Launch the terminal UI: a full control center with 20 workspaces for device inspection, data preparation, training, serving, model operations, benchmarking, evaluation, and job monitoring.
| Tab | Description |
|---|
| Device | GPU/ANE info, Metal feature detection, memory gauge, kernel tuning, UltraFusion topology |
| Models | Browse cached models, HuggingFace Hub search (S), memory fit estimation, download |
| Datasets | Scan and preview local datasets (JSONL, Parquet, CSV) with line counts |
| Tokenize | Convert JSONL text corpora into binary shards for full pretraining |
| Training | Configure and launch SFT/LoRA/QLoRA training runs with sectioned parameter forms |
| EmbedTrain | Configure sentence embedding fine-tuning with contrastive losses |
| Pretrain | Configure full-parameter pretraining from tokenized shards |
| Distillation | Configure knowledge distillation (online, offline, progressive) |
| RLKD | Configure reinforcement learning with knowledge distillation |
| GRPO | Configure GRPO/DAPO reasoning training with reward functions and sampling params |
| Monitor | Live loss curves, LR schedule, throughput sparklines, timing breakdown gauges |
| Inference | Interactive chat interface with markdown rendering and generation settings sidebar |
| DFlash | Configure block-diffusion speculative decoding |
| Serve | Configure an OpenAI-compatible server with KV cache and batching options |
| Quantize | Export GGUF/MLX quantized models |
| Merge | Configure multi-model merge jobs |
| Bench | Configure training, inference, and workload benchmarks |
| Eval | Configure perplexity/evaluation runs |
| Modelfile | Generate Ollama Modelfiles and exports |
| Jobs | Job history with log viewer, status tracking, and metadata |
| Key | Action |
|---|
Tab / Shift+Tab | Switch tabs |
Alt+1-9 | Jump to tab directly |
Ctrl+1-9 | Jump to tab directly on terminals that do not emit Alt-number |
? | Show contextual help |
L | Adjust learning rate mid-run |
S | Search HuggingFace Hub (Models tab) |
q | Quit |
