Workload

Overview

This document describes the JSON schema for a Workload. A Workload defines a concrete, executable instance of a Definition by binding specific values to all variable axes and specifying the data source for all inputs. It represents the exact configuration under which a Solution is benchmarked. Storage Format: In the FlashInfer-Bench dataset, a standalone Workload is stored using the Trace data structure with only the definition and workload fields populated, while solution and evaluation are set to null. In FlashInfer Trace dataset, all workloads of the same definition are stored in a single JSONL file where each line is a Workload object.

JSON Schema Description

Top-Level Object Structure

Field	Type	Required	Description
`uuid`	string	Yes	A unique identifier for this workload configuration.
`axes`	object	Yes	An object mapping `var` axis names from the `Definition` to their concrete integer values.
`inputs`	object	Yes	An object describing the data source for each input.

`inputs` : Input Descriptor Objects

This object maps input names (e.g., "A", "weight", "mask") to input descriptors that explain where the data comes from and (when necessary) how it should be generated or loaded. Each descriptor must contain at least the type field. Additional fields become required or optional depending on the chosen type.

Field	Type	Required	Description
`type`	string	Yes	Data source type. Could be `random`, `scalar`, or `safetensors`.

Additional fields for type scalar:

Field	Type	Required	Description
`value`	int, float, bool	Yes	The concrete value of the input.

Additional fields for type safetensors:

Field	Type	Required	Description
`path`	string	Yes	Relative path or URI of the `.safetensors` file.
`tensor_key`	string	Yes	The key inside the safetensors container that holds this tensor.

Example: RMSNorm Workload

{
  "definition": "rmsnorm_d4096",
  "workload": {
    "uuid": "6120f144-b973-4bd9-b884-77ecb132914e",
    "axes": {
      "batch_size": 32
    },
    "inputs": {
      "input": {
        "type": "safetensors",
        "path": "/data/rmsnorm_evals/b32_input.safetensors",
        "tensor_key": "input"
      },
      "weight": {
        "type": "safetensors",
        "path": "/data/rmsnorm_evals/rmsnorm_weight.safetensors",
        "tensor_key": "weight"
      },
      "eps": {
        "type": "scalar",
        "value": 1e-6
      }
    }
  },
  "solution": null,
  "evaluation": null
}

Example: GEMM Workload with Random Inputs

{
  "definition": "gemm_n_4096_k_4096",
  "workload": {
    "uuid": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
    "axes": {
      "M": 1024
    },
    "inputs": {
      "A": {
        "type": "random"
      },
      "B": {
        "type": "random"
      }
    }
  },
  "solution": null,
  "evaluation": null
}

Getting Started

Tutorials

FlashInfer Trace

Dataset

Op Type Reference

Overview

JSON Schema Description

Top-Level Object Structure

`inputs` : Input Descriptor Objects

Example: RMSNorm Workload

Example: GEMM Workload with Random Inputs

​Overview

​JSON Schema Description

​Top-Level Object Structure

​inputs : Input Descriptor Objects

​Example: RMSNorm Workload

​Example: GEMM Workload with Random Inputs

Overview

JSON Schema Description

Top-Level Object Structure

`inputs` : Input Descriptor Objects

Example: RMSNorm Workload

Example: GEMM Workload with Random Inputs