> ## Documentation Index > Fetch the complete documentation index at: https://togetherai-migration.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Create Job > Use a model to create a fine-tuning job. ## OpenAPI ````yaml POST /fine-tunes openapi: 3.1.0 info: title: Together APIs description: The Together REST API. Please see https://docs.together.ai for more details. version: 2.0.0 termsOfService: https://www.together.ai/terms-of-service contact: name: Together Support url: https://www.together.ai/contact license: name: MIT url: https://github.com/togethercomputer/openapi/blob/main/LICENSE servers: - url: https://api.together.xyz/v1 security: - bearerAuth: [] paths: /fine-tunes: post: tags: - Fine-tuning summary: Create job description: Use a model to create a fine-tuning job. requestBody: required: true content: application/json: schema: type: object required: - training_file - model properties: training_file: type: string description: File-ID of a training file uploaded to the Together API validation_file: type: string description: File-ID of a validation file uploaded to the Together API model: type: string description: Name of the base model to run fine-tune job on n_epochs: type: integer default: 1 description: >- Number of complete passes through the training dataset (higher values may improve results but increase cost and risk of overfitting) n_checkpoints: type: integer default: 1 description: >- Number of intermediate model versions saved during training for evaluation n_evals: type: integer default: 0 description: >- Number of evaluations to be run on a given validation set during training batch_size: oneOf: - type: integer - type: string enum: - max default: max description: >- Number of training examples processed together (larger batches use more memory but may train faster). Defaults to "max". We use training optimizations like packing, so the effective batch size may be different than the value you set. learning_rate: type: number format: float default: 0.00001 description: >- Controls how quickly the model adapts to new information (too high may cause instability, too low may slow convergence) lr_scheduler: $ref: '#/components/schemas/LRScheduler' type: object default: none description: >- The learning rate scheduler to use. It specifies how the learning rate is adjusted during training. warmup_ratio: type: number format: float default: 0 description: >- The percent of steps at the start of training to linearly increase the learning rate. max_grad_norm: type: number format: float default: 1 description: >- Max gradient norm to be used for gradient clipping. Set to 0 to disable. weight_decay: type: number format: float default: 0 description: Weight decay. Regularization parameter for the optimizer. suffix: type: string description: Suffix that will be added to your fine-tuned model name wandb_api_key: type: string description: >- Integration key for tracking experiments and model metrics on W&B platform wandb_base_url: type: string description: The base URL of a dedicated Weights & Biases instance. wandb_project_name: type: string description: >- The Weights & Biases project for your run. If not specified, will use `together` as the project name. wandb_name: type: string description: The Weights & Biases name for your run. train_on_inputs: oneOf: - type: boolean - type: string enum: - auto type: boolean default: auto description: >- Whether to mask the user messages in conversational data or prompts in instruction data. training_method: type: object oneOf: - $ref: '#/components/schemas/TrainingMethodSFT' - $ref: '#/components/schemas/TrainingMethodDPO' description: >- The training method to use. 'sft' for Supervised Fine-Tuning or 'dpo' for Direct Preference Optimization. training_type: type: object oneOf: - $ref: '#/components/schemas/FullTrainingType' - $ref: '#/components/schemas/LoRATrainingType' from_checkpoint: type: string description: >- The checkpoint identifier to continue training from a previous fine-tuning job. Format is `{$JOB_ID}` or `{$OUTPUT_MODEL_NAME}` or `{$JOB_ID}:{$STEP}` or `{$OUTPUT_MODEL_NAME}:{$STEP}`. The step value is optional; without it, the final checkpoint will be used. responses: '200': description: Fine-tuning job initiated successfully content: application/json: schema: $ref: '#/components/schemas/FinetuneResponse' components: schemas: LRScheduler: type: object properties: lr_scheduler_type: type: string enum: - linear - cosine lr_scheduler_args: oneOf: - $ref: '#/components/schemas/LinearLRSchedulerArgs' - $ref: '#/components/schemas/CosineLRSchedulerArgs' required: - lr_scheduler_type TrainingMethodSFT: type: object properties: method: type: string enum: - sft required: - method TrainingMethodDPO: type: object properties: method: type: string enum: - dpo dpo_beta: type: number format: float default: 0.1 required: - method FullTrainingType: type: object properties: type: type: string enum: - Full required: - type LoRATrainingType: type: object properties: type: type: string enum: - Lora lora_r: type: integer lora_alpha: type: integer lora_dropout: type: number format: float default: 0 lora_trainable_modules: type: string default: all-linear required: - type - lora_r - lora_alpha FinetuneResponse: type: object required: - id - status properties: id: type: string format: uuid training_file: type: string validation_file: type: string model: type: string model_output_name: type: string model_output_path: type: string trainingfile_numlines: type: integer trainingfile_size: type: integer created_at: type: string updated_at: type: string n_epochs: type: integer n_checkpoints: type: integer n_evals: type: integer batch_size: oneOf: - type: integer - type: string enum: - max default: max learning_rate: type: number lr_scheduler: $ref: '#/components/schemas/LRScheduler' type: object warmup_ratio: type: number max_grad_norm: type: number format: float weight_decay: type: number format: float eval_steps: type: integer train_on_inputs: oneOf: - type: boolean - type: string enum: - auto default: auto training_method: type: object oneOf: - $ref: '#/components/schemas/TrainingMethodSFT' - $ref: '#/components/schemas/TrainingMethodDPO' training_type: type: object oneOf: - $ref: '#/components/schemas/FullTrainingType' - $ref: '#/components/schemas/LoRATrainingType' status: $ref: '#/components/schemas/FinetuneJobStatus' job_id: type: string events: type: array items: $ref: '#/components/schemas/FineTuneEvent' token_count: type: integer param_count: type: integer total_price: type: integer epochs_completed: type: integer queue_depth: type: integer wandb_project_name: type: string wandb_url: type: string from_checkpoint: type: string LinearLRSchedulerArgs: type: object properties: min_lr_ratio: type: number format: float default: 0 description: The ratio of the final learning rate to the peak learning rate CosineLRSchedulerArgs: type: object properties: min_lr_ratio: type: number format: float default: 0 description: The ratio of the final learning rate to the peak learning rate num_cycles: type: number format: float default: 0.5 description: Number or fraction of cycles for the cosine learning rate scheduler FinetuneJobStatus: type: string enum: - pending - queued - running - compressing - uploading - cancel_requested - cancelled - error - completed FineTuneEvent: type: object required: - object - created_at - message - type - param_count - token_count - total_steps - wandb_url - step - checkpoint_path - model_path - training_offset - hash properties: object: type: string enum: - fine-tune-event created_at: type: string level: anyOf: - $ref: '#/components/schemas/FinetuneEventLevels' message: type: string type: $ref: '#/components/schemas/FinetuneEventType' param_count: type: integer token_count: type: integer total_steps: type: integer wandb_url: type: string step: type: integer checkpoint_path: type: string model_path: type: string training_offset: type: integer hash: type: string FinetuneEventLevels: type: string enum: - null - info - warning - error - legacy_info - legacy_iwarning - legacy_ierror FinetuneEventType: type: string enum: - job_pending - job_start - job_stopped - model_downloading - model_download_complete - training_data_downloading - training_data_download_complete - validation_data_downloading - validation_data_download_complete - wandb_init - training_start - checkpoint_save - billing_limit - epoch_complete - training_complete - model_compressing - model_compression_complete - model_uploading - model_upload_complete - job_complete - job_error - cancel_requested - job_restarted - refund - warning securitySchemes: bearerAuth: type: http scheme: bearer x-bearer-format: bearer x-default: default ````