Sign in
Fine-tuning

List job

List the metadata for a single fine-tuning job.

get/fine-tunes/{id}

Path parameters

idstring required

The ID of the job to retrieve

Response

Fine-tune job details retrieved successfully

idstring uuid required
training_filestring
validation_filestring
modelstring
model_output_namestring
model_output_pathstring
tokenized_dataset_pathstring

Storage path for the tokenized dataset archive generated for this fine-tune job.

tokenized_dataset_uploaded_atstring date-time

Timestamp when the tokenized dataset archive was uploaded.

model_object_idstring

Together model registry object ID for the final model weights (e.g. ml_...).

model_object_namestring

Together model registry name for the final model weights, formatted as <project_slug>/<model_name>.

model_object_revision_idstring

Together model registry revision ID for the final model weights (e.g. rv_...).

adapter_object_idstring

Together model registry object ID for the final adapter weights on LoRA jobs.

adapter_object_namestring

Together model registry name for the final adapter weights on LoRA jobs, formatted as <project_slug>/<model_name>-adapter.

adapter_object_revision_idstring

Together model registry revision ID for the final adapter weights on LoRA jobs.

user_idstring required

ID of the user who created the fine-tune job.

trainingfile_numlinesinteger
trainingfile_sizeinteger
created_atstring date-time
updated_atstring date-time
started_atstring date-time
n_epochsinteger
n_checkpointsinteger
n_evalsinteger
learning_ratenumber
warmup_rationumber
max_grad_normnumber
weight_decaynumber
eval_stepsinteger
status'pending' | 'queued' | 'running' | 'compressing' | 'uploading' | 'cancel_requested' | 'cancelled' | 'error' | 'completed' required
job_idstring
token_countinteger
param_countinteger
total_priceinteger
epochs_completedinteger
queue_depthinteger
wandb_project_namestring
wandb_urlstring
from_checkpointstring
from_hf_modelstring
hf_model_revisionstring
early_stoppedboolean

Whether the early-stopping criterion triggered.

early_stopping_best_stepinteger

Step associated with the selected early-stopping artifact. When early_stopping_best_metric is null, no finite best metric was recorded; this is the halt step, not a best-checkpoint step.

early_stopping_best_metricnumber nullable

Best validation loss observed, corresponding to early_stopping_best_step. Null if no improving evaluation was recorded (for example, a non-finite first evaluation).

Changes

Changed in 2 of the 14 revisions of this API.1119

  • 43e29c4cb27913See the full diff
    • ●

      added the new tokenized_dataset_upload_complete enum value to the events/items/type response property for the response status 200

      response-property-enum-value-added

    • ○

      added the optional property events/items/tokenized_dataset_path to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property tokenized_dataset_path to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property tokenized_dataset_uploaded_at to the response with the 200 status

      response-optional-property-added

  • 0c4fdad77435116See the full diff
    • ▲

      the early_stopping_best_metric response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the events/items/early_stopping_best_metric_value response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the lr_scheduler/lr_scheduler_args/oneOf[CosineLRSchedulerArgs]/min_lr_ratio response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the lr_scheduler/lr_scheduler_args/oneOf[CosineLRSchedulerArgs]/num_cycles response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the lr_scheduler/lr_scheduler_args/oneOf[LinearLRSchedulerArgs]/min_lr_ratio response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the max_grad_norm response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the training_method/oneOf[TrainingMethodDPO]/dpo_beta response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the training_method/oneOf[TrainingMethodDPO]/rpo_alpha response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the training_method/oneOf[TrainingMethodDPO]/simpo_gamma response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the training_type/oneOf[LoRATrainingType]/lora_dropout response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ▲

      the weight_decay response's property type/format changed from number/float to number/ for status 200

      response-property-type-changed

    • ○

      added the optional property adapter_object_id to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property adapter_object_name to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property adapter_object_revision_id to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property model_object_id to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property model_object_name to the response with the 200 status

      response-optional-property-added

    • ○

      added the optional property model_object_revision_id to the response with the 200 status

      response-optional-property-added

    This revision also has 2 changes that name no endpoint, such as unreferenced schemas being removed. See the revision's changelog