MaplePT is Canada's Sovereign Language Model
Built from Canadian data. Trained on Canadian infrastructure.
Tuned for Canadian values.
DOI: 10.36227/techrxiv.176162061.19169845/v1
Why MaplePT?
MaplePT was built as a statement — that advanced Canadian AI doesn't need to be hosted or trained on foreign infrastructure; it doesn't need to be environmentally wasteful. It proves Canada can lead in ethical, local, and sustainable AI innovation.
Sovereignty
Training data curated in-house and aligned with Canadian public knowledge frameworks.
Accessibility
Runs efficiently on consumer GPUs — trained across distributed RTX 4080 and 3090 systems.
Sustainability
Low-carbon fine-tuning and small-model optimization — no mega-datacenters required.
MaplePT Mini v1
Built in Canada. Open. Sovereign.
How We Built It
A practical, transparent approach that balances performance, cost, and sovereignty.
Architecture
Based on a modified Phi-3-style architecture tuned for instruction following and knowledge retention.
Training Infrastructure
Distributed training across low-cost GPUs using PyTorch + Accelerate + QLoRA, optimizing for energy and cost.
Datasets
Balanced Canadian English and French corpus, aligned to public, educational, and government sources.
Alignment
Fine-tuned with custom identity and refusal datasets to strengthen factuality, civility, and bias resistance.
Available for Everyone
Explore the open release and deploy MaplePT locally or on-prem.
DOI: 10.36227/techrxiv.176162061.19169845/v1
Hugging Face
Explore the open-source release and model card.
Docker Image (vLLM)
Run MaplePT locally or on-prem with an OpenAI-compatible API endpoint.
docker pull canxpai/maplept-mini-v1-vllm:latest docker run -p 8686:8686 canxpai/maplept-mini-v1-vllmAt the Edge of Convenience and Conscience
"MaplePT is more than just another AI model — it's a case study in building AI at the edge of convenience and conscience of Canadians."
Transparency
All experiments reproducible — full model card and training corupus available upon approval.
Collaboration
Encouraging universities and startups to extend MaplePT for domain-specific applications.
Innovation
Demonstrating how small-scale AI can lead to big-impact national innovation.
We're looking for Collaborators
We're building small language models (SLMs) that are fast, private, and highly focused. SLMs can run on modest hardware, align tightly to domain knowledge, and reduce operational costs— ideal for Healthcare, Education, Legal, Defence, and Manufacturing.
Partner on sector-specific SLM experts
Explore the sovereign model ecosystem
MaplePT is part of a broader CanXP AI architecture spanning sovereign AI, small language model training, private AI deployment, and sector-specific model programs.
Join the Sovereign AI Movement
Sovereign AI isn't about isolation — it's about empowerment. Let's redefine what it means to build intelligence on Canadian terms.