feat: scaffold a lambda that uses an openai function

Author: lpm0073

Makefile

@@ -65,6 +65,7 @@ api-init:
 	$(ACTIVATE_VENV) && \
 	$(PIP) install --upgrade pip && \
 	$(PIP) install -r requirements.txt && \
+	$(PYTHON) -m spacy download en_core_web_sm
 	deactivate && \
 	pre-commit install
 

api/terraform/lambda_layer.tf renamed to api/terraform/lambda_layer_genai.tf

@@ -0,0 +1,104 @@
+#------------------------------------------------------------------------------
+# written by: Lawrence McDaniel
+#             https://lawrencemcdaniel.com/
+#
+# date:   sep-2023
+#
+# usage:  implement a Python Lambda function to to format and pass
+#         text-based http requests directly to OpenAI API.
+#         - create a Lambda zip archive
+#         - pass openai api key credentials to Lambda in a safe manner
+#         - create a Cloudwatch log for the Lambda
+#         * note that IAM permissions are implemented on the resource(s)
+#           that call this Lambda, rather than here.
+#------------------------------------------------------------------------------
+locals {
+  openai_functionfunction_name     = "lambda_openai_function"
+  openai_functionbuild_path        = "${path.module}/build/distribution_package"
+  openai_functionsource_directory  = "${path.module}/python/openai_api"
+  openai_functionpackaging_script  = "${local.openai_functionsource_directory}/create_pkg.sh"
+  openai_functiondist_package_name = "${local.openai_functionfunction_name}_dist_pkg.zip"
+}
+
+###############################################################################
+# Python package
+# https://alek-cora-glez.medium.com/deploying-aws-lambda-function-with-terraform-custom-dependencies-7874407cd4fc
+###############################################################################
+resource "null_resource" "package_lambda_openai_function" {
+  triggers = {
+    always_redeploy = timestamp()
+  }
+
+  provisioner "local-exec" {
+    interpreter = ["/bin/bash"]
+    command     = local.openai_functionpackaging_script
+
+    environment = {
+      TERRAFORM_ROOT   = path.module
+      SOURCE_CODE_PATH = local.openai_functionsource_directory
+      BUILD_PATH       = local.openai_functionbuild_path
+      PACKAGE_FOLDER   = local.openai_functionfunction_name
+    }
+  }
+}
+
+data "archive_file" "lambda_openai_function" {
+  # see https://registry.terraform.io/providers/hashicorp/archive/latest/docs/data-sources/file
+  source_dir  = local.openai_functionbuild_path
+  output_path = "${path.module}/build/${local.openai_functiondist_package_name}"
+  type        = "zip"
+  depends_on  = [null_resource.package_lambda_openai_function]
+}
+
+###############################################################################
+# OpenAI API key and organization
+###############################################################################
+data "external" "env_lambda_openai_function" {
+  # kluge to read and map the openai api key and org data contained in .env
+  program = ["${local.openai_functionsource_directory}/${local.openai_functionfunction_name}/env.sh"]
+
+  # For Windows (or Powershell core on MacOS and Linux),
+  # run a Powershell script instead
+  #program = ["${path.module}/scripts/env.ps1"]
+}
+
+###############################################################################
+# AWS Lambda function
+###############################################################################
+resource "aws_lambda_function" "lambda_openai_function" {
+  # see https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/lambda_function.html
+  # see https://docs.aws.amazon.com/lambda/latest/dg/lambda-runtimes.html
+  function_name    = local.openai_functionfunction_name
+  description      = "LangChain request handler"
+  role             = aws_iam_role.lambda.arn
+  publish          = true
+  runtime          = var.lambda_python_runtime
+  memory_size      = var.lambda_memory_size
+  timeout          = var.lambda_timeout
+  handler          = "openai_api.lambda_openai_function.lambda_handler.handler"
+  architectures    = var.compatible_architectures
+  filename         = data.archive_file.lambda_openai_function.output_path
+  source_code_hash = data.archive_file.lambda_openai_function.output_base64sha256
+  layers           = [aws_lambda_layer_version.genai.arn]
+  tags             = var.tags
+
+  environment {
+    variables = {
+      DEBUG_MODE                 = var.debug_mode
+      OPENAI_API_ORGANIZATION    = data.external.env_lambda_openai_function.result["OPENAI_API_ORGANIZATION"]
+      OPENAI_API_KEY             = data.external.env_lambda_openai_function.result["OPENAI_API_KEY"]
+      OPENAI_ENDPOINT_IMAGE_N    = var.openai_endpoint_image_n
+      OPENAI_ENDPOINT_IMAGE_SIZE = var.openai_endpoint_image_size
+      AWS_DEPLOYED               = true
+    }
+  }
+}
+
+###############################################################################
+# Cloudwatch logging
+###############################################################################
+resource "aws_cloudwatch_log_group" "lambda_openai_function" {
+  name              = "/aws/lambda/${local.openai_functionfunction_name}"
+  retention_in_days = var.log_retention_days
+  tags              = var.tags
+}

api/terraform/lambda_openai_function.tf

@@ -9,5 +9,8 @@ WORKDIR /var/task
 
 COPY requirements.txt .
 
+ENV PYTHONPATH "${PYTHONPATH}:python/lib/python3.11/site-packages"
+
 RUN yum install -y zip
 RUN pip install -r requirements.txt --target python/lib/python3.11/site-packages
+RUN python3 -m spacy download en_core_web_sm

api/terraform/python/layer_genai/Dockerfile

@@ -9,13 +9,20 @@
 # -----------------------------------------------------------------------------
 
 # generative AI requirements
+# --------------------------
 langchain
 langchain-experimental
+langchain-openai==0.0.3
 openai==1.7.2
-pinecone-client==2.2.4
 
 # general requirements
+# --------------------------
 pydantic==2.5.3
 pydantic-settings==2.1.0
 python-dotenv==1.0.0
 python-hcl2==4.3.2
+
+# NLP requirements
+# --------------------------
+python-Levenshtein==0.23.0
+spacy==3.7.2

api/terraform/python/layer_genai/requirements.txt

@@ -6,9 +6,13 @@
 import datetime
 import json  # library for interacting with JSON data https://www.json.org/json-en.html
 import logging
+import re
+import string
 import sys  # libraries for error management
 import traceback  # libraries for error management
 
+import Levenshtein
+import spacy
 from openai_api.common.const import LANGCHAIN_MESSAGE_HISTORY_ROLES, OpenAIObjectTypes
 from openai_api.common.exceptions import OpenAIAPIValueError
 from openai_api.common.validators import (
@@ -24,6 +28,7 @@
 
 
 logger = logging.getLogger(__name__)
+nlp = spacy.load("en_core_web_sm")
 
 
 class DateTimeEncoder(json.JSONEncoder):
@@ -147,7 +152,7 @@ def get_request_body(event) -> dict:
 
 def parse_request(request_body: dict):
     """Parse the request body and return the endpoint, model, messages, and input_text"""
-    object_type = request_body.get("object")
+    object_type = request_body.get("object_type")
     model = request_body.get("model")
     messages = request_body.get("messages")
     input_text = request_body.get("input_text")
@@ -214,3 +219,61 @@ def get_messages_for_role(messages: list, role: str) -> list:
     """Get the text content from the messages list for a given role"""
     retval = [d.get("content") for d in messages if d["role"] == role]
     return retval
+
+
+def request_meta_data_factory(model, object_type, temperature, max_tokens, input_text):
+    """
+    Return a dictionary of request meta data.
+    """
+    return {
+        "request_meta_data": {
+            "lambda": "lambda_openai_v2",
+            "model": model,
+            "object_type": object_type,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+            "input_text": input_text,
+        }
+    }
+
+
+def does_refer_to(prompt: str, refers_to: str, threshold=3) -> bool:
+    """Check if the prompt refers to the given string."""
+
+    # clean up the prompt by adding spaces before capital letters
+    # converts "WhoIsLawrenceMcDaniel" to "Who Is Lawrence McDaniel"
+    clean_prompt = []
+    for word in prompt.split():
+        word = word.translate(str.maketrans("", "", string.punctuation))
+        doc = nlp(word)
+        words = re.sub("([A-Z][a-z]+)", r" \1", re.sub("([A-Z]+)", r" \1", word)).split()
+        words = (
+            ["".join(re.findall("[a-zA-Z]+", w)) for w in words]
+            if not any(ent.label_ in ["PERSON", "ORG"] for ent in doc.ents)
+            else [word.title()]
+        )
+        clean_prompt.extend(words)
+    prompt = " ".join(clean_prompt)
+
+    # first, try to find the target string in the prompt
+    prompt_words = prompt.lower().split()
+    token_count = len(refers_to.split())
+    found_count = 0
+    for token in refers_to.lower().split():
+        if token.lower() in prompt_words:
+            found_count += 1
+        if found_count >= token_count:
+            return True
+
+    # try to extract any names/titles from the prompt and then use
+    # the Levenshtein distance algorithm to see if any of them are
+    # close enough to the target name
+    words = prompt.split()
+    names = [word for word in words if word.istitle()]
+    for name in names:
+        distance = Levenshtein.distance(refers_to, name)
+        if distance <= threshold:
+            return True
+
+    # bust. we didn't find the target string in the prompt
+    return False

api/terraform/python/openai_api/common/utils.py

@@ -21,14 +21,14 @@
 import json
 
 from langchain.chains import LLMChain
-from langchain.chat_models import ChatOpenAI
 from langchain.memory import ConversationBufferMemory
 from langchain.prompts import (
     ChatPromptTemplate,
     HumanMessagePromptTemplate,
     MessagesPlaceholder,
     SystemMessagePromptTemplate,
 )
+from langchain_openai import ChatOpenAI
 from openai_api.common.conf import settings
 from openai_api.common.const import (  # VALID_EMBEDDING_MODELS,
     VALID_CHAT_COMPLETION_MODELS,
@@ -46,6 +46,7 @@
     get_request_body,
     http_response_factory,
     parse_request,
+    request_meta_data_factory,
 )
 from openai_api.common.validators import (  # validate_embedding_request,
     validate_completion_request,
@@ -77,17 +78,7 @@ def handler(event, context):
         request_body = get_request_body(event=event)
         validate_request_body(request_body=request_body)
         object_type, model, messages, input_text, temperature, max_tokens = parse_request(request_body)
-        request_meta_data = {
-            "request_meta_data": {
-                "lambda": "lambda_langchain",
-                "model": model,
-                "object_type": object_type,
-                "temperature": temperature,
-                "max_tokens": max_tokens,
-                "input_text": input_text,
-            }
-        }
-
+        request_meta_data = request_meta_data_factory(model, object_type, temperature, max_tokens, input_text)
         validate_messages(request_body=request_body)
 
         match object_type:

api/terraform/python/openai_api/lambda_langchain/lambda_handler.py

@@ -0,0 +1,27 @@
+#!/bin/sh
+#------------------------------------------------------------------------------
+# written by: Lawrence McDaniel
+#             https://lawrencemcdaniel.com/
+#
+# date:       sep-2023
+#
+# usage:      reformat environment variables for consumption by Terraform.
+#             Called by Terraform data "external" in lambda_langchain.tf
+#------------------------------------------------------------------------------
+
+# https://support.hashicorp.com/hc/en-us/articles/4547786359571-Reading-and-using-environment-variables-in-Terraform-runs
+
+
+set -o allexport
+source ../../.env set    # this is relative to the calling Terraform resource
++o allexport
+
+# Change the contents of this output to get the environment variables
+# of interest. The output must be valid JSON, with strings for both
+# keys and values.
+cat <<EOF
+{
+  "OPENAI_API_KEY": "$OPENAI_API_KEY",
+  "OPENAI_API_ORGANIZATION": "$OPENAI_API_ORGANIZATION"
+}
+EOF