Mistral AI · Schema

ModelConfig

Properties

Name Type Description
architecture string The model architecture to use.
base_model string The base model to start from. Required for post-training and reinforcement learning jobs.
num_experts integer Number of experts for MoE architecture.
active_experts integer Number of active experts per token for MoE architecture.
View JSON Schema on GitHub

JSON Schema

mistral-modelconfig-schema.json Raw ↑ 
{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "#/components/schemas/ModelConfig",
  "title": "ModelConfig",
  "type": "object",
  "properties": {
    "architecture": {
      "type": "string",
      "description": "The model architecture to use.",
      "enum": [
        "dense",
        "mixture_of_experts"
      ]
    },
    "base_model": {
      "type": "string",
      "description": "The base model to start from. Required for post-training and reinforcement learning jobs."
    },
    "num_experts": {
      "type": "integer",
      "description": "Number of experts for MoE architecture.",
      "minimum": 1
    },
    "active_experts": {
      "type": "integer",
      "description": "Number of active experts per token for MoE architecture.",
      "minimum": 1
    }
  }
}