North Mini Code home pagelight logo
  1. Deployment
  2. Deployment Paths

North Mini Code can be run through Cohere’s managed platform, directly from Hugging Face weights, or through AWS-specific deployment paths.

Cohere Platform

Managed access through Cohere’s Chat APIs and Model Vault.

Transformers

Direct Python loading for first experiments and custom apps.

vLLM

Best-documented local server path for agentic coding.

SGLang

OpenAI-compatible local server with pip and Docker flows.

Docker Model Runner

Lowest-friction container-first local path from the Hugging Face deploy panel.

Amazon SageMaker

Managed AWS deployment using the Hugging Face deploy snippet.

AWS Inferentia & Trainium

Neuron-based AWS accelerator path via Optimum Neuron.

Quantized Local Apps

GGUF, FP8, AWQ, and MLX routes for local apps and smaller footprints.

  • Best for first output

  • Best for coding agents

  • Best for AWS teams

Start with /deployment/transformers or /deployment/docker-model-runner.

These are the lowest-friction paths if you mainly want to verify the model, inspect generations, or test prompts.

Hugging Face’s deploy UI currently surfaces JumpStart and CloudFormation as upcoming options rather than fully documented flows. For now, the concrete AWS routes in source material are the SageMaker SDK snippet and the broader Optimum Neuron documentation.

1

Pick the path that matches your constraint

  • fastest managed start: Cohere Platform
  • best local Python path: Transformers
  • best local agent server: vLLM
  • best AWS managed path: SageMaker
  • best reduced-footprint path: Quantized Local Apps
2

Follow the runtime-specific page

Each deployment page in this section keeps the commands, warnings, and quick links scoped to a single route instead of mixing them together.

3

Return to the guide pages once the model is running

After deployment, use /guides/agentic-coding and /concepts/model-overview to shape your application around North Mini Code’s tool-use and context behavior.