Models API
Prefix: /api/v1/models · UI: Model Management
Model lifecycle
| Method | Path | Description |
|---|---|---|
POST | /download | Download a model from HuggingFace (202); body: repo_id, quantization, trust_remote_code |
POST | /{id}/redownload | Re-download / re-quantize an existing model (202) |
DELETE | /{id}/cancel | Cancel an in-progress download (revokes the running task) |
GET | `` | List models |
GET | /local-cache/list | Enumerate models present in the local HF cache |
GET | /{id} | Get model details |
GET | /{id}/architecture | Discovered layer/hook structure (dynamic layer discovery output) |
PATCH | /{id} | Update model metadata |
DELETE | /{id} | Delete model (204). Returns 409 if a training references it |
GET | /tasks/{task_id} | Raw Celery task status |
Activation extraction (Stage 1)
These run the base model over a tokenized dataset and cache raw activations — see the extraction pipeline for how this differs from SAE feature extraction.
| Method | Path | Description |
|---|---|---|
POST | /{id}/estimate-extraction | Estimate VRAM/disk for an extraction config before running it |
POST | /{id}/extract-activations | Start activation extraction (202); config includes layers, hook types, and optional gpu_id |
GET | /{id}/extractions | List the model's activation extractions |
GET | /{id}/extractions/active | The currently running extraction, if any (200 with null data when idle) |
POST | /{id}/extractions/{eid}/cancel | Cancel a running extraction |
POST | /{id}/extractions/{eid}/retry | Retry a failed extraction |
DELETE | /{id}/extractions | Delete the model's extraction records/artifacts |
Progress channels: models/{id}/progress (events model:progress|completed|error) and models/{id}/extraction.