You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
131 lines
4.9 KiB
131 lines
4.9 KiB
{
|
|
"$schema": "https://json-schema.org/draft-07/schema#",
|
|
"properties": {
|
|
"wandb": {
|
|
"description": "Configures how WandB behaves",
|
|
"properties": {
|
|
"secretValue": {
|
|
"type": ["string", "null"],
|
|
"pattern": "^(|[a-zA-Z0-9]{32})$"
|
|
}
|
|
},
|
|
"required": []
|
|
},
|
|
"customizerConfig": {
|
|
"description": "Configures how Customizer behaves",
|
|
"properties": {
|
|
"training": {
|
|
"properties": {
|
|
"training_timeout": {
|
|
"description": "The timeout in seconds for training.",
|
|
"minimum": 1,
|
|
"type": "integer"
|
|
}
|
|
}
|
|
},
|
|
"models": {
|
|
"description": "A map of model name to model configuration",
|
|
"patternProperties": {
|
|
".*": {
|
|
"properties": {
|
|
"model_uri": {
|
|
"description": "The URI to download the model from to hydrate the model cache for training jobs. To configure downloading the model from NGC, follow the URI format ngc://org/optional-team/model-name:version. The team and version are optional. To configure downloading the model from NeMo Data Store, follow the URI format `hf://namespace/model-name@version`. The version is optional and will default to `main`.",
|
|
"type": "string"
|
|
},
|
|
"model_path": {
|
|
"description": "The file path to the model accessible by the training job.",
|
|
"type": "string"
|
|
},
|
|
"training_options": {
|
|
"description": "Resource configuration for the training option",
|
|
"type": "array",
|
|
"items": {
|
|
"required": ["training_type", "finetuning_type", "num_gpus"],
|
|
"type": "object",
|
|
"properties": {
|
|
"training_type": {
|
|
"description": "the objective",
|
|
"type": "string",
|
|
"enum": ["sft"]
|
|
},
|
|
"finetuning_type": {
|
|
"description": "Training optimization",
|
|
"type": "string",
|
|
"enum": ["lora", "p_tuning", "all_weights"]
|
|
},
|
|
"num_gpus": {
|
|
"description": "The minimum number of GPUs per node to train a model for the training option.",
|
|
"minimum": 1,
|
|
"type": "integer"
|
|
},
|
|
"num_nodes": {
|
|
"description": "The minimum number of nodes to train a model for the training option.",
|
|
"minimum": 1,
|
|
"type": "integer"
|
|
},
|
|
"tensor_parallel_size": {
|
|
"description": "The number of GPUs which a tensor or layer in a neural network is split across.",
|
|
"minimum": 1,
|
|
"type": "integer"
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"micro_batch_size": {
|
|
"description": "The number of examples processed per data parallel rank during training",
|
|
"minimum": 1,
|
|
"type": "integer"
|
|
},
|
|
"max_seq_length": {
|
|
"description": "The maximum sequence length supported by the model (2048, 4096, etc.)",
|
|
"minimum": 2048,
|
|
"type": "integer"
|
|
},
|
|
"sequence_packing": {
|
|
"description": "Sequence Packing Parameters",
|
|
"type": "object",
|
|
"properties": {
|
|
"pack_size": {
|
|
"description": "Max length of packed sequence",
|
|
"minimum": 1,
|
|
"type": "integer"
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"type": "object",
|
|
"required": ["model_path", "training_options", "max_seq_length"]
|
|
}
|
|
},
|
|
"type": "object"
|
|
},
|
|
"openTelemetry": {
|
|
"properties": {
|
|
"enabled": {
|
|
"type": "boolean",
|
|
"default": true
|
|
},
|
|
"tracesExporter": {
|
|
"type": "string",
|
|
"enum": ["otlp", "console", "none"]
|
|
},
|
|
"metricsExporter": {
|
|
"type": "string",
|
|
"enum": ["otlp", "console", "none"]
|
|
},
|
|
"logsExporter": {
|
|
"type": "string",
|
|
"enum": ["otlp", "console", "none"]
|
|
},
|
|
"exporterOtlpEndpoint": {
|
|
"type": "string"
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"type": "object"
|
|
}
|
|
},
|
|
"title": "Values",
|
|
"type": "object"
|
|
}
|
|
|