> ## Documentation Index > Fetch the complete documentation index at: https://docs.phospho.ai/llms.txt > Use this file to discover all available pages before exploring further. # Start training a model > Start training an ACT or gr00t model on the specified dataset. This will upload a trained model to the Hugging Face Hub using the main branch of the specified dataset. ## OpenAPI ````yaml post /training/start openapi: 3.1.0 info: title: FastAPI version: 0.1.0 servers: [] security: [] paths: /training/start: post: tags: - training summary: Start training a model description: >- Start training an ACT or gr00t model on the specified dataset. This will upload a trained model to the Hugging Face Hub using the main branch of the specified dataset. operationId: start_training_training_start_post requestBody: content: application/json: schema: $ref: '#/components/schemas/TrainingRequest' required: true responses: '200': content: application/json: schema: $ref: '#/components/schemas/StartTrainingResponse' description: Successful Response '422': content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' description: Validation Error components: schemas: TrainingRequest: description: >- Pydantic model for training request validation. This version consolidates all model name and parameter logic into a single validator to prevent redundant operations and fix the duplicate suffix bug. properties: dataset_name: description: Dataset repository ID on Hugging Face, should be a public dataset title: Dataset Name type: string model_name: anyOf: - type: string - type: 'null' description: >- Name of the trained model to upload to Hugging Face, should be in the format phospho-app/ or title: Model Name model_type: description: Type of model to train, supports 'ACT', 'gr00t', and 'pi0' enum: - ACT - ACT_BBOX - gr00t - pi0 - custom title: Model Type type: string private_mode: default: false description: Whether to use private training (PRO users only) title: Private Mode type: boolean training_params: anyOf: - $ref: '#/components/schemas/TrainingParamsAct' - $ref: '#/components/schemas/TrainingParamsActWithBbox' - $ref: '#/components/schemas/TrainingParamsGr00T' - $ref: '#/components/schemas/TrainingParamsPi0' - type: 'null' description: Training parameters for the model. title: Training Params user_hf_token: anyOf: - type: string - type: 'null' description: User's personal HF token for private training title: User Hf Token wandb_api_key: anyOf: - type: string - type: 'null' description: >- WandB API key for tracking training, you can find it at https://wandb.ai/authorize title: Wandb Api Key required: - model_type - dataset_name title: TrainingRequest type: object StartTrainingResponse: additionalProperties: true properties: message: anyOf: - type: string - type: 'null' title: Message model_url: anyOf: - type: string - type: 'null' description: URL to the Hugging Face model card. title: Model Url status: default: ok enum: - ok - error title: Status type: string training_id: anyOf: - type: integer - type: 'null' description: >- ID of the training to start. This is the ID returned by the training request. title: Training Id required: - training_id title: StartTrainingResponse type: object HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' title: Detail type: array title: HTTPValidationError type: object TrainingParamsAct: additionalProperties: true description: >- Training parameters are left to None by default and are set depending on the dataset in the training pipeline. properties: batch_size: anyOf: - exclusiveMinimum: 0 maximum: 150 type: integer - type: 'null' description: >- Batch size for training, we run this on an A10G. Leave it to None to auto-detect based on your dataset title: Batch Size save_freq: default: 5000 description: Number of steps between saving the model. exclusiveMinimum: 0 maximum: 1000000 title: Save Freq type: integer steps: anyOf: - exclusiveMinimum: 0 maximum: 1000000 type: integer - type: 'null' description: >- Number of training steps. Leave it to None to auto-detect based on your dataset title: Steps title: TrainingParamsAct type: object TrainingParamsActWithBbox: additionalProperties: true description: Training parameters for ACT with bounding box properties: batch_size: anyOf: - exclusiveMinimum: 0 maximum: 150 type: integer - type: 'null' description: >- Batch size for training, we run this on an A10G. Leave it to None to auto-detect based on your dataset title: Batch Size image_key: default: main description: Key for the image to run detection on, e.g. 'main' or 'images.main' examples: - main - images.main minLength: 1 title: Image Key type: string image_keys_to_keep: description: >- Optional list of image keys to keep. If none, all image keys will be dropped. items: type: string title: Image Keys To Keep type: array save_freq: default: 5000 description: Number of steps between saving the model. exclusiveMinimum: 0 maximum: 1000000 title: Save Freq type: integer steps: anyOf: - exclusiveMinimum: 0 maximum: 1000000 type: integer - type: 'null' description: >- Number of training steps. Leave it to None to auto-detect based on your dataset title: Steps target_detection_instruction: default: 'e.g.: green lego brick, red ball, blue plushy...' description: >- Instruction for the target object to detect, e.g. 'red/orange lego brick' examples: - red/orange lego brick - brown plushy - blue ball minLength: 4 title: Target Detection Instruction type: string title: TrainingParamsActWithBbox type: object TrainingParamsGr00T: additionalProperties: true properties: batch_size: anyOf: - exclusiveMinimum: 0 maximum: 128 type: integer - type: 'null' default: 64 description: >- Batch size for training. Decrease it if you get an Out Of Memory (OOM) error title: Batch Size data_dir: default: data/ description: The directory to save the dataset to title: Data Dir type: string learning_rate: default: 0.0001 description: Learning rate for training. exclusiveMinimum: 0 maximum: 1 title: Learning Rate type: number num_epochs: default: 10 description: Number of epochs to train for. exclusiveMinimum: 0 maximum: 100 title: Num Epochs type: integer output_dir: default: outputs/ description: The directory to save the model to title: Output Dir type: string save_steps: default: 1000 description: Number of steps between saving the model. exclusiveMinimum: 0 maximum: 100000 title: Save Steps type: integer validation_data_dir: anyOf: - type: string - type: 'null' description: >- Optional directory to save the validation dataset to. If None, validation is not run. title: Validation Data Dir validation_dataset_name: anyOf: - type: string - type: 'null' description: Optional dataset repository ID on Hugging Face to use for validation title: Validation Dataset Name title: TrainingParamsGr00T type: object TrainingParamsPi0: additionalProperties: false description: Training parameters for Pi0 model properties: batch_size: anyOf: - exclusiveMinimum: 0 maximum: 128 type: integer - type: 'null' description: >- Batch size for training, leave it to None to auto-detect based on your dataset title: Batch Size data_dir: default: data/ description: The directory to save the dataset to title: Data Dir type: string epochs: default: 10 description: Number of epochs to train for, default is 10 exclusiveMinimum: 0 maximum: 50 title: Epochs type: integer learning_rate: default: 0.0001 description: Learning rate for training, default is 0.0001 exclusiveMinimum: 0 maximum: 1 title: Learning Rate type: number output_dir: default: outputs/ description: The directory to save the model to title: Output Dir type: string path_to_pi0_repo: default: . description: >- The path to the openpi repo. If not provided, will assume we are in the repo. title: Path To Pi0 Repo type: string train_test_split: default: 1 description: >- Train test split ratio, default is 1.0 (no split), should be between 0 and 1 exclusiveMinimum: 0 maximum: 1 title: Train Test Split type: number validation_dataset_name: anyOf: - type: string - type: 'null' description: Optional dataset repository ID on Hugging Face to use for validation title: Validation Dataset Name title: TrainingParamsPi0 type: object ValidationError: properties: loc: items: anyOf: - type: string - type: integer title: Location type: array msg: title: Message type: string type: title: Error Type type: string required: - loc - msg - type title: ValidationError type: object ````