Skip to content

Output Configuration

Output specifications in gpux.yml.

Overview

The outputs section defines model output specifications.

outputs:
  - name: string          # Required: Output name
    type: string          # Required: Data type
    shape: [int]          # Optional: Tensor shape
    labels: [string]      # Optional: Class labels
    description: string   # Optional: Description

Fields

name (required)

Output name matching the model's output.

  • Type: string
  • Required: Yes
outputs:
  - name: logits
  - name: probabilities

type (required)

Data type of the output tensor.

  • Type: string
  • Required: Yes
  • Values: float32, float64, int32, int64, uint8, bool, string
outputs:
  - name: logits
    type: float32
  - name: classes
    type: int64

shape

Tensor shape specification.

  • Type: list[int]
  • Required: No
  • Dynamic: Use -1 for dynamic dimensions
outputs:
  - name: logits
    type: float32
    shape: [1, 2]        # Fixed shape
  - name: boxes
    type: float32
    shape: [-1, 4]       # Dynamic number of boxes

labels

Class labels for classification outputs.

  • Type: list[string]
  • Required: No
outputs:
  - name: sentiment
    type: float32
    shape: [1, 2]
    labels: [negative, positive]

description

Human-readable description.

  • Type: string
  • Required: No
outputs:
  - name: logits
    type: float32
    description: Raw classification logits before softmax

Examples

Classification Output

outputs:
  - name: logits
    type: float32
    shape: [1, 2]
    labels: [negative, positive]
    description: Sentiment classification logits

Detection Outputs

outputs:
  - name: boxes
    type: float32
    shape: [-1, 4]
    description: Bounding boxes [x1, y1, x2, y2]
  - name: scores
    type: float32
    shape: [-1]
    description: Confidence scores
  - name: classes
    type: int64
    shape: [-1]
    description: Class indices

Multi-Label Classification

outputs:
  - name: probabilities
    type: float32
    shape: [1, 1000]
    labels: [cat, dog, bird, ...]  # 1000 ImageNet classes

Alternative Syntax

Dict-style outputs:

outputs:
  logits:
    type: float32
    shape: [1, 2]
    labels: [negative, positive]
  embeddings:
    type: float32
    shape: [1, 768]

See Also