本文為英文版的機器翻譯版本，如內容有任何歧義或不一致之處，概以英文版為準。

# Amazon Nova 配方
<a name="nova-model-recipes"></a>

您可以從配方儲存庫取得 Amazon Nova [ 配方](https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes)。Amazon Nova 配方是 [YAML](https://yaml.org/) 組態檔案，可提供有關如何執行模型自訂任務的詳細資訊給 SageMaker。它提供基本模型名稱、設定訓練超參數、定義最佳化設定，並包含成功微調或訓練模型所需的任何其他選項。

您也可以透過 Amazon SageMaker Studio 和導覽至模型中樞 AWS、選取和瀏覽 Amazon Nova 模型以尋找其相關聯的配方，來存取 Amazon Nova 配方。Amazon SageMaker Studio 和 都會為每個配方提供範例筆記本，其中包括修改配方並使用 SageMaker 訓練任務或 SageMaker HyperPod 環境執行自訂任務的所有必要步驟。

若要存取 Amazon SageMaker Studio 中的配方頁面，執行角色必須具有下列許可。

------
#### [ JSON ]

****  

```
{
    "Version":"2012-10-17",		 	 	 
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "s3:GetObject"
            ],
            "Resource": [
                "arn:aws:s3:::*model-customization-recipes*"
            ]
        }
    ]
}
```

------

若要在 SageMaker 訓練任務 和 上執行範例筆記本，請使用下列其中一個 SageMaker 分佈映像版本：`2.7.1+`、`2.8.0+`、`3.2.1+`、`3.3.0+`。這同時適用於 Amazon SageMaker Studio 和 。

**Topics**
+ [取得 Amazon Nova 配方](#nova-model-get-recipes)
+ [取得 Amazon Nova Forge 配方](#nova-model-get-forge-recipes)
+ [可用的模型和演算法](#nova-model-algorithm)
+ [Amazon Nova Lite](#nova-model-recipes-reference-novalite)
+ [Amazon Nova Micro](#nova-model-recipes-reference-novamicro)
+ [Amazon Nova Pro](#nova-model-recipes-reference-novapro)

## 取得 Amazon Nova 配方
<a name="nova-model-get-recipes"></a>

若要取得基本 Amazon Nova 配方，請執行下列命令來複製[配方](https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes)儲存庫。

```
git clone https://github.com/aws/sagemaker-hyperpod-recipes.git
```

基本配方可在 [https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes/](https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes/) 取得。

```
cd recipes_collection/recipes/
```

Amazon Nova 自訂配方位於下列資料夾中。


****  

| 配方類型  | 資料夾 | 
| --- | --- | 
| SFT (全秩和 PEFT)、PPO、DPO (全秩和 PEFT) | [fine-tuning/nova](https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes/fine-tuning/nova) | 
| 評估 | [evaluation/nova](https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes/evaluation/nova) | 
| CPT | [training/nova](https://github.com/aws/sagemaker-hyperpod-recipes/tree/main/recipes_collection/recipes/training) | 

## 取得 Amazon Nova Forge 配方
<a name="nova-model-get-forge-recipes"></a>

若要存取任務專用的 Amazon Nova Forge 配方，請[設定您對 Amazon Nova Forge 的存取權](nova-forge-access.md)，然後遵循[此研討會](https://catalog.us-east-1.prod.workshops.aws/workshops/dcac6f7a-3c61-4978-8344-7535526bf743/en-US/02-smhp-rig/03-nova-forge)來設定 CLI 的 Forge 版本。對於 SageMaker 訓練任務， AWS 主控台和 Nova Forge SDK 會自動存取 Forge 配方。

## 可用的模型和演算法
<a name="nova-model-algorithm"></a>

下表摘要說明使用 SageMaker 自訂 Amazon Nova 模型和支援演算法的可用性。


| 模型名稱 | 模型 ID | 微調 | 備註 | 
| --- |--- |--- |--- |
| Amazon Nova Lite 2.0 | amazon.nova-2-lite-v1：0：256k | 是 |    SFT - 接受文字和/或影像或文字和/或影片做為輸入，並產生文字做為輸出。單一任務無法在同一個執行中結合文字、影像和影片。   DPO - 接受文字和影像做為輸入，並產生文字做為輸出。   RFT - 接受文字和影像作為單轉對話的輸入，並產生文字作為輸出。改善獎勵最佳化功能。    | 

## Amazon Nova Lite
<a name="nova-model-recipes-reference-novalite"></a>

下表列出 Amazon Nova Lite 配方參考的詳細資訊。


****  

| 模型 | 類別/子類別 | 技術 | 配方名稱 | 映像 URI （訓練任務） | 映像 URI (SageMaker HyperPod) | 運算執行個體 | 
| --- | --- | --- | --- | --- | --- | --- | 
| Nova Lite | 訓練/微調 | 監督式微調 (LoRA) |  `nova_lite_1_0_g5_g6_12x_gpu_lora_sft.yaml` `nova_lite_1_0_g5_g6_48x_gpu_lora_sft.yaml` `nova_lite_1_0_p5_p4d_gpu_lora_sft.yaml`  | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-SFT-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-SFT-latest |  `ml.p5.48xlarge`, `ml.p5en.48xlarge`, `ml.g5.12xlarge`, `ml.g6.12xlarge`, `ml.g5.48xlarge`, `ml.g6.48xlarge`  | 
| Nova Lite | 訓練/微調 | 監督式微調 (完整) | nova\$1lite\$11\$10\$1p5\$1p4d\$1gpu\$1sft.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-SFT-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-SFT-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Lite | 訓練/微調 | 直接偏好最佳化 (完整) | nova\$1lite\$11\$10\$1p5\$1p4d\$1gpu\$1dpo.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-DPO-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-DPO-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Lite | 訓練/微調 | 直接偏好最佳化 (LoRA) |  `nova_lite_1_0_p5_p4d_gpu_lora_dpo.yaml` `nova_lite_1_0_g5_g6_48x_gpu_lora_dpo.yaml`  | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-DPO-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-DPO-latest |  `ml.p5.48xlarge`, `ml.p5en.48xlarge`, `ml.g5.48xlarge`, `ml.g6.48xlarge`  | 
| Nova Lite | 訓練/強化學習 | 強化微調 (RFT) | nova\$1lite\$11\$10\$1p5\$1gpu\$1ppo.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SMHP-PPO-TRAIN-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Lite | 訓練/持續預先訓練 | 持續預先訓練 (基本模型) | nova\$1lite\$1gpu\$1p5x16\$1pretrain.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:HP-CPT-latest | ml.p5.48xlarge | 
| Nova Lite | 評估/評估 | 標準文字基準 | nova\$1lite\$1p5\$148xl\$1general\$1text\$1benchmark\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Lite | 評估/評估 | 自訂資料集評估 | nova\$1lite\$1p5\$148xl\$1bring\$1your\$1own\$1dataset\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Lite | 評估/評估 | 多模態基準 | nova\$1lite\$1p5\$148\$1general\$1multi\$1modal\$1benchmark\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Lite | 評估/評估 | LLM 即評審 | nova\$1lite\$1p5\$148xl\$1llm\$1judge\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 

## Amazon Nova Micro
<a name="nova-model-recipes-reference-novamicro"></a>

下表列出 Amazon Nova Micro 配方參考的詳細資訊。


****  

| 模型 | 類別/子類別 | 技術 | 配方名稱 | 映像 URI （訓練任務） | 映像 URI (SageMaker HyperPod) | 運算執行個體 | 
| --- | --- | --- | --- | --- | --- | --- | 
| Nova Micro | 訓練/微調 | 監督式微調 (LoRA) |  `nova_micro_1_0_p5_p4d_gpu_lora_sft.yaml` `nova_micro_1_0_g5_g6_12x_gpu_lora_sft.yaml` `nova_micro_1_0_g5_g6_48x_gpu_lora_sft.yaml`  | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-SFT-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-SFT-latest |  `ml.p5.48xlarge`, `ml.p5en.48xlarge`, `ml.p4d.24xlarge`, `ml.g5.12xlarge`, `ml.g6.12xlarge`, `ml.g5.48xlarge`, `ml.g6.48xlarge`  | 
| Nova Micro | 訓練/微調 | 監督式微調 (完整) |  `nova_micro_1_0_p5_p4d_gpu_sft.yaml` `nova_micro_1_0_g5_g6_48x_gpu_sft.yaml`  | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-SFT-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-SFT-latest |  `ml.p5.48xlarge`, `ml.p5en.48xlarge`, `ml.p4d.24xlarge`, `ml.g5.48xlarge`, `ml.g6.48xlarge`  | 
| Nova Micro | 訓練/微調 | 直接偏好最佳化 (LoRA) |  `nova_micro_1_0_g5_g6_12x_gpu_lora_dpo.yaml` `nova_micro_1_0_g5_g6_48x_gpu_lora_dpo.yaml` `nova_micro_1_0_p5_p4d_gpu_lora_dpo.yaml`  | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-DPO-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-DPO-latest |  `ml.p5.48xlarge`, `ml.p5en.48xlarge`, `ml.p4d.24xlarge`, `ml.g5.12xlarge`, `ml.g6.12xlarge`, `ml.g5.48xlarge`, `ml.g6.48xlarge`  | 
| Nova Micro | 訓練/微調 | 直接偏好最佳化 (完整) | nova\$1micro\$11\$10\$1p5\$1p4d\$1gpu\$1dpo.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-DPO-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-DPO-latest | ml.p5.48xlarge, ml.p5en.48xlarge, ml.p4d.24xlarge | 
| Nova Micro | 訓練/強化學習 | 強化微調 (RFT) | nova\$1micro\$11\$10\$1p5\$1gpu\$1ppo.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SMHP-PPO-TRAIN-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Micro | 訓練/持續預先訓練 | 持續預先訓練 (基本模型) | nova\$1micro\$1gpu\$1p5x8\$1pretrain.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:HP-CPT-latest | ml.p5.48xlarge | 
| Nova Micro | 評估/評估 | 一般文字基準 | nova\$1micro\$1p5\$148xl\$1general\$1text\$1benchmark\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Micro | 評估/評估 | 使用自有資料集 (gen\$1qa) 基準 | nova\$1micro\$1p5\$148xl\$1bring\$1your\$1own\$1dataset\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Micro | 評估/評估 | LLM 即評審 | nova\$1micro\$1p5\$148xl\$1llm\$1judge\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 

## Amazon Nova Pro
<a name="nova-model-recipes-reference-novapro"></a>

下表列出 Amazon Nova Pro 配方參考的詳細資訊。


****  

| 模型 | 類別/子類別 | 技術 | 配方名稱 | 映像 URI （訓練任務） | 映像 URI (SageMaker HyperPod) | 運算執行個體 | 
| --- | --- | --- | --- | --- | --- | --- | 
| Nova Pro | 訓練/微調 | 監督式微調 (LoRA) | nova\$1pro\$11\$10\$1p5\$1p4d\$1gpu\$1lora\$1sft.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-SFT-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-SFT-latest | ml.p5.48xlarge, ml.p5en.48xlarge, ml.p4d.24xlarge | 
| Nova Pro | 訓練/微調 | 監督式微調 (完整) | nova\$1pro\$11\$10\$1p5\$1p4d\$1gpu\$1sft.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-SFT-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-SFT-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Pro | 訓練/微調 | 直接偏好最佳化 (完整) | nova\$1pro\$11\$10\$1p5\$1gpu\$1dpo.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-DPO-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-DPO-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Pro | 訓練/微調 | 直接偏好最佳化 (LoRA) | nova\$1pro\$11\$10\$1p5\$1p4d\$1gpu\$1lora\$1dpo.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-TJ-DPO-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SM-HP-DPO-latest | ml.p5.48xlarge, ml.p5en.48xlarge, ml.p4d.24xlarge | 
| Nova Pro | 訓練/強化學習 | 強化微調 (RFT) | nova\$1pro\$11\$10\$1p5\$1gpu\$1ppo.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:SMHP-PPO-TRAIN-latest | ml.p5.48xlarge, ml.p5en.48xlarge | 
| Nova Pro | 訓練/持續預先訓練 | 持續預先訓練 (基本模型) | nova\$1pro\$1gpu\$1p5x24\$1pretrain.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-fine-tune-repo:HP-CPT-latest | ml.p5.48xlarge | 
| Nova Pro | 訓練/資料增強 | 用於訓練後模型蒸餾 | nova\$1pro\$1r5\$1cpu\$1distill.yaml | N/A | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-distillation-repo:SM-TJ-DISTILL-LATEST | ml.r5.24xlarge | 
| Nova Pro | 評估/評估 | 標準文字基準 | nova\$1pro\$1p5\$148xl\$1general\$1text\$1benchmark\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Pro | 評估/評估 | 自訂資料集評估 | nova\$1pro\$1p5\$148xl\$1bring\$1your\$1own\$1dataset\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Pro | 評估/評估 | 多模態基準 | nova\$1pro\$1p5\$148xl\$1general\$1multi\$1modal\$1benchmark\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge | 
| Nova Pro | 評估/評估 | LLM 即評審 | nova\$1pro\$1p5\$148xl\$1llm\$1judge\$1eval.yaml | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-TJ-Eval-latest | 708977205387.dkr.ecr.us-east-1.amazonaws.com/nova-evaluation-repo:SM-HP-Eval-latest | ml.p5.48xlarge |