added AWS Bedrock provider for Llama 3 - with model input converted t…

…o using Llama 3 prompting format and output converted to use simulated OpenAI chat completion output type

.env.sample

@@ -6,4 +6,6 @@ OPENAI_API_KEY=""
 OLLAMA_API_URL="http://localhost:11434"
 REPLICATE_API_KEY=""
 TOGETHER_API_KEY=""
-OCTO_API_KEY=""
+OCTO_API_KEY=""
+AWS_ACCESS_KEY_ID=""
+AWS_SECRET_ACCESS_KEY=""

aimodels/client/multi_fm_client.py

@@ -11,6 +11,7 @@
     ReplicateInterface,
     TogetherInterface,
     OctoInterface,
+    AWSBedrockInterface,
 )
 
 
@@ -46,6 +47,7 @@ def __init__(self):
             "replicate": ReplicateInterface,
             "together": TogetherInterface,
             "octo": OctoInterface,
+            "aws": AWSBedrockInterface
         }
 
     def get_provider_interface(self, model):

aimodels/providers/__init__.py

@@ -9,3 +9,4 @@
 from .replicate_interface import ReplicateInterface
 from .together_interface import TogetherInterface
 from .octo_interface import OctoInterface
+from .aws_bedrock_interface import AWSBedrockInterface

aimodels/providers/aws_bedrock_interface.py

@@ -0,0 +1,101 @@
+"""The interface to the Together API."""
+
+import os
+from urllib.request import urlopen
+import boto3
+import json
+
+from ..framework.provider_interface import ProviderInterface
+
+def convert_messages_to_llama3_prompt(messages):
+    """
+    Convert a list of messages to a prompt in Llama 3 instruction format.
+    
+    Args:
+    messages (list of dict): List of messages where each message is a dictionary 
+                            with 'role' ('system', 'user', 'assistant') and 'content'.
+    
+    Returns:
+    str: Formatted prompt for Llama 3.
+    """
+    prompt = "<|begin_of_text|>"
+    for message in messages:
+        prompt += f"<|start_header_id|>{message['role']}<|end_header_id|>{message['content']}<|eot_id|>\n"
+
+    prompt += "<|start_header_id|>assistant<|end_header_id|>"
+
+    return prompt    
+
+class RecursiveNamespace:
+    """
+    Convert dictionaries to objects with attribute access, including nested dictionaries.
+    This class is used to simulate the OpenAI chat.completions.create's return type, so
+    response.choices[0].message.content works consistenly for AWS Bedrock's LLM return of a string.
+    """
+    def __init__(self, data):
+        for key, value in data.items():
+            if isinstance(value, dict):
+                value = RecursiveNamespace(value)
+            elif isinstance(value, list):
+                value = [RecursiveNamespace(item) if isinstance(item, dict) else item for item in value]
+            setattr(self, key, value)
+
+    @classmethod
+    def from_dict(cls, data):
+        return cls(data)
+
+    def to_dict(self):
+        result = {}
+        for key, value in self.__dict__.items():
+            if isinstance(value, RecursiveNamespace):
+                value = value.to_dict()
+            elif isinstance(value, list):
+                value = [item.to_dict() if isinstance(item, RecursiveNamespace) else item for item in value]
+            result[key] = value
+        return result
+
+class AWSBedrockInterface(ProviderInterface):
+    """Implements the ProviderInterface for interacting with AWS Bedrock's APIs."""
+
+    def __init__(self):
+        """Set up the AWS Bedrock client using the AWS access key id and secret access key obtained from the user's environment."""
+        self.aws_bedrock_client = boto3.client(
+            service_name="bedrock-runtime",
+            region_name="us-west-2",
+            aws_access_key_id=os.getenv("AWS_ACCESS_KEY_ID"),
+            aws_secret_access_key=os.getenv("AWS_SECRET_ACCESS_KEY"),
+    )
+
+    def chat_completion_create(self, messages=None, model=None, temperature=0):
+        """Request chat completions from the AWS Bedrock API.
+
+        Args:
+        ----
+            model (str): Identifies the specific provider/model to use.
+            messages (list of dict): A list of message objects in chat history.
+            temperature (float): The temperature to use in the completion.
+
+        Returns:
+        -------
+            The API response with the completion result.
+
+        """
+        body = json.dumps({
+            "prompt": convert_messages_to_llama3_prompt(messages),
+            "temperature": temperature
+        })
+        accept = 'application/json'
+        content_type = 'application/json'
+        response = self.aws_bedrock_client.invoke_model(body=body, modelId=model, accept=accept, contentType=content_type)
+        response_body = json.loads(response.get('body').read())
+        generation = response_body.get('generation')
+
+        response_data = {
+            "choices": [
+                {
+                    "message": {"content": generation},
+                }
+            ],
+        }        
+
+        return RecursiveNamespace.from_dict(response_data)

examples/multi_fm_client.ipynb

@@ -45,7 +45,9 @@
     "os.environ['FIREWORKS_API_KEY'] = 'xxx' # get a free key at https://fireworks.ai/api-keys\n",
     "os.environ['REPLICATE_API_KEY'] = 'xxx' # get a free key at https://replicate.com/account/api-tokens\n",
     "os.environ['TOGETHER_API_KEY'] = 'xxx' # get a free key at https://api.together.ai\n",
-    "os.environ['OCTO_API_KEY'] = 'xxx' # get a free key at https://octoai.cloud/settings"
+    "os.environ['OCTO_API_KEY'] = 'xxx' # get a free key at https://octoai.cloud/settings\n",
+    "os.environ['AWS_ACCESS_KEY_ID'] = 'xxx' # get or create at https://console.aws.amazon.com/iam/home\n",
+    "os.environ['AWS_SECRET_ACCESS_KEY'] = 'xxx' # get or create at https://console.aws.amazon.com/iam/home"
    ]
   },
   {
@@ -70,6 +72,31 @@
     "]"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1ffe9a49-638e-4304-b9de-49ee21d9ac8d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#!pip install boto3"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "9893c7e4-799a-42c9-84de-f9e643044462",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "aws_bedrock_llama3_8b = \"aws:meta.llama3-8b-instruct-v1:0\"\n",
+    "#aws_bedrock_llama3_8b = \"aws:meta.llama3-70b-instruct-v1:0\"\n",
+    "\n",
+    "response = client.chat.completions.create(model=aws_bedrock_llama3_8b, messages=messages)\n",
+    "\n",
+    "print(response.choices[0].message.content)"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,