> ## Documentation Index
> Fetch the complete documentation index at: https://docs.portkey.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Fine-tuning

> Run your fine-tuning jobs with Prisma AIRS AI Gateway

<Note>
  [Data Service](/docs/aigw/changelog/data-service) must be enabled for **AI Gateway Managed Fine-tuning** or when **cost-attribution** is required.
</Note>

AI Gateway supports fine-tuning in two ways:

1. Fine-tuning through the AI Gateway supports multiple providers with a single unified API.
2. Fine-tuning with the AI Gateway as a client, supports following providers:
   * [OpenAI](/docs/aigw/integrations/llms/openai/fine-tuning)
   * [Bedrock](/docs/aigw/integrations/llms/bedrock/fine-tuning)
   * [Azure OpenAI](/docs/aigw/integrations/llms/azure-openai/fine-tuning)
   * [Fireworks](/docs/aigw/integrations/llms/fireworks/fine-tuning)
   * [Vertex](/docs/aigw/integrations/llms/vertex-ai/fine-tuning)

## 1. Fine-tuning with the AI Gateway as a client

With the AI Gateway acting as a client, gives you the following benefits:

1. Provider specific fine-tuning.
2. More control over the fine-tuning process.
3. Unified API for all providers.

## 2. Fine-tuning with the AI Gateway as a provider

With the AI Gateway acting as provider, gives you the following benefits:

1. Unified API for all providers
2. No need to manage multiple endpoints and keys
3. Easy to switch between providers with limited changes
4. Easier integration with the AI Gateway's other features like batching, etc.

## How to use fine-tuning

The AI Gateway supports a Unified Fine-tuning API for all providers which is based on OpenAI's Fine-tuning API.

While the signature of API is same for all providers, there are places where a specific provider require some more information which can be passed via headers if used as a client, or by `portkey_options` params if used as a provider.

### Fine-tuning with the AI Gateway as a client

**Upload a File**

```sh theme={"system"}
curl -X POST --header 'Authorization: Bearer <portkey_api_key>' \
 --form 'file=@dataset.jsonl' \
 --form 'purpose=fine-tune' \
 'https://aigw.portkey.ai/v1/files'
```

<Info>Learn more about [files](/docs/aigw/product/ai-gateway/files)</Info>

**Create a Fine-tuning Job**

```sh theme={"system"}
curl -X POST --header 'Content-Type: application/json' \
 --header 'Authorization: Bearer <portkey_api_key>' \
 --data \
 $'{"model": "<base_model>", "suffix": "<finetune_name>", "training_file": "<file_id>", "portkey_options": {"x-portkey-provider": "@openai-prod"}, "method": {"type": "supervised", "supervised": {"hyperparameters": {"n_epochs": 1}}}}\n' \
'https://aigw.portkey.ai/v1/fine_tuning/jobs'
```

<Tip>
  `portkey_options` supports all the options that are supported by each provider
  in the gateway. Values can be provider specific, refer to provider's
  documentation for each provider's specific options required for fine-tuning.
</Tip>

**List Fine-tuning Jobs**

```sh theme={"system"}
curl -X GET --header 'Authorization: Bearer <portkey_api_key>' \
'https://aigw.portkey.ai/v1/fine_tuning/jobs'
```

**Get Fine-tuning Job**

```sh theme={"system"}
curl -X GET --header 'Authorization: Bearer <portkey_api_key>' \
'https://aigw.portkey.ai/v1/fine_tuning/jobs/<job_id>'
```

**Cancel Fine-tuning Job**

```sh theme={"system"}
curl -X POST --header 'Authorization: Bearer <portkey_api_key>' \
'https://aigw.portkey.ai/v1/fine_tuning/jobs/<job_id>/cancel'
```

When you use the AI Gateway as a provider for fine-tuning, everything is managed by the AI Gateway hosted solution, this includes job submission with provider, monitoring job status and able to use the fine-tuned model with the AI Gateway's Prompt Playground.

<Note>
  Providers like OpenAI, Azure OpenAPI does provider a straight forward approach
  to use the fine-tuned model for inference whereas providers like `Bedrock` abd
  `Vertex` does require you to manually provide compute which is hard for
  the AI Gateway to manage.
</Note>

**Notes**:

* For `Bedrock`, `model` param should point to the `modelId` from the AI Gateway and the `modelId` to submit to the provider should be passed via `provider_options`. Example is available below.

```sh theme={"system"}
  curl -X POST --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <portkey_api_key>' \
  --data \
  $'{"model": "amazon.titan-text-lite-v1:0", "suffix": "<finetune_name>", "training_file": "<file_id>", "provider_options": {"model": "amazon.titan-text-lite-v1:0:4k"}, "method": {"type": "supervised", "supervised": {"hyperparameters": {"n_epochs": 1}}}}\n' \
  'https://aigw.portkey.ai/v1/fine_tuning/jobs'
```

> For more info about `modelId`, please refer to the provider's documentation.

Please refer to the provider's documentation for fine-tuning with the AI Gateway as a gateway.

1. [OpenAI](/docs/aigw/integrations/llms/openai/fine-tuning)
2. [Bedrock](/docs/aigw/integrations/llms/bedrock/fine-tuning)
3. [Azure OpenAI](/docs/aigw/integrations/llms/azure-openai/fine-tuning)
4. [Fireworks](/docs/aigw/integrations/llms/fireworks/fine-tuning)
5. [Vertex](/docs/aigw/integrations/llms/vertex-ai/fine-tuning)


## Related topics

- [OpenAI](/docs/aigw/integrations/llms/openai.md)
- [List fine tuning events](/docs/aigw/api-reference/fine-tuning/list-fine-tuning-events.md)
- [Cancel fine tuning job](/docs/aigw/api-reference/fine-tuning/cancel-fine-tuning-job.md)
- [Retrieve fine tuning job](/docs/aigw/api-reference/fine-tuning/retrieve-fine-tuning-job.md)
- [Create fine tuning job](/docs/aigw/api-reference/fine-tuning/create-fine-tuning-job.md)