Merge branch 'aws-samples:main' into main

cli/magic-config.ts

@@ -630,7 +630,10 @@ async function processCreateOptions(options: any): Promise<void> {
           options.kendraExternal.length > 0) ||
         false,
       skip(): boolean {
-        return !(this as any).state.answers.enableRag;
+        if (!(this as any).state.answers.enableRag){
+          return true;
+        }
+        return !(this as any).state.answers.ragsToEnable.includes("kendra");
       },
     },
   ];

lib/model-interfaces/langchain/functions/request-handler/adapters/bedrock/__init__.py

@@ -4,3 +4,4 @@
 from .cohere import *
 from .llama2_chat import *
 from .mistral import *
+from .llama3 import *

lib/model-interfaces/langchain/functions/request-handler/adapters/bedrock/llama3.py

@@ -0,0 +1,67 @@
+import genai_core.clients
+
+# from langchain.llms import Bedrock (pending https://github.com/langchain-ai/langchain/issues/13316)
+from .base import Bedrock
+
+from langchain.prompts.prompt import PromptTemplate
+
+
+from ..shared.meta.llama3_instruct import (
+    Llama3PromptTemplate,
+    Llama3QAPromptTemplate,
+    Llama3CondensedQAPromptTemplate,
+)
+from ..shared.meta.llama3_instruct import Llama3ConversationBufferMemory
+
+from ..base import ModelAdapter
+from genai_core.registry import registry
+
+
+class BedrockMetaLLama3InstructAdapter(ModelAdapter):
+    def __init__(self, model_id, *args, **kwargs):
+        self.model_id = model_id
+
+        super().__init__(*args, **kwargs)
+
+    def get_memory(self, output_key=None, return_messages=False):
+        return Llama3ConversationBufferMemory(
+            memory_key="chat_history",
+            chat_memory=self.chat_history,
+            return_messages=return_messages,
+            output_key=output_key,
+        )
+
+    def get_llm(self, model_kwargs={}):
+        bedrock = genai_core.clients.get_bedrock_client()
+
+        params = {}
+        if "temperature" in model_kwargs:
+            params["temperature"] = model_kwargs["temperature"]
+        if "topP" in model_kwargs:
+            params["top_p"] = model_kwargs["topP"]
+        if "maxTokens" in model_kwargs:
+            params["max_gen_len"] = model_kwargs["maxTokens"]
+
+        return Bedrock(
+            client=bedrock,
+            model_id=self.model_id,
+            model_kwargs=params,
+            streaming=model_kwargs.get("streaming", False),
+            callbacks=[self.callback_handler],
+        )
+
+    def get_prompt(self):
+        return Llama3PromptTemplate
+
+    def get_qa_prompt(self):
+        return Llama3QAPromptTemplate
+
+    def get_condense_question_prompt(self):
+        return Llama3CondensedQAPromptTemplate
+
+
+# Register the adapter
+registry.register(
+    r"^bedrock.meta.llama3-.*-instruct.*",
+    BedrockMetaLLama3InstructAdapter,
+)

lib/model-interfaces/langchain/functions/request-handler/adapters/shared/meta/__init__.py

@@ -1 +1,2 @@
 from .llama2_chat import *
+from .llama3_instruct import *

lib/model-interfaces/langchain/functions/request-handler/adapters/shared/meta/llama3_instruct.py

@@ -0,0 +1,61 @@
+import json
+
+from langchain.schema import AIMessage, HumanMessage, SystemMessage
+from langchain.memory import ConversationBufferMemory
+from langchain.prompts import PromptTemplate
+
+BEGIN_OF_TEXT = "<|begin_of_text|>"
+SYSTEM_HEADER = "<|start_header_id|>system<|end_header_id|>"
+USER_HEADER = "<|start_header_id|>user<|end_header_id|>"
+ASSISTANT_HEADER = "<|start_header_id|>assistant<|end_header_id|>"
+EOD = "<|eot_id|>"
+
+Llama3Prompt = f"""{BEGIN_OF_TEXT}{SYSTEM_HEADER}
+
+You are an helpful assistant that provides concise answers to user questions with as little sentences as possible and at maximum 3 sentences. You do not repeat yourself. You avoid bulleted list or emojis.{EOD}{{chat_history}}{USER_HEADER}
+
+{{input}}{EOD}{ASSISTANT_HEADER}"""
+
+Llama3QAPrompt = f"""{BEGIN_OF_TEXT}{SYSTEM_HEADER}
+
+Use the following conversation history and pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer. You do not repeat yourself. You avoid bulleted list or emojis.{EOD}{{chat_history}}{USER_HEADER}
+
+Context: {{context}}
+
+{{question}}{EOD}{ASSISTANT_HEADER}"""
+
+Llama3CondensedQAPrompt = f"""{BEGIN_OF_TEXT}{SYSTEM_HEADER}
+
+Given the following conversation and the question at the end, rephrase the follow up input to be a standalone question, in the same language as the follow up input. You do not repeat yourself. You avoid bulleted list or emojis.{EOD}{{chat_history}}{USER_HEADER}
+
+{{question}}{EOD}{ASSISTANT_HEADER}"""
+
+
+Llama3PromptTemplate = PromptTemplate.from_template(Llama3Prompt)
+Llama3QAPromptTemplate = PromptTemplate.from_template(Llama3QAPrompt)
+Llama3CondensedQAPromptTemplate = PromptTemplate.from_template(Llama3CondensedQAPrompt)
+
+
+class Llama3ConversationBufferMemory(ConversationBufferMemory):
+    @property
+    def buffer_as_str(self) -> str:
+        return self.get_buffer_string()
+
+    def get_buffer_string(self) -> str:
+        # See https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3/
+        string_messages = []
+        for m in self.chat_memory.messages:
+            if isinstance(m, HumanMessage):
+                message = f"""{USER_HEADER}
+
+{m.content}{EOD}"""
+
+            elif isinstance(m, AIMessage):
+                message = f"""{ASSISTANT_HEADER}
+
+{m.content}{EOD}"""
+            else:
+                raise ValueError(f"Got unsupported message type: {m}")
+            string_messages.append(message)
+
+        return "".join(string_messages)

lib/shared/layers/python-sdk/python/genai_core/documents.py

@@ -485,6 +485,7 @@ def _process_document(
 
             try:
                 urls_to_crawl = genai_core.websites.extract_urls_from_sitemap(path)
+                limit = min(limit, len(urls_to_crawl))
 
                 if len(urls_to_crawl) == 0:
                     set_status(workspace_id, document_id, "error")
@@ -512,7 +513,7 @@ def _process_document(
                     "priority_queue": priority_queue,
                     "processed_urls": [],
                     "follow_links": follow_links,
-                    "limit": min(limit, len(urls_to_crawl)),
+                    "limit": limit,
                     "done": False,
                 },
                 cls=genai_core.utils.json.CustomEncoder,