Train Robotic AI Models using Qualia
by @fabbe1999
We handle the GPU training infra. Fine-tune robot foundation models (VLA, vision-language-action) on cloud GPUs: pi0, pi0.5 (π0.5), GR00T N1.5, ACT, SmolVLA, SARM reward models. Robotics and robot training with LeRobot-format HuggingFace datasets. Launch, monitor, and cancel fine-tune jobs from the CLI. Agent-native: --json output a
clawhub install qualia-skill📖 About This Skill
name: qualia description: "Fine-tune robot foundation models on cloud GPUs — π0.5, π0, GR00T, SmolVLA, ACT, and more." metadata: {"clawdis":{"emoji":"🤖","requires":{"env":["QUALIA_API_KEY"]},"tags":["robotics","robot-learning","foundation-models","vla","fine-tuning","imitation-learning","manipulation","embodied-ai","ml-training","gpu","reward-model"],"categories":["robotics","ai-ml","developer-tools"],"homepage":"https://qualiastudios.dev"}}
Qualia
Fine-tune Vision-Language-Action (VLA) models for robotics on cloud GPUs.
Setup
1. Sign up at app.qualiastudios.dev 2. Create an API key (Settings → API Keys) 3. Set the env var:
export QUALIA_API_KEY="your-api-key"
When Someone Asks to Train a Model
They probably won't give you everything upfront. Here's what you need and how to get it:
1. Dataset — ask for their HuggingFace dataset ID (e.g. your-org/your-dataset)
2. Model type — if they don't specify, run models and help them choose:
- Quick prototyping → suggest ACT (fast, no base model needed)
- Production quality → suggest π0.5 or π0
- Humanoid robots → suggest GR00T N1.5
- Resource-conscious → suggest SmolVLA
3. Training duration — if unspecified, suggest 2–4 hours for a first run
4. Camera mapping — run dataset-keys on their dataset, then models to see required slots, and map them automatically. Confirm with the user before launching.
If the user already has a project, use it. Otherwise create one.
When Things Go Wrong
| Symptom | Likely cause | Fix |
|---------|-------------|-----|
| Job stuck at credit_validation | Insufficient credits | Run credits, tell user to top up |
| Fails at dataset_preprocessing | Bad camera mapping or invalid dataset | Re-check dataset-keys output, verify mapping |
| Fails at instance_booting | GPU capacity issue | Try a different instance type or region |
| Job failed with no clear error | Check phase events | Run status and read the event messages |
Always run status and share the full phase history with the user when debugging.
Quick Start
# See what models are available (always check — new ones are added regularly)
python3 {baseDir}/scripts/qualia.py modelsCheck GPU options and pricing
python3 {baseDir}/scripts/qualia.py instancesCheck your credit balance
python3 {baseDir}/scripts/qualia.py credits
Train a Model
# 1. Discover image keys in your dataset
python3 {baseDir}/scripts/qualia.py dataset-keys your-org/your-dataset2. Create a project
python3 {baseDir}/scripts/qualia.py project-create "My Robot"3. Launch training
python3 {baseDir}/scripts/qualia.py finetune your-org/your-dataset 4 \
'{"cam_1": "observation.images.top"}' \
--model \
--name "My run"4. Monitor
python3 {baseDir}/scripts/qualia.py status
Notes:
models first to see which VLA types require --model and which don'tmodels) to dataset image keys (from dataset-keys)cam_1, cam_2, cam_3) but the underlying models have a specific input order. Map semantically using these known orders:cam_1 = base/overview camera, cam_2 = left wrist/arm, cam_3 = right wrist/arm
- GR00T N1.5: cam_1 = base/overview camera, cam_2 = left wrist/arm, cam_3 = right wrist/arm
- ACT / SmolVLA: cam_1 = primary camera, cam_2/cam_3 = secondary views
- Fuzzy-match dataset keys to these roles: context_camera or base_0 → cam_1; left_wrist ≈ left_arm → cam_2; right_wrist ≈ right_arm → cam_3
--model for types that don't support custom modelsinstances to get credits/hr, multiply by hours. Tell the user the estimated cost before confirming.Manage Jobs & Projects
python3 {baseDir}/scripts/qualia.py projects # List projects and jobs
python3 {baseDir}/scripts/qualia.py status # Job status and phase history
python3 {baseDir}/scripts/qualia.py cancel # Cancel a running job
python3 {baseDir}/scripts/qualia.py project-delete # Delete a project
Custom Hyperparameters
# Get defaults
python3 {baseDir}/scripts/qualia.py hyperparams [model_id]Validate overrides
python3 {baseDir}/scripts/qualia.py hyperparams-validate '{"learning_rate": 1e-4}'Use in training
python3 {baseDir}/scripts/qualia.py finetune ... --hyper-spec '{"learning_rate": 1e-4, "num_epochs": 50}'
Finetune Flags
| Flag | Description |
|------|-------------|
| --model | Base model ID (required for some VLA types) |
| --name | Job display name |
| --instance | GPU instance type |
| --region | Cloud region |
| --batch-size | Batch size (1–512, default 32) |
| --hyper-spec ' | Custom hyperparameters |
| --rabc | Enable RA-BC with SARM reward model (HF path) |
| --rabc-image-key | Image key for reward annotations |
| --rabc-head-mode | RA-BC head mode (e.g. sparse) |
RA-BC (Reward-Aware Behavior Cloning)
Use a trained SARM reward model to weight training samples. Supported on smolvla, pi0, pi05.
python3 {baseDir}/scripts/qualia.py finetune \
pi0 your-org/your-dataset 4 \
'{"cam_1": "observation.images.top"}' \
--model lerobot/pi0 \
--rabc your-org/sarm-reward-model \
--rabc-image-key observation.images.top \
--rabc-head-mode sparse
Job Phases
queuing → credit_validation → instance_booting → instance_activation → instance_setup → dataset_preprocessing → training_running → model_uploading → completed
Terminal: completed, failed, cancelled
Live Docs
For the latest models, endpoints, and capabilities — always check the live documentation:
Links
X-API-Key header)💡 Examples
# See what models are available (always check — new ones are added regularly)
python3 {baseDir}/scripts/qualia.py modelsCheck GPU options and pricing
python3 {baseDir}/scripts/qualia.py instancesCheck your credit balance
python3 {baseDir}/scripts/qualia.py credits
⚙️ Configuration
1. Sign up at app.qualiastudios.dev 2. Create an API key (Settings → API Keys) 3. Set the env var:
export QUALIA_API_KEY="your-api-key"