Skip to main content
The Models page manages Base Models, which are the starting point for training, and LoRA Models, which are checkpoints produced by training runs and can be deployed through Osmosis inference.

​
Base Models

Base models are imported from Hugging Face and used as the starting point for training on Osmosis.

​
Default Base Models

New workspaces include these default base models:
The Models page in your workspace is the source of truth for the base models available there.
If Add Model is visible on the Models page, you can use it to import a Hugging Face model repository that you can access.

​
List Base Models

The base model list shows model name, creation date, and creator.

​
Inspect a Base Model

The CLI shows the Hugging Face path, parameter count, context window, Hugging Face URL, and inference prices in USD per million tokens for input, cached input, and output. Unset prices show –; see model info for the JSON fields. In the Platform, click a base model card on the Models page to inspect its details.

​
LoRA Models

LoRA models are trained checkpoints produced by training runs. The Models page lists them separately from base models and shows training run, checkpoint step, training reward, creation date, and deployment status when inference deployment is available for your account.

​
Inspect LoRA Models

List LoRA models:
Show details for a single LoRA model:
Model details include the base model, training run, checkpoint step, training reward, Hugging Face publish status, and deployment status when deployment info is available. From the detail page, you can publish the model to Hugging Face, download its artifacts, rename it, or delete it. Undeploy an active LoRA model before renaming or deleting it. Deleted models and their artifacts are retained for 30 days before permanent deletion. List base models and LoRA models side by side:
When deployment info is available, the LoRA section also shows the workspace’s deployment-quota summary (for example, 2 of 5 inference deployments used).

​
Deploy a LoRA Model

After a training run finishes, list its LoRA models to find one to deploy:
Deploy a LoRA model by name:
Deploying an inactive LoRA model reactivates it. Deploying an already-active LoRA model is a no-op.

​
Call the Inference Endpoint

Deployed LoRA models are served through the OpenAI-compatible chat completions endpoint:
Use your Osmosis API key and the canonical model value from the model detail page or osmosis model info. The model value has the form <base_model_path>:<lora-model-name>.
If inference deployment is not available for your account, deployment status, deployment quota, and endpoint snippets may be hidden.
A deployed model that has not been called for a while is no longer loaded, so its next request waits for the model to start. The start takes about 20 seconds when the model’s files are cached nearby (a warm start). It takes about 2 to 3 minutes when Osmosis fetches the files from storage (a cold start). The Inference section of the model’s detail page shows a Readiness, last 7 days strip with one cell per hour. Each cell shows the state the model spent the most time in during that hour. Ready means requests start right away. Warm start or Cold start means the model was not loaded then, so a request would have waited for that start. No data means more of the hour went unmeasured than was spent in any one state. Hover over a cell to see the minutes spent in each state.

​
Tag Requests

Add an X-Osmosis-Tags header to group usage by your own labels, such as a job or experiment ID. On the Usage page, open By Tag and choose the tag key to see requests and tokens per value.
With the OpenAI Python SDK, set the header once on the client: OpenAI(base_url="https://inference.osmosis.ai/v1", api_key=..., default_headers={"X-Osmosis-Tags": "job=f26225ad-9110"}).
  • Up to 16 key=value pairs per request, separated by commas.
  • Keys use lowercase letters, digits, _, ., and -, start with a letter, and are at most 64 characters. Keys starting with osmosis are reserved.
  • Values use letters, digits, _, ., :, /, and -, are at most 256 characters, and cannot be empty. Each key can appear only once.
  • A workspace records up to 2,000 distinct tag keys and pairs per hour; usage past that limit is counted under __overflow__, which is reserved as a value.
  • An invalid header never fails the request: its tags are ignored and the response carries X-Osmosis-Tags-Rejected with the reason.
Do not put secrets or personal data in tags.

​
Undeploy

To transition a LoRA model’s deployment to inactive:
The LoRA model remains in the training run history; undeploy only transitions the serving deployment to inactive. undeploy is idempotent: calling it on an already-inactive model is a no-op. If an active evaluation run or benchmark run is using the model, Osmosis prevents undeployment until the run finishes.

​
Requirements

  • Run model commands from the workspace directory so the CLI can resolve the connected workspace from Git origin.
  • The LoRA model must belong to a training run in the same workspace.
  • Inference deployment must be available for your account. Deploying models also requires workspace billing to be in good standing (for self-serve workspaces, a valid payment method on file).
  • GitHub setup must be healthy before training runs can produce new LoRA models.

​
Next Steps

Training Runs

Submit training runs and inspect their LoRA models.

Command Reference

Review model and deployment commands and options.
Last modified on October 27, 2026