Merge branch 'develop' v0.1.15

CHANGELOG.md

@@ -6,6 +6,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.1.15] - 2024-03-07
+### Added
+- Amazon Bedrock LLM plugin now suports anthropic.claude-3-sonnet model, and deprecates anthropic.claude-v1  - PR #23.
+
 ## [0.1.14] - 2023-12-22
 ### Added
 - Amazon Q Business Expert plugin now suports optional file attachments via Lex Web UI (v0.20.4) attach option and the new userFileUpload session attribute - PR #23.
@@ -81,7 +85,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Initial release
 
 [Unreleased]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/compare/v0.1.14...develop
-[0.1.14]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/releases/tag/v0.1.13
+[0.1.15]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/releases/tag/v0.1.15
+[0.1.14]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/releases/tag/v0.1.14
 [0.1.13]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/releases/tag/v0.1.13
 [0.1.12]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/releases/tag/v0.1.12
 [0.1.11]: https://github.com/aws-samples/qnabot-on-aws-plugin-samples/releases/tag/v0.1.11

lambdas/anthropic-llm/src/settings.py

@@ -13,6 +13,18 @@ def getModelSettings(model):
         "minTokens": 0,
         "topP": 1
     }
+    # claude-3 message API params are slightly different
+    provider = modelId.split(".")[0]
+    if provider == "anthropic":
+        if modelId.startswith("anthropic.claude-3"):
+            params = {
+                "model": model,
+                "temperature": 0,
+                "max_tokens": 256,
+                "top_p": 1,
+                "system": "You are a helpful AI assistant."
+            }            
+
     settings = {
         'LLM_GENERATE_QUERY_MODEL_PARAMS': json.dumps(params),
         'LLM_QA_MODEL_PARAMS': json.dumps(params),

lambdas/bedrock-embeddings-and-llm/src/lambdahook.py

@@ -90,7 +90,7 @@ def format_prompt(modelId, prompt):
     provider = modelId.split(".")[0]
     if provider == "anthropic":
         print("Model provider is Anthropic. Checking prompt format.")
-        if not prompt.startswith("\n\nHuman:"):
+        if not prompt.startswith("\n\nHuman:") or not prompt.startswith("\n\nSystem:"):
             prompt = "\n\nHuman: " + prompt
             print("Prepended '\\n\\nHuman:'")
         if not prompt.endswith("\n\nAssistant:"):

lambdas/bedrock-embeddings-and-llm/src/llm.py

@@ -20,11 +20,20 @@ def get_request_body(modelId, parameters, prompt):
     provider = modelId.split(".")[0]
     request_body = None
     if provider == "anthropic":
-        request_body = {
-            "prompt": prompt,
-            "max_tokens_to_sample": DEFAULT_MAX_TOKENS
-        } 
-        request_body.update(parameters)
+        # claude-3 models use new messages format
+        if modelId.startswith("anthropic.claude-3"):
+            request_body = {
+                "anthropic_version": "bedrock-2023-05-31",
+                "messages": [{"role": "user", "content": [{'type':'text','text': prompt}]}],
+                "max_tokens": DEFAULT_MAX_TOKENS
+            }
+            request_body.update(parameters)
+        else:
+            request_body = {
+                "prompt": prompt,
+                "max_tokens_to_sample": DEFAULT_MAX_TOKENS
+            } 
+            request_body.update(parameters)
     elif provider == "ai21":
         request_body = {
             "prompt": prompt,
@@ -59,20 +68,21 @@ def get_request_body(modelId, parameters, prompt):
 def get_generate_text(modelId, response):
     provider = modelId.split(".")[0]
     generated_text = None
+    response_body = json.loads(response.get("body").read())
+    print("Response body: ", json.dumps(response_body))
     if provider == "anthropic":
-        response_body = json.loads(response.get("body").read().decode())
-        generated_text = response_body.get("completion")
+        # claude-3 models use new messages format
+        if modelId.startswith("anthropic.claude-3"):
+            generated_text = response_body.get("content")[0].get("text")
+        else:
+            generated_text = response_body.get("completion")
     elif provider == "ai21":
-        response_body = json.loads(response.get("body").read())
         generated_text = response_body.get("completions")[0].get("data").get("text")
     elif provider == "amazon":
-        response_body = json.loads(response.get("body").read())
         generated_text = response_body.get("results")[0].get("outputText")
     elif provider == "cohere":
-        response_body = json.loads(response.get('body').read())
         generated_text = response_body.get("generations")[0].get("text")
     elif provider == "meta":
-        response_body = json.loads(response.get('body').read())
         generated_text = response_body.get("generation")
     else:
         raise Exception("Unsupported provider: ", provider)
@@ -95,8 +105,9 @@ def call_llm(parameters, prompt):
 {
   "prompt": "\n\nHuman:Why is the sky blue?\n\nAssistant:",
   "parameters": {
-    "modelId": "anthropic.claude-v1",
-    "temperature": 0
+    "modelId": "anthropic.claude-3-sonnet-20240229-v1:0",
+    "temperature": 0,
+    "system": "You are an AI assistant that always answers in ryhming couplets"
   }
 }
 For supported parameters for each provider model, see Bedrock docs: https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/providers

lambdas/bedrock-embeddings-and-llm/src/settings.py

@@ -33,10 +33,29 @@ def getModelSettings(modelId):
         "modelId": modelId,
         "temperature": 0
     }
+    params_qa = params.copy()
+    # claude-3 message API params are slightly different
+    provider = modelId.split(".")[0]
+    if provider == "anthropic":
+        if modelId.startswith("anthropic.claude-3"):
+            params = {
+                "modelId": modelId,
+                "temperature": 0,
+                "max_tokens": 256,
+                "top_p": 1
+            }
+            # add optional system prompt to qa params
+            params_qa = {
+                "modelId": modelId,
+                "temperature": 0,
+                "max_tokens": 256,
+                "top_p": 1,
+                "system": "You are a helpful AI assistant."
+            }
     lambdahook_args = {"Prefix":"LLM Answer:", "Model_params": params}
     settings = {
         'LLM_GENERATE_QUERY_MODEL_PARAMS': json.dumps(params),
-        'LLM_QA_MODEL_PARAMS': json.dumps(params),
+        'LLM_QA_MODEL_PARAMS': json.dumps(params_qa),
         'QNAITEM_LAMBDAHOOK_ARGS': json.dumps(lambdahook_args)
     }
     provider = modelId.split(".")[0]

lambdas/bedrock-embeddings-and-llm/template.yml

@@ -1,5 +1,5 @@
 AWSTemplateFormatVersion: "2010-09-09"
-Description: QnABot on AWS LLM Plugin for Bedrock - v0.1.13
+Description: QnABot on AWS LLM Plugin for Bedrock - v0.1.15
 
 Parameters:
 
@@ -18,10 +18,10 @@ Parameters:
       - amazon.titan-text-lite-v1
       - ai21.j2-ultra-v1
       - ai21.j2-mid-v1
-      - anthropic.claude-v1
       - anthropic.claude-instant-v1
       - anthropic.claude-v2
       - anthropic.claude-v2:1
+      - anthropic.claude-3-sonnet-20240229-v1:0
       - cohere.command-text-v14
       - cohere.command-light-text-v14
       - meta.llama2-13b-chat-v1
@@ -303,7 +303,7 @@ Resources:
       ServiceToken: !GetAtt OutputSettingsFunction.Arn
       EmbeddingsModelId: !Ref EmbeddingsModelId
       LLMModelId: !Ref LLMModelId
-      LastUpdate: '11/07/2023' 
+      LastUpdate: '03/07/2024 11:45' 
 
   TestBedrockModelFunction:
     Type: AWS::Lambda::Function