feat: support amazon nova (#623)

* support amazon nova * コメント対応 * CI対応 * CI対応 * コメント反映 * update readme --------- Co-authored-by: statefb <[email protected]>
aws-samples · Dec 4, 2024 · aabc8b6 · aabc8b6
1 parent 2063700
commit aabc8b6
Show file tree

Hide file tree

Showing 12 changed files with 166 additions and 42 deletions.
diff --git a/README.md b/README.md
@@ -1,8 +1,9 @@
-# Bedrock Claude Chat
+# Bedrock Claude Chat (Nova)
 
 ![](https://github.com/aws-samples/bedrock-claude-chat/actions/workflows/cdk.yml/badge.svg)
 
-> [!Warning] > **V2 released. To update, please carefully review the [migration guide](./docs/migration/V1_TO_V2.md).** Without any care, **BOTS FROM V1 WILL BECOME UNUSABLE.**
+> [!Warning]
+> **V2 released. To update, please carefully review the [migration guide](./docs/migration/V1_TO_V2.md).** Without any care, **BOTS FROM V1 WILL BECOME UNUSABLE.**
 
 This repository is a sample chatbot using the Anthropic company's LLM [Claude](https://www.anthropic.com/), one of the foundational models provided by [Amazon Bedrock](https://aws.amazon.com/bedrock/) for generative AI.
 
@@ -12,8 +13,6 @@ This repository is a sample chatbot using the Anthropic company's LLM [Claude](h
 
 ### Basic Conversation
 
-Not only text but also images are available with [Anthropic's Claude 3](https://www.anthropic.com/news/claude-3-family). Currently we support `Haiku`, `Sonnet` and `Opus`.
-
 ![](./docs/imgs/demo.gif)
 
 ### Bot Personalization
@@ -64,7 +63,7 @@ By using the [Agent functionality](./docs/AGENT.md), your chatbot can automatica
 
 ## 🚀 Super-easy Deployment
 
-- In the us-east-1 region, open [Bedrock Model access](https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/modelaccess) > `Manage model access` > Check `Anthropic / Claude 3 Haiku`, `Anthropic / Claude 3 Sonnet`, `Anthropic / Claude 3.5 Sonnet`, `Anthropic / Claude 3.5 Sonnet v2`, `Anthropic / Claude 3.5 Haiku`, `Amazon / Titan Text Embeddings V2` and `Cohere / Embed Multilingual` then `Save changes`.
+- In the us-east-1 region, open [Bedrock Model access](https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/modelaccess) > `Manage model access` > Check all of `Anthropic / Claude 3`, all of `Amazon / Nova`, `Amazon / Titan Text Embeddings V2` and `Cohere / Embed Multilingual` then `Save changes`.
 
 <details>
 <summary>Screenshot</summary>

diff --git a/backend/app/bedrock.py b/backend/app/bedrock.py
@@ -1,6 +1,6 @@
 import logging
 import os
-from typing import TypeGuard
+from typing import TypeGuard, Dict, Any, Optional, Tuple
 
 from app.agents.tools.agent_tool import AgentTool
 from app.config import BEDROCK_PRICING
@@ -46,6 +46,48 @@ def _is_conversation_role(role: str) -> TypeGuard[ConversationRoleType]:
     return role in ["user", "assistant"]
 
 
+def _is_nova_model(model: type_model_name) -> bool:
+    """Check if the model is an Amazon Nova model"""
+    return model in ["amazon-nova-pro", "amazon-nova-lite", "amazon-nova-micro"]
+
+
+def _prepare_nova_model_params(
+    model: type_model_name, generation_params: Optional[GenerationParamsModel] = None
+) -> Tuple[InferenceConfigurationTypeDef, Dict[str, Any]]:
+    """
+    Prepare inference configuration and additional model request fields for Nova models
+    > Note that Amazon Nova expects inference parameters as a JSON object under a inferenceConfig attribute. Amazon Nova also has an additional parameter "topK" that can be passed as an additional inference parameters. This parameter follows the same structure and is passed through the additionalModelRequestFields, as shown below.
+    https://docs.aws.amazon.com/nova/latest/userguide/getting-started-converse.html
+    """
+    # Base inference configuration
+    inference_config: InferenceConfigurationTypeDef = {
+        "maxTokens": (
+            generation_params.max_tokens
+            if generation_params
+            else DEFAULT_GENERATION_CONFIG["max_tokens"]
+        ),
+        "temperature": (
+            generation_params.temperature
+            if generation_params
+            else DEFAULT_GENERATION_CONFIG["temperature"]
+        ),
+        "topP": (
+            generation_params.top_p
+            if generation_params
+            else DEFAULT_GENERATION_CONFIG["top_p"]
+        ),
+    }
+
+    # Additional model request fields specific to Nova models
+    additional_fields: Dict[str, Any] = {"inferenceConfig": {}}
+
+    # Add top_k if specified in generation params
+    if generation_params and generation_params.top_k is not None:
+        additional_fields["inferenceConfig"]["topK"] = generation_params.top_k
+
+    return inference_config, additional_fields
+
+
 def compose_args_for_converse_api(
     messages: list[SimpleMessageModel],
     model: type_model_name,
@@ -88,39 +130,55 @@ def process_content(c: ContentModel, role: str) -> list[ContentBlockTypeDef]:
         if _is_conversation_role(message.role)
     ]
 
-    inference_config: InferenceConfigurationTypeDef
-    if generation_params:
-        inference_config = {
-            "maxTokens": generation_params.max_tokens,
-            "temperature": generation_params.temperature,
-            "topP": generation_params.top_p,
-            "stopSequences": generation_params.stop_sequences,
-        }
-        additional_model_request_fields = {
-            "top_k": generation_params.top_k,
-        }
-
+    # Prepare model-specific parameters
+    if _is_nova_model(model):
+        # Special handling for Nova models
+        inference_config, additional_model_request_fields = _prepare_nova_model_params(
+            model, generation_params
+        )
     else:
+        # Standard handling for non-Nova models
         inference_config = {
-            "maxTokens": DEFAULT_GENERATION_CONFIG["max_tokens"],
-            "temperature": DEFAULT_GENERATION_CONFIG["temperature"],
-            "topP": DEFAULT_GENERATION_CONFIG["top_p"],
-            "stopSequences": DEFAULT_GENERATION_CONFIG["stop_sequences"],
+            "maxTokens": (
+                generation_params.max_tokens
+                if generation_params
+                else DEFAULT_GENERATION_CONFIG["max_tokens"]
+            ),
+            "temperature": (
+                generation_params.temperature
+                if generation_params
+                else DEFAULT_GENERATION_CONFIG["temperature"]
+            ),
+            "topP": (
+                generation_params.top_p
+                if generation_params
+                else DEFAULT_GENERATION_CONFIG["top_p"]
+            ),
+            "stopSequences": (
+                generation_params.stop_sequences
+                if generation_params
+                else DEFAULT_GENERATION_CONFIG.get("stop_sequences", [])
+            ),
         }
         additional_model_request_fields = {
-            "top_k": DEFAULT_GENERATION_CONFIG["top_k"],
+            "top_k": (
+                generation_params.top_k
+                if generation_params
+                else DEFAULT_GENERATION_CONFIG["top_k"]
+            )
         }
 
+    # Construct the base arguments
     args: ConverseStreamRequestRequestTypeDef = {
         "inferenceConfig": inference_config,
-        "additionalModelRequestFields": additional_model_request_fields,
         "modelId": get_model_id(model),
         "messages": arg_messages,
         "system": [
             {"text": instruction}
             for instruction in instructions
             if len(instruction) > 0
         ],
+        "additionalModelRequestFields": additional_model_request_fields,
     }
 
     if guardrail and guardrail.guardrail_arn and guardrail.guardrail_version:
@@ -193,6 +251,10 @@ def get_model_id(
         "mistral-7b-instruct": "mistral.mistral-7b-instruct-v0:2",
         "mixtral-8x7b-instruct": "mistral.mixtral-8x7b-instruct-v0:1",
         "mistral-large": "mistral.mistral-large-2402-v1:0",
+        # New Amazon Nova models
+        "amazon-nova-pro": "amazon.nova-pro-v1:0",
+        "amazon-nova-lite": "amazon.nova-lite-v1:0",
+        "amazon-nova-micro": "amazon.nova-micro-v1:0",
     }
 
     # Ref: https://docs.aws.amazon.com/bedrock/latest/userguide/cross-region-inference-support.html
@@ -203,6 +265,9 @@ def get_model_id(
         "claude-v3.5-sonnet",
         "claude-v3.5-sonnet-v2",
         "claude-v3.5-haiku",
+        "amazon-nova-pro",
+        "amazon-nova-lite",
+        "amazon-nova-micro",
     }
 
     supported_region_prefixes = {

diff --git a/backend/app/config.py b/backend/app/config.py
@@ -58,6 +58,9 @@ class EmbeddingConfig(TypedDict):
         "mistral-7b-instruct": {"input": 0.00015, "output": 0.0002},
         "mixtral-8x7b-instruct": {"input": 0.00045, "output": 0.0007},
         "mistral-large": {"input": 0.008, "output": 0.024},
+        "amazon-nova-pro": {"input": 0.0008, "output": 0.0032},
+        "amazon-nova-lite": {"input": 0.00006, "output": 0.00024},
+        "amazon-nova-micro": {"input": 0.000035, "output": 0.00014},
     },
     "us-west-2": {
         "claude-instant-v1": {
@@ -73,6 +76,9 @@ class EmbeddingConfig(TypedDict):
         "mistral-7b-instruct": {"input": 0.00015, "output": 0.0002},
         "mixtral-8x7b-instruct": {"input": 0.00045, "output": 0.0007},
         "mistral-large": {"input": 0.008, "output": 0.024},
+        "amazon-nova-pro": {"input": 0.0008, "output": 0.0032},
+        "amazon-nova-lite": {"input": 0.00006, "output": 0.00024},
+        "amazon-nova-micro": {"input": 0.000035, "output": 0.00014},
     },
     "ap-northeast-1": {
         "claude-instant-v1": {
@@ -102,5 +108,8 @@ class EmbeddingConfig(TypedDict):
         "mistral-7b-instruct": {"input": 0.00015, "output": 0.0002},
         "mixtral-8x7b-instruct": {"input": 0.00045, "output": 0.0007},
         "mistral-large": {"input": 0.008, "output": 0.024},
+        "amazon-nova-pro": {"input": 0.0008, "output": 0.0032},
+        "amazon-nova-lite": {"input": 0.00006, "output": 0.00024},
+        "amazon-nova-micro": {"input": 0.000035, "output": 0.00014},
     },
 }
diff --git a/backend/app/routes/schemas/conversation.py b/backend/app/routes/schemas/conversation.py
@@ -21,6 +21,10 @@
     "mistral-7b-instruct",
     "mixtral-8x7b-instruct",
     "mistral-large",
+    # New Amazon Nova models
+    "amazon-nova-pro",
+    "amazon-nova-lite",
+    "amazon-nova-micro",
 ]
 
 

diff --git a/docs/README_ja.md b/docs/README_ja.md
@@ -1,12 +1,12 @@
-# Bedrock Claude Chat
+# Bedrock Claude Chat (Nova)
 
-> [!Warning] > **V2 がリリースされました。** 更新する際は、[移行ガイド](./migration/V1_TO_V2.md)を必ずご確認ください。 **注意を払わずに進めると、V1 の BOT は使用できなくなります。**
+> [!Warning]
+> **V2 がリリースされました。** 更新する際は、[移行ガイド](./migration/V1_TO_V2.md)を必ずご確認ください。 **注意を払わずに進めると、V1 の BOT は使用できなくなります。**
 
 このリポジトリは、生成系 AI を提供する[Amazon Bedrock](https://aws.amazon.com/jp/bedrock/)の基盤モデルの一つである、Anthropic 社製 LLM [Claude](https://www.anthropic.com/)を利用したチャットボットのサンプルです。
 
 ### 基本的な会話
 
-[Claude 3](https://www.anthropic.com/news/claude-3-family)によるテキストと画像の両方を利用したチャットが可能です。現在`Haiku`および`Sonnet`、または`Opus`をサポートしています。
 ![](./imgs/demo_ja.gif)
 
 ### ボットのカスタマイズ
@@ -33,7 +33,7 @@
 
 ## 🚀 まずはお試し
 
-- us-east-1 リージョンにて、[Bedrock Model access](https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/modelaccess) > `Manage model access` > `Anthropic / Claude 3 Haiku`, `Anthropic / Claude 3 Sonnet`, `Anthropic / Claude 3.5 Sonnet` `Cohere / Embed Multilingual`をチェックし、`Save changes`をクリックします
+- us-east-1 リージョンにて、[Bedrock Model access](https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/modelaccess) > `Manage model access` > すべての`Anthropic / Claude 3`、 すべての `Amazon / Nova`、`Cohere / Embed Multilingual`をチェックし、`Save changes`をクリックします
 
 <details>
 <summary>スクリーンショット</summary>

diff --git a/frontend/src/@types/conversation.d.ts b/frontend/src/@types/conversation.d.ts
@@ -10,7 +10,10 @@ export type Model =
   | 'claude-v3.5-haiku'
   | 'mistral-7b-instruct'
   | 'mixtral-8x7b-instruct'
-  | 'mistral-large';
+  | 'mistral-large'
+  | 'amazon-nova-pro'
+  | 'amazon-nova-lite'
+  | 'amazon-nova-micro';
 
 export type Content = TextContent | ImageContent | AttachmentContent;
 

diff --git a/frontend/src/components/InputChatContent.tsx b/frontend/src/components/InputChatContent.tsx
@@ -459,10 +459,11 @@ const InputChatContent = forwardRef<HTMLElement, Props>((props, focusInputRef) =
         )}>
         <div className="flex w-full">
           <Textarea
-            className="m-1  bg-transparent pr-12 scrollbar-thin scrollbar-thumb-light-gray"
-            placeholder={props.placeholder ?? t('app.inputMessage')}
+            className="m-1 bg-transparent pr-12 scrollbar-thin scrollbar-thumb-light-gray"
+            placeholder={t('app.inputMessage')}
             disabled={props.disabled}
             noBorder
+            rows={3}
             value={content}
             onChange={setContent}
             ref={focusInputRef}

diff --git a/frontend/src/constants/index.ts b/frontend/src/constants/index.ts
@@ -48,7 +48,7 @@ export const EDGE_MISTRAL_GENERATION_PARAMS = {
 
 export const DEFAULT_GENERATION_CONFIG: GenerationParams = {
   maxTokens: 2000,
-  topK: 250,
+  topK: 128,
   topP: 0.999,
   temperature: 0.6,
   stopSequences: ['Human: ', 'Assistant: '],

diff --git a/frontend/src/hooks/useModel.ts b/frontend/src/hooks/useModel.ts
@@ -14,6 +14,13 @@ const CLAUDE_SUPPORTED_MEDIA_TYPES = [
   'image/webp',
 ];
 
+const NOVA_SUPPORTED_MEDIA_TYPES = [
+  'image/jpeg',
+  'image/png',
+  'image/gif',
+  'image/webp',
+]
+
 const useModelState = create<{
   modelId: Model;
   setModelId: (m: Model) => void;
@@ -75,6 +82,25 @@ const useModel = () => {
             description: t('model.opus3.description'),
             supportMediaType: CLAUDE_SUPPORTED_MEDIA_TYPES,
           },
+          // New Amazon Nova models
+          {
+            modelId: 'amazon-nova-pro',
+            label: t('model.novaPro.label'),
+            description: t('model.novaPro.description'),
+            supportMediaType: NOVA_SUPPORTED_MEDIA_TYPES,
+          },
+          {
+            modelId: 'amazon-nova-lite',
+            label: t('model.novaLite.label'),
+            description: t('model.novaLite.description'),
+            supportMediaType: NOVA_SUPPORTED_MEDIA_TYPES,
+          },
+          {
+            modelId: 'amazon-nova-micro',
+            label: t('model.novaMicro.label'),
+            description: t('model.novaMicro.description'),
+            supportMediaType: [],
+          }
         ]
       : [
           {

diff --git a/frontend/src/i18n/en/index.ts b/frontend/src/i18n/en/index.ts
@@ -8,7 +8,7 @@ const translation = {
     app: {
       name: 'Bedrock Claude Chat',
       nameWithoutClaude: 'Bedrock Chat',
-      inputMessage: 'Send a message',
+      inputMessage: 'Can I Help You?',
       starredBots: 'Starred Bots',
       recentlyUsedBots: 'Recently Used Bots',
       conversationHistory: 'History',
@@ -53,6 +53,18 @@ const translation = {
       mistralLarge: {
         label: 'Mistral Large',
       },
+      novaPro: {
+        label: "Amazon Nova Pro",
+        description: 'A highly capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks.',
+      },
+      novaLite: {
+        label: 'Amazon Nova Lite',
+        description: 'A very low-cost multimodal model that is lightning fast for processing image, video, and text inputs.',
+      },
+      novaMicro: {
+        label: 'Amazon Nova Micro',
+        description: 'A text-only model that delivers the lowest latency responses in the Amazon Nova family of models at a very low cost.',
+      }
     },
     agent: {
       label: 'Agent',

diff --git a/frontend/src/i18n/ja/index.ts b/frontend/src/i18n/ja/index.ts
@@ -11,7 +11,7 @@ const translation = {
     app: {
       name: 'Bedrock Claude Chat',
       nameWithoutClaude: 'Bedrock Chat',
-      inputMessage: '入力してください',
+      inputMessage: 'お手伝いできることはありますか？',
       starredBots: 'スター付きのボット',
       recentlyUsedBots: '最近使用したボット',
       conversationHistory: '会話履歴',
@@ -55,6 +55,18 @@ const translation = {
       mistralLarge: {
         label: 'Mistral Large',
       },
+      novaPro: {
+        label: "Amazon Nova Pro",
+        description: '精度、速度、コストのバランスが最も優れた高性能マルチモーダルモデル',
+      },
+      novaLite: {
+        label: 'Amazon Nova Lite',
+        description: '非常に低コストで高速なマルチモーダルモデルで、リアルタイム処理に最適',
+      },
+      novaMicro: {
+        label: 'Amazon Nova Micro',
+        description: '最も低いレイテンシーと低コストで提供される軽量なテキストモデル',
+      }
     },
     agent: {
       label: 'エージェント',