Dashboard Training
Use the Halo Forge dashboard as the primary operator surface
The dashboard is the normal path for local workstation training. The CLI remains the automation path, but the dashboard is designed for operators who want to choose a goal, check preflight, launch, monitor, serve, and inspect results without memorizing every flag.
Start vs Train
- Start is the beginner-safe path. It stays SFT-only and chooses conservative defaults.
- Train is the full method surface. It supports SFT, RAFT, DPO, ORPO, RM, GRPO, VLM, audio, reasoning, and agentic training.
- Runs monitors active and completed work.
- Results answers what to do next: open the run, serve the final model, compare, or inspect local paths.
Default Output Path
Dashboard launches save under:
~/.halo-forge/runs/<method>-<goal-or-template>-<model-slug>
This avoids installed-app permission failures from repo-relative models/... paths.
Method Preconditions
| Method | Needs |
|---|---|
| SFT | model, dataset, writable output path |
| RAFT | model, prompt file, verifier |
| DPO/ORPO | model, preference dataset |
| RM | model, preference dataset |
| GRPO | model, prompt dataset, verifier |
| VLM | compatible VLM family and image-text data |
| Audio | audio dependencies, task, audio data |
| Reasoning | compatible text model and reasoning data |
| Agentic | tool-call traces or structured-output data |
When a method is capability-gated, the dashboard shows the reason and keeps the CLI path documented.
Serving After Training
When a run produces a final model or adapter path, open Results and choose Serve model. Halo Forge manages one local serve process at a time and sends Playground to the managed endpoint by default.