distributed through GitHub and Hugging Face.
An open-source toolkit and model project for building agents that browse the web by following instructions and talking with users. Its first model, Llama-3-8B-Web, is fine-tuned on WebLINX interactions and is intended for web navigation and dialogue.
A real-world website-navigation benchmark and dataset used by WebLlama, covering multi-turn dialogue, web actions and generalization across websites, domains and locations.
A recipe for interpretable, generative embeddings that encode the potential answer of an LLM to a query rather than the query itself.
Market Position
McGill NLP positions itself as an academic, open-source research group rather than a commercial product company. LLM2Vec is differentiated from traditional encoder-only embedding models by transforming decoder-only LLMs into encoders and, according to the project’s paper and ServiceNow write-up, achieving strong or state-of-the-art results on relevant MTEB comparisons without synthetic GPT-4 data. WebLlama competes in open web-agent research and reports performance comparisons against GPT-4V on WebLINX.
Leadership
Executive Team
Siva Reddy
Assistant Professor, McGill University; Facebook CIFAR AI Chair; co-lead of the McGill NLP Group
McGill NLP faculty member working in natural language processing, machine learning and computational linguistics; his official profile says he co-leads the group.
Jackie Cheung
Associate Professor, McGill University
McGill NLP faculty member and associate professor in the School of Computer Science.
Business Model
Revenue Model
Academic research group that releases open-source code, datasets and model checkpoints; the reviewed sources describe research and distribution, not a paid subscription, API or licensing business model.
Target Markets
- Academic NLP and machine-learning researchers
- Developers building embedding, retrieval and classification systems
- Teams building web-browsing and web-navigation agents
- Organizations evaluating LLM and web-agent safety, reasoning and generalization
- Users of open-source LLMs and Hugging Face tooling
- Text embedding and semantic representation
- Semantic search and information retrieval
- Clustering and classification
- Word-level NLP tasks
- Web browsing agents that follow instructions and interact through dialogue
- Research reproducibility, model training and benchmark evaluation