ModelOps
Know which model is which.
Keep a record of every experiment, find the best settings automatically, and register the models you trust before they go live. The paper trail auditors ask for, built in.
Who it's for
Teams running models in production.
Regulated teams
Show which data and settings produced the model in production.
ML teams
Compare experiments side by side and stop losing good results.
Managers
See which models are live, who approved them and when.
What teams use it for
- A model registry with an approval step before go-live
- Automatic search for the best training settings overnight
- An audit answer: which data trained this model?
How it works
Three steps.
- 01
Track
Every run records its settings, data and results in MLflow.
- 02
Tune
Katib tries settings for you and reports the best.
- 03
Register
Promote the model you trust, with its history attached.
What you get
ModelOps, in detail.
Experiment tracking
MLflow records parameters, metrics and files for every run.
Model registry
Versioned models with stages, from test to production.
Automatic tuning
Katib searches for better settings while you do other work.
Notebooks into pipelines
Kale turns the notebook that worked into a repeatable pipeline.
An audit trail
A clear record of what was trained, on what, and by whom.
Under the hood
The key numbers.
- Location
- Kathmandu Valley, Nepal
- Billing
- Per second
- Location
- Kathmandu Valley, Nepal
- Billing
- Per second
- GPU
- NVIDIA H200 NVL
- GPU memory
- 141 GB
- Memory bandwidth
- 4.8 TB/s
- Currency
- Nepalese rupees (NPR)
- Ways to pay
- Digital payment, bank transfer or invoice
Billing
Included with your workspace; GPU time in NPR
Commitment: none. Rates are being finalised; every request gets a firm quote in rupees.
How billing worksQuestions
About modelops.
Is this separate from the workspace?
No. ModelOps tools are part of your workspace and work with its notebooks and pipelines.
Can we keep using our own MLflow habits?
Yes. It is standard MLflow, so existing code and habits carry over.
Start with modelops.
Tell us what you want to run. It takes about a minute.