Deployment & custom pairings

Deploy your way.

Dedicated endpoints, cloud GPUs, or on-premises. Bring your models and we’ll scope a pairing around your workload and serving environment.

Dedicated endpoints: A compatible model pair in your inference setup.
Your GPUs / on-premises: Models and trained mapper on your own infrastructure.

Choose your model pair

Playground

Try a prompt and compare responses, time to first token, and total runtime across three inference paths.

PlaygroundCompare 3 paths

For developers

Run mapped prefill and target decode in one blocking Python call. Supported customer aliases are qwen3-8b and qwen3.8-27b.

Python · one_triangle 0.1.0

import one_triangle

result = one_triangle.prefill(
    model="qwen3-8b",
    context=context,
    max_tokens=256,
)

print(result.decode())

The service performs mapped prefill and target decode. Customers do not manually transfer a cache.