Error creating fine-tuning job

[jwerre]

The documentation doesn't really clarify how to provide the arguments when creating fine-tuning jobs.

Can you help me resolve the following error

SDKError: API error occurred: Status 422 Content-Type application/json; charset=utf-8 Body 
{"detail": [{"type": "value_error", "loc": ["body", "job_in"], "msg": "Value error, At least one training file or repository is required.", "ctx": {"error": "At least one training file or repository is required."}}]}
    at matchFunc (node_modules/@mistralai/mistralai/lib/matchers.js:132:28)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async fineTuningJobsCreate (node_modules/@mistralai/mistralai/funcs/fineTuningJobsCreate.js:91:22)
    at async unwrapAsync (node_modules/@mistralai/mistralai/types/fp.js:31:15)
    at async _createFineTuningModel (file://lib/support_model.js:247:20) 

The function I'm using looks something like this:

async function _createFineTuningModel(id, options = {}) {
	let mistralApiKey = 'XXX';

	const mistral = new Mistral({
		apiKey: mistralApiKey,
	});

	const hyperparameters = {
		epochs: options.epochs || 3,
		batch_size: options.batchSize || 8,
		learning_rate: options.learningRate || 5e-5,
		weight_decay: options.weightDecay || 0.01,
		warmup_steps: options.warmupSteps || 500,
		gradient_accumulation_steps: options.gradientAccumulationSteps || 1,
	};

	const payload = {
		dryRun: options.dryRun || false,
		model: options.model || 'mistral-large-latest',
		trainingFiles: [id],
		hyperparameters,
	};


	const tuningRes = await mistral.fineTuning.jobs.create(payload);

	return tuningRes;
}

What is the correct structure of the arguments for this function?

[sophiamyang]

Thanks for reporting! There is a bug in our docs and we will update the docs soon. Please use the following instead.

const createdJob = await client.fineTuning.jobs.create({
    model: 'open-mistral-7b',
    trainingFiles: [{fileId: training_data.id, weight: 1}],
    validationFiles: [validation_data.id],
    hyperparameters: {
      trainingSteps: 10,
      learningRate: 0.0001,
    },
    autoStart:false,
  });
console.log("Created fine-tuning job:", createdJob);

[jwerre]

Thanks. I was able to find this in the examples directory. Can you also clarify how dry_run works for sanity checks?

[sophiamyang]

Hi please don't use dry_run. You can just use auto_start=False to just create the job without starting it.