Every evaluation and inference run starts from a model in your workspace. Adding one is a quick two-part flow: give the model a name, then choose where its weights come from. You can upload your own archive or import a model straight from HuggingFace.
Two ways to add a model
| Source | Use when | What you provide |
|---|
| File | You have your own weights, such as a fine-tuned checkpoint or a merged adapter. | A .zip archive containing config.json and the model weights. |
| HF model | You want a public model from the HuggingFace Hub. | A search term, then pick the model from the results. |
Steps
1. Open the Models page and click Add Model
In the Eval tab, click Models in the left sidebar, then click + Add Model in the top right of the AI Models page.
2. Enter the basic details
The Add Model panel slides in. Give the model a name (for example myqwen3-0.6b-model) and, optionally, a short description. You can edit the description later, so it’s fine to leave it blank for now.
Click Add Model to continue. You’ll choose where the weights come from on the next step.
Pick a name you’ll recognize later in model lists and leaderboards. Avoid throwaway names like test1.
3. Choose where the model comes from
The model is created in a Draft state and opens to its model card. The Add Model card prompts you to choose a source. Pick one of the two tabs.
Option A — Upload a file
On the File tab, drag and drop a model archive onto the upload area, or click Browse Files to pick it. The archive should be a .zip containing config.json and the model weights.
Once the file is attached, click Add model.
Option B — Import from HuggingFace
On the HF model tab, type a model name into Search HuggingFace models (for example llama, qwen, or mistral).
As you type, matching models appear with their download count, likes, and task tag. Click the one you want.
The selected model shows as a chip, and the Add model button becomes active. Click Add model to import it.
4. Confirm the model is ready
BenchGen registers the model and provisions its endpoint. When it finishes, the status badge changes to ready and the model card’s endpoint panel fills in with its LiteLLM Name, Endpoint URL, access Token, and basic information such as format and visibility.
Your model now appears in the Models list and is available for inference and evaluation.
Next Steps
