AI SDK Type Definitions

Common Types

Error

Type Definition

zif_peng_azoai_sdk_types=>ty_error

Field Name	Type	Description
code	ErrorCode	ErrorCode
message	string	The message of this error.
target	string	The location where the error happened if available.

Error Code

Name	Type	Description
conflict	string	The requested operation conflicts with the current resource state.
fileImportFailed	string	Import of file failed.
forbidden	string	The operation is forbidden for the current user/api key.
internalFailure	string	Internal error. Please retry.
invalidPayload	string	The request data is invalid for this operation.
itemDoesAlreadyExist	string	The item does already exist.
jsonlValidationFailed	string	Validation of jsonl data failed.
notFound	string	The resource is not found.
quotaExceeded	string	Quota exceeded.
serviceUnavailable	string	The service is currently not available.
unexpectedEntityState	string	The operation cannot be executed in the current resource's state.

Models

Model Get

Type Definition

zif_peng_azoai_sdk_types=>ty_model_get

Field Name	Type	Description
capabilities	Capabilities	Capabilities. The capabilities of a model. It defines the modes for scaling and the reserved capacity.
created_at	integer	A timestamp when this job or item was created (in unix epochs).
deprecation	Deprecation	Deprecation. The deprecation of a model.
id	string	The identity of this item.
lifecycle_status	Lifecycle Status	LifecycleStatus. The lifecycle status of a model.
object	Type Discriminator	Defines the type of an object.
status	State	State. The state of a job or item.
updated_at	integer	A timestamp when this job or item was modified last (in unix epochs).

Model List

Type Definition

zif_peng_azoai_sdk_types=>ty_model_list

Field Name	Type	Description
data	Model[]	Internal Table of Model

Deployments

Deployment Create Info

Type Definition

zif_peng_azoai_sdk_types=>ty_deployments_create

Field Name	Type	Description
model	string	The OpenAI model identifier (model-id) to deploy. Can be a base model or a fine tune.
scale_settings	Scale Settings	ScaleSettings. The scale settings of a deployment. It defines the modes for scaling and the reserved capacity.

Deployment

Type Definition

zif_peng_azoai_sdk_types=>ty_deployments

Field Name	Type	Description
created_at	integer	A timestamp when this job or item was created (in unix epochs).
error	Error	Error Information
id	string	The identity of this item.
model	string	The OpenAI model identifier (model-id) to deploy. Can be a base model or a fine tune.
object	TypeDiscriminator	Type of the object.
owner	string	The owner of this deployment. For Azure OpenAI only "organization-owner" is supported.
scale_settings	Scale Settings	ScaleSettings. The scale settings of a deployment. It defines the modes for scaling and the reserved capacity.
status	State	State. The state of a job or item.
updated_at	integer	A timestamp when this job or item was modified last (in unix epochs).

Deployment List

Type Definition

zif_peng_azoai_sdk_types=>ty_deployments_list

Field Name	Type	Description
data	Deployment[]	Internal Table of Deployment
object	Type Discriminator	Defines the type of an object.

Files

File

Type Definition

zif_peng_azoai_sdk_types=>ty_file_get

Field Name	Type	Description
bytes	integer	The size of this file when available (can be null).
created_at	integer	A timestamp when this job or item was created (in unix epochs).
error	Error	Error
filename	string	The name of the file.
id	string	The identity of this item.
object	Type Discriminator	Defines the type of an object.
purpose	Purpose	Purpose The intended purpose of the uploaded documents. Use "fine-tune" for fine-tuning. This allows validation of format of the uploaded file.
statistics	File Statistics	FileStatistics A file is a document usable for training and validation. It can also be a service generated document with result details.
status	State	State The state of a job or item.
updated_at	integer	A timestamp when this job or item was modified last (in unix epochs).

File Import

Type Definition

zif_peng_azoai_sdk_types=>ty_file_import

Field Name	Type	Description
content_url	string	The url to download the document from (can be SAS url of a blob or any other external url accessible with a GET request).
filename	string	The name of the JSON Lines file to be uploaded. If the purpose is set to "fine-tune", each line is a JSON record with "prompt" and "completion" fields representing your training examples.
purpose	Purpose	Purpose. The intended purpose of the uploaded documents. Use "fine-tune" for fine-tuning.

File List

Type Definition

zif_peng_azoai_sdk_types=>ty_file_list

Field Name	Type	Description
data	File[]	The list of items.
object	Type Discriminator	Defines the type of an object.

File Upload

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_int_file_upload

Field Name	Type	Description
prompt	string	Training Data Prompt
completion	string	Training Data Completion.

Fine Tunes

Fine Tune Create

Type Definition

zif_peng_azoai_sdk_types=>ty_finetune_create

Field Name	Type	Description
model	string	The identifier (model-id) of the base model used for this fine-tune.
training_file	string	The file identity (file-id) that is used for training this fine tuned model.
batch_size	integer	The batch size to use for training. The batch size is the number of training examples used to train a single forward and backward pass. In general, we've found that larger batch sizes tend to work better for larger datasets. The default value as well as the maximum value for this property are specific to a base model.
classification_betas	number[]	The classification beta values. If this is provided, we calculate F-beta scores at the specified beta values. The F-beta score is a generalization of F-1 score. This is only used for binary classification. With a beta of 1 (i.e.the F-1 score), precision and recall are given the same weight. A larger beta score puts more weight on recall and less on precision. A smaller beta score puts more weight on precision and less on recall.
classification_n_classes	integer	The number of classes in a classification task. This parameter is required for multiclass classification.
classification_positive_class	string	The positive class in binary classification. This parameter is needed to generate precision, recall, and F1 metrics when doing binary classification.
compute_classification_metrics	boolean	A value indicating whether to compute classification metrics. If set, we calculate classification-specific metrics such as accuracy and F-1 score using the validation set at the end of every epoch. These metrics can be viewed in the results file. In order to compute classification metrics, you must provide a validation_file.Additionally, you must specify classification_n_classes for multiclass classification or classification_positive_class for binary classification.
learning_rate_multiplier	number	The learning rate multiplier to use for training. The fine-tuning learning rate is the original learning rate used for pre-training multiplied by this value. Larger learning rates tend to perform better with larger batch sizes. We recommend experimenting with values in the range 0.02 to 0.2 to see what produces the best results.
n_epochs	integer	The number of epochs to train the model for. An epoch refers to one full cycle through the training dataset.
prompt_loss_weight	number	The weight to use for loss on the prompt tokens. This controls how much the model tries to learn to generate the prompt (as compared to the completion which always has a weight of 1.0), and can add a stabilizing effect to training when completions are short. If prompts are extremely long (relative to completions), it may make sense to reduce this weight so as to avoid over-prioritizing learning the prompt.
suffix	string	The suffix used to identify the fine-tuned model. The suffix can contain up to 40 characters (a-z, A-Z, 0-9,- and _) that will be added to your fine-tuned model name.
validation_file	string	The file identity (file-id) that is used to evaluate the fine tuned model during training.

Fine Tune Events

Type Definition

zif_peng_azoai_sdk_types=>ty_finetune_events

Field Name	Type	Description
data	Fine Tune Event[]	The list of items.
object	Type Discriminator	Defines the type of an object.

Fine Tune List

Type Definition

zif_peng_azoai_sdk_types=>ty_finetune_list

Field Name	Type	Description
data	Fine Tune Result[]	Internal table of Fine Tunes.
object	Type Discriminator	Defines the type of an object.

Fine Tune Result

Type Definition

zif_peng_azoai_sdk_types=>ty_finetune_result

Field Name	Type	Description
created_at	integer	A timestamp when this job or item was created (in unix epochs).
error	Error	Error
events	Fine Tune Events[]	The events that show the progress of the fine-tune run including queued, running and completed.
fine_tuned_model	string	The identifier (model-id) of the resulting fine tuned model. This property is only populated for successfully completed fine-tune runs. Use this identifier to create a deployment for inferencing.
hyperparams	HyperParameter	HyperParameters The hyper parameter settings used in a fine tune job.
id	string	The identity of this item.
model	string	The identifier (model-id) of the base model used for the fine-tune.
object	Type Discriminator	Defines the type of an object.
organisation_id	string	The organisation id of this fine tune job. Unused on Azure OpenAI; compatibility for OpenAI only.
result_files	File[]	The result file identities (file-id) containing training and evaluation metrics in csv format. The file is only available for successfully completed fine-tune runs.
status	State	State. The state of a job or item.
suffix	string	The suffix used to identify the fine-tuned model.
training_files	File[]	The file identities (file-id) that are used for training the fine tuned model.
updated_at	integer	A timestamp when this job or item was modified last (in unix epochs).
user_id	string	The user id of this fine tune job. Unused on Azure OpenAI; compatibility for OpenAI only.
validation_files	File[]	The file identities (file-id) that are used to evaluate the fine tuned model during training.

Embeddings

Embedding Input

Type Definition

zif_peng_azoai_sdk_types=>ty_embeddings_input

Field Name	Type	Description
input	string[]	Input text to get embeddings for, encoded as a string. The number of input tokens varies depending on what model you are using.

Embedding Output

Type Definition

zif_peng_azoai_sdk_types=>ty_embeddings_output

Field Name	Type	Description
object	Type Discriminator	Defines the type of an object. For embedding, this will be "list"
data	Embeddings[]	An array of embeddings, based on the number of inputs.
model	string	The identifier of the model/deployment used to generate the embeddings.
usage	Embedding Usage	Useage Information for this embedding run.

Completions

Completion Input

Type Definition

zif_peng_azoai_sdk_types=>ty_completion_input

Field Name	Type	Description
prompt	string[]	The prompt(s) to generate completions for, encoded as a string, a list of strings, or a list of token lists. Note that <\|endoftext\|> is the document separator that the model sees during training, so if a prompt isn't specified the model will generate as if from the beginning of a new document.
max_tokens	integer	The maximum number of tokens to generate.
user	string	A unique identifier representing your end-user, which can help monitoring and detecting abuse. Do not pass PII identifiers instead use pseudoanonymized values such as GUIDs
n	integer	The number of completions to generate for each prompt.
temperature	number	What sampling temperature to use, between 0 and 2. Higher values means the model will take more risks. Try 0.9 for more creative applications, and 0 (argmax sampling) for ones with a well-defined answer. We generally recommend altering this or top_p but not both.
top_p	number	An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or temperature but not both.
stop	string[]	Up to four sequences where the API will stop generating further tokens. The returned text won't contain the stop sequence.
frequency_penalty	number	Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.
presence_penalty	number	Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.
best_of	integer	Generates best_of completions server-side and returns the "best" (the one with the lowest log probability per token). When used with n, best_of controls the number of candidate completions and n specifies how many to return from those - best_of must be greater than n

caution

Careful using the parameters "n" and "best_of". Because this parameter generates many completions, it can quickly consume your token quota. Use carefully and ensure that you have reasonable settings for max_tokens and stop.

Completion Output

Type Definition

zif_peng_azoai_sdk_types=>ty_completion_output

Field Name	Type	Description
id	string	The identity of this item.
object	Type Discriminator	Defines the type of an object.
created	integer	A timestamp when this job or item was created (in unix epochs).
model	string	The identifier (model-id) of the model used for this completion.
choices	Completion Choice[]	The completions that are generated
usage	Completion Usage	Useage Information for this completion run.

Completion Choice

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_completion_resobj_choices

Field Name	Type	Description
text	string	The text of the completion.
index	integer	The index of the prompt this completion is generated for.
finish_reason	string	Reason for the response text to be output.

Usage

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_completion_resobj_usage

Field Name	Type	Description
completion_tokens	integer	The number of tokens generated by the completion/embedding.
prompt_tokens	integer	The number of tokens used as prompt for the completion/embedding.
total_tokens	integer	The total number of tokens (prompt + completion) generated by the completion/embedding.

Chat Completions

Chat Completion Message

Type Definition

zif_peng_azoai_sdk_types=>ty_chatcompletion_message

Field Name	Type	Description
role	Role	The role of the message.
content	string	The content of the message.

Chat Completion Input

Type Definition

zif_peng_azoai_sdk_types=>ty_chatcompletion_input

Note

The current Completion input structure has only a subset of fields, compared to all the fields supported by AI Engines. In future, we will expand this capability

Field Name	Type	Description
messages	Chat Completion Message[]	A list of messages that will be used as context to generate the completion.
max_tokens	integer	The maximum number of tokens to generate.
user	string	A unique identifier representing your end-user, which can help monitoring and detecting abuse. Do not pass PII identifiers instead use pseudoanonymized values such as GUIDs
n	integer	The number of completions to generate for each prompt.
temperature	number	What sampling temperature to use, between 0 and 2. Higher values means the model will take more risks. Try 0.9 for more creative applications, and 0 (argmax sampling) for ones with a well-defined answer. We generally recommend altering this or top_p but not both.
top_p	number	An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or temperature but not both.
stop	string[]	Up to four sequences where the API will stop generating further tokens. The returned text won't contain the stop sequence.
frequency_penalty	number	Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.
presence_penalty	number	Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.

Chat Completion Output

Type Definition

zif_peng_azoai_sdk_types=>ty_chatcompletion_output

Field Name	Type	Description
id	string	The identity of this item.
object	Type Discriminator	Defines the type of an object.
created	integer	A timestamp when this job or item was created (in unix epochs).
model	string	The identifier (model-id) of the model used for this completion.
choices	Chat Completion Choice[]	The completions that are generated
usage	Completion Usage	Useage Information for this completion run.

Chat Completion Choice

Type Definition

zif_peng_azoai_sdk_typinternal=>tty_chatcompl_resobj_choices

Field Name	Type	Description
index	integer	The index of the prompt this completion is generated for.
finish_reason	string	Reason for the response text to be output.
message	Chat Completion Message	The messages generated by the completion.

Internal Types

The following are internal types of AI SDK, and chances are, you wont be declaring these types in your use cases or ABAP programs. However, you will still come across these as fields in other exposed data types.

Capabilities

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_int_mod_get_capabilities

Value	Type	Description
completion	boolean	Whether the model supports completion.
embeddings	boolean	Whether the model supports embeddings.
fine_tune	boolean	Whether the model supports fine-tuning.
inference	boolean	Whether the model supports inference.
scale_types	string[]	The supported scale types.

Deprecation

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_int_mod_get_deprecation

Value	Type	Description
fine_tune	integer	The end date of fine tune support of this model. Will be null for fine tune models.
inference	integer	The end date of inference support of this model.

Embedding

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_embedding

Field Name	Type	Description
object	Type Discriminator	Defines the type of an object.
index	integer	The index of the input this embedding is generated for.
embedding	float[]	The embedding vector.

File Statistics

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_int_file_stats

Field Name	Type	Description
examples	integer	The number of contained training examples in files of kind "fine-tune" once validation of file content is complete.
tokens	integer	The number of tokens used in prompts and completions for files of kind "fine-tune" once validation of file content is complete.

Finetune Event

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_int_finetune_event

Field Name	Type	Description
created_at	integer	A timestamp when this event was created (in unix epochs).
level	Log Level	LogLevel. The verbosity level of an event.
message	string	The message describing the event. This can be a change of state, e.g., enqueued, started, failed or completed, or other events like uploaded results.
object	Type Discriminator	Defines the type of an object.

Hyper Parameters

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_int_finetune_hyperparams

Field Name	Type	Description
batch_size	integer	The batch size to use for training. The batch size is the number of training examples used to train a single forward and backward pass. In general, we've found that larger batch sizes tend to work better for larger datasets. The default value as well as the maximum value for this property are specific to a base model.
classification_betas	number[]	The classification beta values. If this is provided, we calculate F-beta scores at the specified beta values. The F-beta score is a generalization of F-1 score. This is only used for binary classification. With a beta of 1 (i.e.the F-1 score), precision and recall are given the same weight. A larger beta score puts more weight on recall and less on precision. A smaller beta score puts more weight on precision and less on recall.
classification_n_classes	integer	The number of classes in a classification task. This parameter is required for multiclass classification.
classification_positive_class	string	The positive class in binary classification. This parameter is needed to generate precision, recall, and F1 metrics when doing binary classification.
compute_classification_metrics	boolean	A value indicating whether to compute classification metrics. If set, we calculate classification-specific metrics such as accuracy and F-1 score using the validation set at the end of every epoch. These metrics can be viewed in the results file. In order to compute classification metrics, you must provide a validation_file.Additionally, you must specify classification_n_classes for multiclass classification or classification_positive_class for binary classification.
learning_rate_multiplier	number	The learning rate multiplier to use for training. The fine-tuning learning rate is the original learning rate used for pre-training multiplied by this value. Larger learning rates tend to perform better with larger batch sizes. We recommend experimenting with values in the range 0.02 to 0.2 to see what produces the best results.
n_epochs	integer	The number of epochs to train the model for. An epoch refers to one full cycle through the training dataset.
prompt_loss_weight	number	The weight to use for loss on the prompt tokens. This controls how much the model tries to learn to generate the prompt (as compared to the completion which always has a weight of 1.0), and can add a stabilizing effect to training when completions are short. If prompts are extremely long (relative to completions), it may make sense to reduce this weight so as to avoid over-prioritizing learning the prompt.

Log Level

Type Definition

string

Value	Description
error	This message represents a non recoverable issue.
info	This message represents an informational message.
warning	This event represents a mitigated issue.

Lifecycle Status

Type Definition

string

Value	Type	Description
generally-available	string	The model is generally available.
preview	string	Model is in preview and covered by the service preview terms.

Purpose

Type Definition

string

Value	Description
fine-tune	This file contains training data for a fine tune job.
fine-tune-results	This file contains the results of a fine tune job.

Role

Type Definition

string

Value	Description
system	System Role, also known as System message - provides initial instructions to the model.
user	User Message. When invoking the AI Engine with Chat Completion, the last message should have the role "user".
assistant	This role is used to for previous or current responses from AI Engine

Scale Settings

Type Definition

zif_peng_azoai_sdk_typinternal=>ty_scale_settings

Field Name	Type	Description
capacity	Integer	The constant reserved capacity of the inference endpoint for this deployment. Only used when scale_type is manual
scale_type	String manual standard	Defines how scaling operations will be executed. If set to manual, capacity must be specified.

State

Type Definition

string

Value	Description
canceled	The operation has been canceled and is incomplete.
deleted	The entity has been deleted but may still be referenced by other entities predating the deletion.
failed	The operation has completed processing with a failure and cannot be further consumed.
notRunning	The operation was created and is not queued to be processed in the future.
running	The operation has started to be processed.
succeeded	The operation has successfully be processed and is ready for consumption.

Type Discriminator

Type Definition

string

Value	Description
model	This object represents a model (can be a base models or fine tune job result).
deployment	This object represents a deployment.
file	This object represents a file.
fine-tune	This object represents a fine tune job.
fine-tune-event	This object represents an event of a fine tune job.
list	This object represents a list of other objects.
completion	This object represents a completion.
chat.completion	This object represents a chat completion.