# How to access model customization
<a name="customizing-models-access"></a>

Amazon SageMaker AI provides multiple interfaces for model customization. The following table summarizes your options — choose the interface that best fits your workflow and level of control.


| Interface | Infrastructure | Best for | Control level | 
| --- | --- | --- | --- | 
| Studio UI | [Serverless](customize-model.md) | Quick experimentation, no-code users | Guided — select model, technique, upload data | 
| Python SDK (SFTTrainer, etc.) | [Serverless](customize-model.md) | Reproducible pipelines, automation | Programmatic — specify all params in code | 
| Python SDK (ModelTrainer) | [SageMaker AI Training Jobs](customizing-models-training-jobs.md) | Custom instances, advanced configs | Full — choose instance type, count, image | 
| Recipes | [Training Jobs](customizing-models-training-jobs.md) \+ [HyperPod](customizing-models-hyperpod.md) | Pre-configured best practices | Recipe-driven — modify YAML config | 
| HP-CLI | [HyperPod](customizing-models-hyperpod.md) (EKS) | Kubernetes-native teams | Full — Kubernetes job submission | 
| Agent (Model customization skills) | [Serverless](customize-model.md) | Natural language workflow, guided experience | Conversational — describe what you need | 
| Python SDK (ServiceClient) | Serverless ([Training session APIs](customizing-models-training-sessions.md)) | Custom training loops, multi-turn reinforcement learning (RL), multi-stage pipelines | Operation-level — orchestrate forward\_backward(), sample(), optim\_step() as API calls | 

## Studio UI
<a name="access-studio-ui"></a>

The SageMaker AI Studio console provides visual workflows for model customization without writing code.

**Customize with UI**  
The primary entry point for serverless model customization. Walk through a guided workflow:  

1. Select a foundation model from the catalog

1. Choose a customization technique (SFT, DPO, RLVR, RLAIF)

1. Select a training type (LoRA or FFT)

1. Upload or specify your training dataset (S3 path)

1. Configure hyperparameters

1. Launch the training job
Monitor progress in real time with live metrics and logs.

**JumpStart model hub**  
Browse and discover foundation models available for customization. Filter by provider, task, or model size. From a model detail card, you can launch customization directly.

**Agent-Guided experience**  
Use Amazon SageMaker AI Skills for a conversational approach to model customization. Describe what you want to achieve in natural language, and the agent guides you through model selection, dataset preparation, and job configuration. For more information, see [Model customization agent skills](https://docs.aws.amazon.com/sagemaker/latest/dg/model-customize-agent-skills.html).

## Python SDK V3
<a name="access-python-sdk"></a>

Programmatic access for serverless and SageMaker AI Training Jobs workflows using the [SageMaker AI Python SDK](https://sagemaker.readthedocs.io/en/stable/model_customization/index.html).

### Serverless trainers
<a name="access-serverless-trainers"></a>

Use technique-specific trainer classes for fully managed customization:

```
from sagemaker.train import SFTTrainer, DPOTrainer, RLVRTrainer, RLAIFTrainer

# Example: Serverless SFT
trainer = SFTTrainer(
    model_id="meta-textgeneration-llama-3-1-8b-instruct",
    train_data="s3://my-bucket/train.jsonl",
    hyperparameters={
        "max_epochs": 3,
        "learning_rate": 5e-5,
        "lora_rank": 16,
    }
)
trainer.train()
```

### SageMaker AI Training Jobs
<a name="access-modeltrainer"></a>

Use `ModelTrainer` for full control over instance selection:

```
from sagemaker.train import ModelTrainer
from sagemaker.train.configs import Compute, InputData

trainer = ModelTrainer(
    training_image="my-training-image",
    role="arn:aws:iam::123456789012:role/SageMakerRole",
    compute=Compute(instance_type="ml.p4d.24xlarge", instance_count=2)
)
trainer.train(input_data_config=[
    InputData(channel_name="training", data_source="s3://my-bucket/train")
])
```

## Training session APIs
<a name="access-training-sessions"></a>

Use the SageMaker AI Training session APIs when a predefined workflow is too rigid. Instead of submitting a complete training script, you open a stateful session and call training operations — `forward_backward()`, `optim_step()`, `sample()` — as individual API calls, then compose them into whatever procedure your technique requires. The model stays loaded between calls, so you can write custom loss functions, run multi-turn reinforcement learning with elastic rollouts, and chain multiple procedures in one session. SageMaker AI manages the underlying infrastructure and tenant isolation.

```
from sagemaker.train.training_session import ServiceClient, AdamParams

svc = ServiceClient(region="us-west-2", role_arn="arn:aws:iam::111122223333:role/MyRole")
tc = svc.create_lora_training_client(base_model="openai/gpt-oss-20b", rank=32, alpha=64)

for batch in data_loader:
    tc.forward_backward(batch, loss_fn="cross_entropy")
    tc.optim_step(AdamParams(learning_rate=5e-5))

print(tc.save_state().result().model_package_arn)
tc.stop()
```

For more information, see [Training session APIs](customizing-models-training-sessions.md).

## Recipes (Training Jobs \+ HyperPod)
<a name="access-launcher"></a>

Run pre-configured training recipes on [SageMaker AI Training Jobs](customizing-models-training-jobs.md) or [HyperPod](customizing-models-hyperpod.md) clusters. Recipes handle distributed training configuration, dataset loading, and checkpoint management automatically.

Recipes are available for all supported models and techniques. See [SageMaker AI HyperPod recipes](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-hyperpod-recipes.html) for the full catalog.

```
git clone --recursive https://github.com/aws/sagemaker-hyperpod-recipes.git
cd sagemaker-hyperpod-recipes
pip3 install -r requirements.txt

# Launch a fine-tuning recipe
python3 launcher/launch.py --recipe recipes_collection/recipes/fine-tuning/llama/sft_lora.yaml
```

## HP-CLI
<a name="access-hp-cli"></a>

Command-line interface for submitting training jobs to HyperPod EKS clusters. HP-CLI integrates with Kubernetes-based workload orchestration for containerized training on persistent clusters.

For setup and usage, see [HyperPod with Amazon EKS](https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-hyperpod-eks.html).