Pause fine-tuning
POST
/fine_tuning/jobs/{fine_tuning_job_id}/pausePause a fine-tune job.
Pause fine-tuning
curl -X POST https://api.openai.com/v1/fine_tuning/jobs/ftjob-abc123/pause \
-H "Authorization: Bearer $OPENAI_API_KEY"
Response example
{
"object": "fine_tuning.job",
"id": "ftjob-abc123",
"model": "o4-mini",
"created_at": 1721764800,
"fine_tuned_model": null,
"organization_id": "org-123",
"result_files": [],
"status": "pausing",
"validation_file": "file-123",
"training_file": "file-abc",
"error": {},
"finished_at": null,
"seed": null,
"trained_tokens": null,
"method": {
"type": "reinforcement",
"reinforcement": {
"grader": {
"type": "string_check",
"name": "Example string check grader",
"input": "{{sample.output_text}}",
"reference": "{{item.label}}",
"operation": "eq"
},
"hyperparameters": {
"reasoning_effort": "medium"
}
}
}
}
Returns
FineTuningJob object { id, created_at, error, 16 more } The fine_tuning.job object represents a fine-tuning job that has been created through the API.
FineTuningJob object { id, created_at, error, 16 more }
The fine_tuning.job object represents a fine-tuning job that has been created through the API.
id: string The object identifier, which can be referenced in the API endpoints.
created_at: integer The Unix timestamp (in seconds) for when the fine-tuning job was created.
formatunixtime error: object or unknown or null For fine-tuning jobs that have failed, this will contain more information on the cause of the failure.
One of the following: { code, message, param } For fine-tuning jobs that have failed, this will contain more information on the cause of the failure.
code: string A machine-readable error code.
message: string A human-readable error message.
param: string or null The parameter that was invalid, usually training_file or validation_file. This field will be null if the failure was not parameter-specific.
unknown fine_tuned_model: string or null The name of the fine-tuned model that is being created. The value will be null if the fine-tuning job is still running.
finished_at: integer or null The Unix timestamp (in seconds) for when the fine-tuning job was finished. The value will be null if the fine-tuning job is still running.
formatunixtime model: string The base model that is being fine-tuned.
object: "fine_tuning.job" The object type, which is always “fine_tuning.job”.
organization_id: string The organization that owns the fine-tuning job.
result_files: array of string The compiled results file ID(s) for the fine-tuning job. You can retrieve the results with the Files API.
seed: integer or null The seed used for the fine-tuning job.
status: "validating_files" or "queued" or "running" or 5 more The current status of the fine-tuning job, which can be validating_files, queued, running, pausing, paused, succeeded, failed, or cancelled.
One of the following: "validating_files" "queued" "running" "succeeded" "failed" "cancelled" "pausing" "paused" trained_tokens: integer or null The total number of billable tokens processed by this fine-tuning job. The value will be null if the fine-tuning job is still running.
training_file: string The file ID used for training. You can retrieve the training data with the Files API.
validation_file: string or null The file ID used for validation. You can retrieve the validation results with the Files API.
estimated_finish: optional integer or null The Unix timestamp (in seconds) for when the fine-tuning job is estimated to finish. The value will be null if the fine-tuning job is not running.
formatunixtime hyperparameters: optional object or null The hyperparameters used for the fine-tuning job. This value will only be returned when running supervised jobs.
batch_size: optional "auto" or number or null (default: "auto") Number of examples in each batch. A larger batch size means that model parameters
are updated less frequently, but with lower variance.
One of the following: "auto" number minimum1 maximum256 learning_rate_multiplier: optional "auto" or number (default: "auto") Scaling factor for the learning rate. A smaller learning rate may be useful to avoid
overfitting.
One of the following: "auto" number exclusiveMinimum0 n_epochs: optional "auto" or number (default: "auto") The number of epochs to train the model for. An epoch refers to one full cycle
through the training dataset.
One of the following: "auto" number minimum1 maximum50 integrations: optional array of FineTuningJobWandbIntegrationObject or null A list of integrations to enable for this fine-tuning job.
maxItems5 type: "wandb" The type of the integration being enabled for the fine-tuning job
wandb: FineTuningJobWandbIntegration { project, entity, name, tags } The settings for your integration with Weights and Biases. This payload specifies the project that
metrics will be sent to. Optionally, you can set an explicit display name for your run, add tags
to your run, and set a default entity (team, username, etc) to be associated with your run.
project: string The name of the project that the new run will be created under.
entity: optional string or null The entity to use for the run. This allows you to set the team or username of the WandB user that you would
like associated with the run. If not set, the default entity for the registered WandB API key is used.
name: optional string or null A display name to set for the run. If not set, we will use the Job ID as the name.
tags: optional array of string A list of tags to be attached to the newly created run. These tags are passed through directly to WandB. Some
default tags are generated by OpenAI: “openai/finetune”, “openai/{base-model}”, “openai/{ftjob-abcdef}”.
metadata: optional Metadata or null Set of 16 key-value pairs that can be attached to an object. This can be
useful for storing additional information about the object in a structured
format, and querying for objects via API or the dashboard.
Keys are strings with a maximum length of 64 characters. Values are strings
with a maximum length of 512 characters.
method: optional object or null The method used for fine-tuning.
type: "supervised" or "dpo" or "reinforcement" The type of method. Is either supervised, dpo, or reinforcement.
One of the following: "supervised" "dpo" "reinforcement" dpo: optional DpoMethod { hyperparameters } Configuration for the DPO fine-tuning method.
hyperparameters: optional DpoHyperparameters { batch_size, beta, learning_rate_multiplier, n_epochs } The hyperparameters used for the DPO fine-tuning job.
batch_size: optional "auto" or number (default: "auto") Number of examples in each batch. A larger batch size means that model parameters are updated less frequently, but with lower variance.
One of the following: "auto" number minimum1 maximum256 beta: optional "auto" or number (default: "auto") The beta value for the DPO method. A higher beta value will increase the weight of the penalty between the policy and reference model.
One of the following: "auto" number exclusiveMinimum0 maximum2 learning_rate_multiplier: optional "auto" or number (default: "auto") Scaling factor for the learning rate. A smaller learning rate may be useful to avoid overfitting.
One of the following: "auto" number exclusiveMinimum0 n_epochs: optional "auto" or number (default: "auto") The number of epochs to train the model for. An epoch refers to one full cycle through the training dataset.
One of the following: "auto" number minimum1 maximum50 reinforcement: optional ReinforcementMethod { grader, hyperparameters } Configuration for the reinforcement fine-tuning method.
grader: StringCheckGrader or TextSimilarityGrader or PythonGrader or 2 more The grader used for the fine-tuning job.
One of the following: StringCheckGrader { input, name, operation, 2 more } A StringCheckGrader object that performs a string comparison between input and reference using a specified operation.
input: string The input text. This may include template strings.
name: string The name of the grader.
operation: "eq" or "ne" or "like" or "ilike" The string check operation to perform. One of eq, ne, like, or ilike.
One of the following: "eq" "ne" "like" "ilike" reference: string The reference text. This may include template strings.
type: "string_check" The object type, which is always string_check.
TextSimilarityGrader { evaluation_metric, input, name, 2 more } A TextSimilarityGrader object which grades text based on similarity metrics.
evaluation_metric: "cosine" or "fuzzy_match" or "bleu" or 8 more The evaluation metric to use. One of cosine, fuzzy_match, bleu,
gleu, meteor, rouge_1, rouge_2, rouge_3, rouge_4, rouge_5,
or rouge_l.
One of the following: "cosine" "fuzzy_match" "bleu" "gleu" "meteor" "rouge_1" "rouge_2" "rouge_3" "rouge_4" "rouge_5" "rouge_l" input: string The text being graded.
name: string The name of the grader.
reference: string The text being graded against.
type: "text_similarity" (default: "text_similarity") The type of grader.
PythonGrader { name, source, type, image_tag } A PythonGrader object that runs a python script on the input.
name: string The name of the grader.
source: string The source code of the python script.
type: "python" The object type, which is always python.
image_tag: optional string The image tag to use for the python script.
ScoreModelGrader { input, model, name, 3 more } A ScoreModelGrader object that uses a model to assign a score to the input.
input: array of { content, role, type } The input messages evaluated by the grader. Supports text, output text, input image, and input audio content blocks, and may include template strings.
content: string or ResponseInputText or OutputText or 3 more Inputs to the model - can contain template strings. Supports text, output text, input images, and input audio, either as a single item or an array of items.
One of the following: TextInput = string A text input to the model.
ResponseInputText { text, type, prompt_cache_breakpoint } A text input to the model.
text: string The text input to the model.
type: "input_text" (default: "input_text") The type of the input item. Always input_text.
prompt_cache_breakpoint: optional { mode, ttl } Marks the exact end of a reusable prompt prefix. When ttl is omitted, the breakpoint inherits the request’s prompt_cache_options.ttl; the boundary is not rounded to a token block.
mode: "explicit" (default: "explicit") The breakpoint mode. Always explicit.
ttl: optional "30m" or "12h" The minimum lifetime of this cache breakpoint. Defaults to the request TTL.
One of the following: "30m" "12h" OutputText { text, type } A text output from the model.
text: string The text output from the model.
type: "output_text" The type of the output text. Always output_text.
InputImage { image_url, type, detail } An image input block used within EvalItem content arrays.
image_url: string The URL of the image input.
formaturi type: "input_image" The type of the image input. Always input_image.
detail: optional string The detail level of the image to be sent to the model. One of high, low, or auto. Defaults to auto.
ResponseInputAudio { input_audio, type } An audio input to the model.
input_audio: { data, format } data: string Base64-encoded audio data.
format: "mp3" or "wav" The format of the audio data. Currently supported formats are mp3 and
wav.
One of the following: "mp3" "wav" type: "input_audio" The type of the input item. Always input_audio.
GraderInputs = array of string or ResponseInputText or OutputText or 2 more A list of inputs, each of which may be either an input text, output text, input
image, or input audio object.
One of the following: TextInput = string A text input to the model.
ResponseInputText { text, type, prompt_cache_breakpoint } A text input to the model.
text: string The text input to the model.
type: "input_text" (default: "input_text") The type of the input item. Always input_text.
prompt_cache_breakpoint: optional { mode, ttl } Marks the exact end of a reusable prompt prefix. When ttl is omitted, the breakpoint inherits the request’s prompt_cache_options.ttl; the boundary is not rounded to a token block.
mode: "explicit" (default: "explicit") The breakpoint mode. Always explicit.
ttl: optional "30m" or "12h" The minimum lifetime of this cache breakpoint. Defaults to the request TTL.
One of the following: "30m" "12h" OutputText { text, type } A text output from the model.
text: string The text output from the model.
type: "output_text" The type of the output text. Always output_text.
InputImage { image_url, type, detail } An image input block used within EvalItem content arrays.
image_url: string The URL of the image input.
formaturi type: "input_image" The type of the image input. Always input_image.
detail: optional string The detail level of the image to be sent to the model. One of high, low, or auto. Defaults to auto.
ResponseInputAudio { input_audio, type } An audio input to the model.
input_audio: { data, format } data: string Base64-encoded audio data.
format: "mp3" or "wav" The format of the audio data. Currently supported formats are mp3 and
wav.
One of the following: "mp3" "wav" type: "input_audio" The type of the input item. Always input_audio.
role: "user" or "assistant" or "system" or "developer" The role of the message input. One of user, assistant, system, or
developer.
One of the following: "user" "assistant" "system" "developer" type: optional "message" The type of the message input. Always message.
model: string The model to use for the evaluation.
name: string The name of the grader.
type: "score_model" The object type, which is always score_model.
range: optional array of number The service requires two numbers for the score range. Defaults to [0, 1].
sampling_params: optional { max_completions_tokens, reasoning_effort, seed, 2 more } The sampling parameters for the model.
max_completions_tokens: optional integer or null The maximum number of tokens the grader model may generate in its response.
minimum1 reasoning_effort: optional ReasoningEffort or null (default: "medium") Constrains effort on reasoning for reasoning models. Currently supported
values are none, minimal, low, medium, high, xhigh, and max.
Reducing reasoning effort can result in faster responses and fewer tokens
used on reasoning in a response. Not all reasoning models support every
value. See the
reasoning guide
for model-specific support.
One of the following: "none" "minimal" "low" "medium" "high" "xhigh" "max" seed: optional integer or null A seed value to initialize the randomness, during sampling.
temperature: optional number or null A higher temperature increases randomness in the outputs.
top_p: optional number or null (default: 1) An alternative to temperature for nucleus sampling; 1.0 includes all tokens.
MultiGrader { calculate_output, graders, name, type } A MultiGrader object combines the output of multiple graders to produce a single score.
calculate_output: string A formula to calculate the output based on grader results.
graders: map[StringCheckGrader or TextSimilarityGrader or PythonGrader or 2 more] One of the following: StringCheckGrader { input, name, operation, 2 more } A StringCheckGrader object that performs a string comparison between input and reference using a specified operation.
input: string The input text. This may include template strings.
name: string The name of the grader.
operation: "eq" or "ne" or "like" or "ilike" The string check operation to perform. One of eq, ne, like, or ilike.
One of the following: "eq" "ne" "like" "ilike" reference: string The reference text. This may include template strings.
type: "string_check" The object type, which is always string_check.
TextSimilarityGrader { evaluation_metric, input, name, 2 more } A TextSimilarityGrader object which grades text based on similarity metrics.
evaluation_metric: "cosine" or "fuzzy_match" or "bleu" or 8 more The evaluation metric to use. One of cosine, fuzzy_match, bleu,
gleu, meteor, rouge_1, rouge_2, rouge_3, rouge_4, rouge_5,
or rouge_l.
One of the following: "cosine" "fuzzy_match" "bleu" "gleu" "meteor" "rouge_1" "rouge_2" "rouge_3" "rouge_4" "rouge_5" "rouge_l" input: string The text being graded.
name: string The name of the grader.
reference: string The text being graded against.
type: "text_similarity" (default: "text_similarity") The type of grader.
PythonGrader { name, source, type, image_tag } A PythonGrader object that runs a python script on the input.
name: string The name of the grader.
source: string The source code of the python script.
type: "python" The object type, which is always python.
image_tag: optional string The image tag to use for the python script.
ScoreModelGrader { input, model, name, 3 more } A ScoreModelGrader object that uses a model to assign a score to the input.
input: array of { content, role, type } The input messages evaluated by the grader. Supports text, output text, input image, and input audio content blocks, and may include template strings.
content: string or ResponseInputText or OutputText or 3 more Inputs to the model - can contain template strings. Supports text, output text, input images, and input audio, either as a single item or an array of items.
One of the following: TextInput = string A text input to the model.
ResponseInputText { text, type, prompt_cache_breakpoint } A text input to the model.
text: string The text input to the model.
type: "input_text" (default: "input_text") The type of the input item. Always input_text.
prompt_cache_breakpoint: optional { mode, ttl } Marks the exact end of a reusable prompt prefix. When ttl is omitted, the breakpoint inherits the request’s prompt_cache_options.ttl; the boundary is not rounded to a token block.
mode: "explicit" (default: "explicit") The breakpoint mode. Always explicit.
ttl: optional "30m" or "12h" The minimum lifetime of this cache breakpoint. Defaults to the request TTL.
One of the following: "30m" "12h" OutputText { text, type } A text output from the model.
text: string The text output from the model.
type: "output_text" The type of the output text. Always output_text.
InputImage { image_url, type, detail } An image input block used within EvalItem content arrays.
image_url: string The URL of the image input.
formaturi type: "input_image" The type of the image input. Always input_image.
detail: optional string The detail level of the image to be sent to the model. One of high, low, or auto. Defaults to auto.
ResponseInputAudio { input_audio, type } An audio input to the model.
input_audio: { data, format } data: string Base64-encoded audio data.
format: "mp3" or "wav" The format of the audio data. Currently supported formats are mp3 and
wav.
One of the following: "mp3" "wav" type: "input_audio" The type of the input item. Always input_audio.
GraderInputs = array of string or ResponseInputText or OutputText or 2 more A list of inputs, each of which may be either an input text, output text, input
image, or input audio object.
One of the following: TextInput = string A text input to the model.
ResponseInputText { text, type, prompt_cache_breakpoint } A text input to the model.
text: string The text input to the model.
type: "input_text" (default: "input_text") The type of the input item. Always input_text.
prompt_cache_breakpoint: optional { mode, ttl } Marks the exact end of a reusable prompt prefix. When ttl is omitted, the breakpoint inherits the request’s prompt_cache_options.ttl; the boundary is not rounded to a token block.
mode: "explicit" (default: "explicit") The breakpoint mode. Always explicit.
ttl: optional "30m" or "12h" The minimum lifetime of this cache breakpoint. Defaults to the request TTL.
One of the following: "30m" "12h" OutputText { text, type } A text output from the model.
text: string The text output from the model.
type: "output_text" The type of the output text. Always output_text.
InputImage { image_url, type, detail } An image input block used within EvalItem content arrays.
image_url: string The URL of the image input.
formaturi type: "input_image" The type of the image input. Always input_image.
detail: optional string The detail level of the image to be sent to the model. One of high, low, or auto. Defaults to auto.
ResponseInputAudio { input_audio, type } An audio input to the model.
input_audio: { data, format } data: string Base64-encoded audio data.
format: "mp3" or "wav" The format of the audio data. Currently supported formats are mp3 and
wav.
One of the following: "mp3" "wav" type: "input_audio" The type of the input item. Always input_audio.
role: "user" or "assistant" or "system" or "developer" The role of the message input. One of user, assistant, system, or
developer.
One of the following: "user" "assistant" "system" "developer" type: optional "message" The type of the message input. Always message.
model: string The model to use for the evaluation.
name: string The name of the grader.
type: "score_model" The object type, which is always score_model.
range: optional array of number The service requires two numbers for the score range. Defaults to [0, 1].
sampling_params: optional { max_completions_tokens, reasoning_effort, seed, 2 more } The sampling parameters for the model.
max_completions_tokens: optional integer or null The maximum number of tokens the grader model may generate in its response.
minimum1 reasoning_effort: optional ReasoningEffort or null (default: "medium") Constrains effort on reasoning for reasoning models. Currently supported
values are none, minimal, low, medium, high, xhigh, and max.
Reducing reasoning effort can result in faster responses and fewer tokens
used on reasoning in a response. Not all reasoning models support every
value. See the
reasoning guide
for model-specific support.
One of the following: "none" "minimal" "low" "medium" "high" "xhigh" "max" seed: optional integer or null A seed value to initialize the randomness, during sampling.
temperature: optional number or null A higher temperature increases randomness in the outputs.
top_p: optional number or null (default: 1) An alternative to temperature for nucleus sampling; 1.0 includes all tokens.
LabelModelGrader { input, labels, model, 3 more } A LabelModelGrader object which uses a model to assign labels to each item
in the evaluation.
input: array of { content, role, type } content: string or ResponseInputText or OutputText or 3 more Inputs to the model - can contain template strings. Supports text, output text, input images, and input audio, either as a single item or an array of items.
One of the following: TextInput = string A text input to the model.
ResponseInputText { text, type, prompt_cache_breakpoint } A text input to the model.
text: string The text input to the model.
type: "input_text" (default: "input_text") The type of the input item. Always input_text.
prompt_cache_breakpoint: optional { mode, ttl } Marks the exact end of a reusable prompt prefix. When ttl is omitted, the breakpoint inherits the request’s prompt_cache_options.ttl; the boundary is not rounded to a token block.
mode: "explicit" (default: "explicit") The breakpoint mode. Always explicit.
ttl: optional "30m" or "12h" The minimum lifetime of this cache breakpoint. Defaults to the request TTL.
One of the following: "30m" "12h" OutputText { text, type } A text output from the model.
text: string The text output from the model.
type: "output_text" The type of the output text. Always output_text.
InputImage { image_url, type, detail } An image input block used within EvalItem content arrays.
image_url: string The URL of the image input.
formaturi type: "input_image" The type of the image input. Always input_image.
detail: optional string The detail level of the image to be sent to the model. One of high, low, or auto. Defaults to auto.
ResponseInputAudio { input_audio, type } An audio input to the model.
input_audio: { data, format } data: string Base64-encoded audio data.
format: "mp3" or "wav" The format of the audio data. Currently supported formats are mp3 and
wav.
One of the following: "mp3" "wav" type: "input_audio" The type of the input item. Always input_audio.
GraderInputs = array of string or ResponseInputText or OutputText or 2 more A list of inputs, each of which may be either an input text, output text, input
image, or input audio object.
One of the following: TextInput = string A text input to the model.
ResponseInputText { text, type, prompt_cache_breakpoint } A text input to the model.
text: string The text input to the model.
type: "input_text" (default: "input_text") The type of the input item. Always input_text.
prompt_cache_breakpoint: optional { mode, ttl } Marks the exact end of a reusable prompt prefix. When ttl is omitted, the breakpoint inherits the request’s prompt_cache_options.ttl; the boundary is not rounded to a token block.
mode: "explicit" (default: "explicit") The breakpoint mode. Always explicit.
ttl: optional "30m" or "12h" The minimum lifetime of this cache breakpoint. Defaults to the request TTL.
One of the following: "30m" "12h" OutputText { text, type } A text output from the model.
text: string The text output from the model.
type: "output_text" The type of the output text. Always output_text.
InputImage { image_url, type, detail } An image input block used within EvalItem content arrays.
image_url: string The URL of the image input.
formaturi type: "input_image" The type of the image input. Always input_image.
detail: optional string The detail level of the image to be sent to the model. One of high, low, or auto. Defaults to auto.
ResponseInputAudio { input_audio, type } An audio input to the model.
input_audio: { data, format } data: string Base64-encoded audio data.
format: "mp3" or "wav" The format of the audio data. Currently supported formats are mp3 and
wav.
One of the following: "mp3" "wav" type: "input_audio" The type of the input item. Always input_audio.
role: "user" or "assistant" or "system" or "developer" The role of the message input. One of user, assistant, system, or
developer.
One of the following: "user" "assistant" "system" "developer" type: optional "message" The type of the message input. Always message.
labels: array of string The labels to assign to each item in the evaluation.
model: string The model to use for the evaluation. Must support structured outputs.
name: string The name of the grader.
passing_labels: array of string The labels that indicate a passing result. Must be a subset of labels.
type: "label_model" The object type, which is always label_model.
name: string The name of the grader.
type: "multi" (default: "multi") The object type, which is always multi.
hyperparameters: optional ReinforcementHyperparameters { batch_size, compute_multiplier, eval_interval, 4 more } The hyperparameters used for the reinforcement fine-tuning job.
batch_size: optional "auto" or number (default: "auto") Number of examples in each batch. A larger batch size means that model parameters are updated less frequently, but with lower variance.
One of the following: "auto" number minimum1 maximum256 compute_multiplier: optional "auto" or number (default: "auto") Multiplier on amount of compute used for exploring search space during training.
One of the following: "auto" number exclusiveMinimum0.00001 maximum10 eval_interval: optional "auto" or number (default: "auto") The number of training steps between evaluation runs.
One of the following: "auto" number minimum1 eval_samples: optional "auto" or number (default: "auto") Number of evaluation samples to generate per training step.
One of the following: "auto" number minimum1 learning_rate_multiplier: optional "auto" or number (default: "auto") Scaling factor for the learning rate. A smaller learning rate may be useful to avoid overfitting.
One of the following: "auto" number exclusiveMinimum0 n_epochs: optional "auto" or number (default: "auto") The number of epochs to train the model for. An epoch refers to one full cycle through the training dataset.
One of the following: "auto" number minimum1 maximum50 reasoning_effort: optional "default" or "low" or "medium" or "high" (default: "default") Level of reasoning effort.
One of the following: "default" "low" "medium" "high" supervised: optional SupervisedMethod { hyperparameters } Configuration for the supervised fine-tuning method.
hyperparameters: optional SupervisedHyperparameters { batch_size, learning_rate_multiplier, n_epochs } The hyperparameters used for the fine-tuning job.
batch_size: optional "auto" or number (default: "auto") Number of examples in each batch. A larger batch size means that model parameters are updated less frequently, but with lower variance.
One of the following: "auto" number minimum1 maximum256 learning_rate_multiplier: optional "auto" or number (default: "auto") Scaling factor for the learning rate. A smaller learning rate may be useful to avoid overfitting.
One of the following: "auto" number exclusiveMinimum0 n_epochs: optional "auto" or number (default: "auto") The number of epochs to train the model for. An epoch refers to one full cycle through the training dataset.
One of the following: "auto" number minimum1 maximum50
Returns Examples
Response example
{
"object": "fine_tuning.job",
"id": "ftjob-abc123",
"model": "o4-mini",
"created_at": 1721764800,
"fine_tuned_model": null,
"organization_id": "org-123",
"result_files": [],
"status": "pausing",
"validation_file": "file-123",
"training_file": "file-abc",
"error": {},
"finished_at": null,
"seed": null,
"trained_tokens": null,
"method": {
"type": "reinforcement",
"reinforcement": {
"grader": {
"type": "string_check",
"name": "Example string check grader",
"input": "{{sample.output_text}}",
"reference": "{{item.label}}",
"operation": "eq"
},
"hyperparameters": {
"reasoning_effort": "medium"
}
}
}
}