Caching strategies for cost reduction in LangChain - Mini Project: Build & Apply

Caching Strategies for Cost Reduction with LangChain

📖 Scenario: You are building a chatbot using LangChain that calls an expensive language model API. To save money, you want to store answers for repeated questions so the API is not called every time.

🎯 Goal: Build a simple LangChain setup that caches responses for repeated inputs to reduce API calls and costs.

📋 What You'll Learn

Create a dictionary called cache to store question-answer pairs

Create a variable called max_cache_size set to 3

Write a function get_answer(question) that checks cache before calling the API

Add logic to remove the oldest cached item when cache exceeds max_cache_size

💡 Why This Matters

🌍 Real World

Caching is used in chatbots and apps to save money and speed up responses by reusing previous answers.

💼 Career

Understanding caching helps developers optimize costs and improve performance in real-world software projects.

1

Create the cache dictionary

Create an empty dictionary called cache to store question-answer pairs.

LangChain

# Create an empty dictionary called cache
# Your code here

Hint

Use {} to create an empty dictionary in Python.

2

Set the maximum cache size

Create a variable called max_cache_size and set it to 3.

LangChain

cache = {}
# Set max_cache_size to 3
# Your code here

Hint

Just assign the number 3 to the variable max_cache_size.

3

Write the caching function

Write a function called get_answer(question) that does the following:
1. Checks if question is in cache and returns the cached answer if found.
2. Otherwise, calls a placeholder function call_api(question) to get the answer.
3. Adds the new question and answer to cache.

LangChain

cache = {}
max_cache_size = 3

# Define get_answer function
# Your code here

Hint

Use if question in cache to check cache, else call call_api(question) and store result.

4

Limit cache size by removing oldest entry

Update the get_answer(question) function to remove the oldest cached item when the size of cache exceeds max_cache_size after adding a new answer.

LangChain

cache = {}
max_cache_size = 3

def call_api(question):
    # Placeholder for expensive API call
    return f"Answer to '{question}'"

def get_answer(question):
    if question in cache:
        return cache[question]
    answer = call_api(question)
    cache[question] = answer
    # Remove oldest cache item if cache is too big
    # Your code here
    return answer

Hint

Use len(cache) to check size and next(iter(cache)) to get oldest key.

Practice

1. What is the main benefit of using caching in Langchain to reduce costs?

easy

A. It automatically upgrades the Langchain version

B. It stores previous results to avoid repeated expensive calls

C. It deletes all data after each request to save memory

D. It increases the number of API calls to improve speed

Solution

1. Step 1: Understand caching purpose

   Caching saves results from previous operations to reuse them later.

2. Step 2: Connect caching to cost reduction

   By reusing stored results, it avoids repeated expensive API calls, lowering costs.

3. Final Answer:

   It stores previous results to avoid repeated expensive calls -> Option B

4. Quick Check:

   Caching = Avoid repeated calls [OK]

Hint: Caching saves results to skip repeated work [OK]

Common Mistakes:

• Thinking caching increases API calls
• Believing caching deletes data immediately
• Confusing caching with version upgrades

2. Which of the following is the correct way to use caching with Langchain's get_or_set method?

easy

A. cache.get_or_set(key, lambda: expensive_call())

B. cache.set_or_get(key, expensive_call())

C. cache.get_or_set(expensive_call(), key)

D. cache.get(key, expensive_call())

Solution

1. Step 1: Recall get_or_set syntax

   The method takes a key and a function to call if the key is missing.

2. Step 2: Match correct argument order

   Correct usage is cache.get_or_set(key, lambda: expensive_call()) to delay call until needed.

3. Final Answer:

   cache.get_or_set(key, lambda: expensive_call()) -> Option A

4. Quick Check:

   Correct method and argument order = B [OK]

Hint: Remember: key first, then function in get_or_set [OK]

Common Mistakes:

• Swapping key and function arguments
• Calling expensive function immediately instead of lazy
• Using wrong method names like set_or_get

3. Given this code snippet using Langchain caching:

cache = InMemoryCache()
result1 = cache.get_or_set('key1', lambda: 'data1')
result2 = cache.get_or_set('key1', lambda: 'data2')

What will be the value of result2?

medium

A. 'data2'

B. None

C. 'data1'

D. Raises an error

Solution

1. Step 1: Understand get_or_set behavior

   If the key exists, it returns cached value without calling the function.

2. Step 2: Apply to given code

   First call caches 'data1' under 'key1'. Second call finds 'key1' and returns 'data1', ignoring 'data2'.

3. Final Answer:

   'data1' -> Option C

4. Quick Check:

   Cache returns stored value, not new function result [OK]

Hint: get_or_set returns cached value if key exists [OK]

Common Mistakes:

• Assuming second lambda runs and returns 'data2'
• Expecting None if key exists
• Thinking it raises an error on duplicate keys

4. You wrote this code but it raises an error:

cache = RedisCache()
result = cache.get_or_set('key', expensive_call())

What is the likely cause of the error?

medium

A. You passed the result of expensive_call() instead of a function

B. RedisCache does not support get_or_set method

C. The key must be an integer, not a string

D. expensive_call() must be imported from langchain.cache

Solution

1. Step 1: Check get_or_set argument types

   get_or_set expects a key and a function to call if missing, not the function result.

2. Step 2: Identify error cause

   Passing expensive_call() calls it immediately, causing error or unwanted behavior.

3. Final Answer:

   You passed the result of expensive_call() instead of a function -> Option A

4. Quick Check:

   Pass function, not result, to get_or_set [OK]

Hint: Use lambda or function, not direct call, in get_or_set [OK]

Common Mistakes:

• Calling function instead of passing it
• Assuming RedisCache lacks get_or_set
• Using wrong key types

5. You want to reduce API costs by caching results in Langchain. Your app runs on multiple servers. Which caching strategy is best to share cached data across servers?

hard

A. Cache results only in local files on each server

B. Use an in-memory cache on each server separately

C. Disable caching to avoid stale data

D. Use a Redis cache shared by all servers

Solution

1. Step 1: Understand multi-server caching needs

   Multiple servers require a shared cache to avoid duplicate expensive calls.

2. Step 2: Evaluate cache types

   In-memory caches are local to each server; Redis is a shared external cache accessible by all servers.

3. Final Answer:

   Use a Redis cache shared by all servers -> Option D

4. Quick Check:

   Shared cache for multi-server = Redis [OK]

Hint: Use shared Redis cache for multi-server apps [OK]

Common Mistakes:

• Using local in-memory cache for multi-server apps
• Disabling caching unnecessarily
• Relying on local files which are not shared