Scores

Retrieve Training Dataset

GET/models/{modelId}/scores/training-dataset

ModelsExpand Collapse

ScoreRetrievePromptResponse object { details, modelId, score, 2 more }

details: object { captionsSimilarity, embeddingSimilarity, promptsSimilarity }

Different components used to compute the final score

captionsSimilarity: object { mean, median, min }

The Dice-Sørensen similarity between the prompt and each captions from the model’s training images

The values are between 0 and 1. The higher the value, the more similar the prompt is to the training images’ captions

mean: number

median: number

min: number

embeddingSimilarity: number

The cosine similarity between the prompt and the model’s caption embedding

The value is a number between -1 and 1. The higher the value, the more similar the prompt is to the model’s caption embedding

promptsSimilarity: optional object { mean, median, min }

The Dice-Sørensen similarity between the prompt and each prompts from the model’s examples

The values are between 0 and 1. The higher the value, the more similar the prompt is to the prompts’ examples

mean: number

median: number

min: number

modelId: string

The model ID (example: “model_eyVcnFJcR92BxBkz7N6g5w”)

score: number

The prompt’s score for the input modelId

The score is a number between 0 and 1. The higher the score, the better the prompt is for the model

maximum1

minimum0

status: "complete" or "incomplete" or "unknown"

The status of the prompt’s score

Details if all the necessary data is present to compute the score

One of the following:

"complete"

"incomplete"

"unknown"

syntaxError: optional "Prompt weight should be between 0 and 2" or "Prompt weight symbol should be - or + with a maximum length of 5. Ex: --, ++, ---, +++++" or "Unsupported prompt syntax: '[to:when], [from::when]'" or 4 more

Present only if the prompt has one or more syntax errors

The API checks several syntax errors in the prompt

One of the following:

"Prompt weight should be between 0 and 2"

"Prompt weight symbol should be - or + with a maximum length of 5. Ex: --, ++, ---, +++++"

"Unsupported prompt syntax: '[to:when], [from::when]'"

"Unsupported prompt syntax: 'word:<weight>' or '(word):<weight>'. Please use 'word<weight>' or '(word)<weight>' instead"

"Unsupported prompt syntax: --(s|c)ref"

"Unsupported prompt syntax: <name:weight> or <lora:name:weight>"

"Unsupported url in prompt"

ScoreRetrieveTrainingDatasetResponse object { details, modelId, score, 3 more }

details: object { assetCouplesCaptionsDistance, assetCouplesSimilarity, assetsEmbeddingSimilarity, 4 more }

Different components used to compute the final score

assetCouplesCaptionsDistance: array of object { couple, distance }

For each couple of assets in the training dataset, the distance between the captions

couple: array of string

The couple of AssetIds (example: [‘asset_GTrL3mq4SXWyMxkOHRxlpw’, ‘asset_6VrLs8nC1DHWCfKGaVne0gBn’])

distance: number

The Dice-Sørensen distance between the captions of the assetIds of the couple.

maximum1

minimum0

assetCouplesSimilarity: array of object { couple, similarity }

For each couple of assets in the training dataset, the similarity between the assets

couple: array of string

The couple of AssetIds (example: [‘asset_GTrL3mq4SXWyMxkOHRxlpw’, ‘asset_6VrLs8nC1DHWCfKGaVne0gBn’])

similarity: number

The cosine similarity between the assetIds of the couple.

The value is a number between -1 and 1. The higher the value, the more similar the assets are to each other

maximum1

minimum-1

assetsEmbeddingSimilarity: array of object { assetId, similarity }

For each asset, the similarity with the model’s embedding (excluding the current asset)

assetId: string

The AssetId associated with their similarity (example: “asset_GTrL3mq4SXWyMxkOHRxlpw”)

similarity: number

The cosine similarity between the assetId and the model’s embedding minus the asset’s embedding

The value is a number between -1 and 1. The higher the value, the more similar the asset is to the model’s embedding

maximum1

minimum-1

assetsSimilarity: object { mean, median, min }

The min, mean, median of the assetsEmbeddingSimilarity

mean: number

maximum1

minimum-1

median: number

maximum1

minimum-1

min: number

maximum1

minimum-1

couplesCaptionsDistance: object { mean, median, min }

The min, mean, median of the assetCouplesCaptionsDistance

mean: number

maximum1

minimum0

median: number

maximum1

minimum0

min: number

maximum1

minimum0

couplesSimilarity: object { mean, median, min }

The min, mean, median of the assetCouplesSimilarity

mean: number

maximum1

minimum-1

median: number

maximum1

minimum-1

min: number

maximum1

minimum-1

metadataScore: number

Takes into account all missing data and warnings to compute a metadata score

maximum1

minimum-1

modelId: string

The model ID (example: “model_eyVcnFJcR92BxBkz7N6g5w”)

score: number

The training dataset’s score for the input modelId

The score is a number between 0 and 1. The higher the score, the most chance the model has to train well

maximum1

minimum0

status: "complete" or "incomplete" or "unknown"

The status of the training dataset’s score

Details if all the necessary data is present to compute the score

One of the following:

"complete"

"incomplete"

"unknown"

assetCouplesWarnings: optional array of object { couple, message }

Present only if their is a warning between the assets of the couples

couple: array of string

The couple of AssetIds (example: [‘asset_GTrL3mq4SXWyMxkOHRxlpw’, ‘asset_6VrLs8nC1DHWCfKGaVne0gBn’])

message: array of "Captions are identical" or "Captions are too close" or "Empty caption" or 6 more

An indicative message concerning the couple of assets

One of the following:

"Captions are identical"

"Captions are too close"

"Empty caption"

"Missing caption"

"Missing metadata"

"Same asset"

"Similarity is too high"

"Similarity with other assets is too low"

"Width and height must be larger than 1024"

assetsWarnings: optional array of object { assetId, message }

Present only if the training image assets of the modelId have one or more warnings

assetId: string

The AssetId which has a warning (example: “asset_GTrL3mq4SXWyMxkOHRxlpw”)

message: array of "Captions are identical" or "Captions are too close" or "Empty caption" or 6 more

An indicative message concerning the asset

One of the following:

"Captions are identical"

"Captions are too close"

"Empty caption"

"Missing caption"

"Missing metadata"

"Same asset"

"Similarity is too high"

"Similarity with other assets is too low"

"Width and height must be larger than 1024"

Scores

Retrieve Prompt

Retrieve Training Dataset

ModelsExpand Collapse