Caching strategies for cost reduction in LangChain

Introduction

Caching saves previous results so you don't pay or wait for the same work twice. It helps reduce costs and speeds up your app.

When you call an expensive API multiple times with the same input

When you want faster responses for repeated questions or tasks

When you want to limit usage of paid services to save money

When you build a chatbot that often repeats answers

When you want to avoid unnecessary processing for unchanged data

Syntax

LangChain

from langchain.cache import InMemoryCache

cache = InMemoryCache()

# Use cache to store and retrieve results
result = cache.get_or_set(key, lambda: expensive_function(input))

Use get_or_set to check cache first, then run function if missing.

LangChain supports different cache backends like memory, Redis, or custom stores.

Examples

This caches the answer to 'What is AI?' so next time it returns instantly without calling the model.

LangChain

from langchain.cache import InMemoryCache

cache = InMemoryCache()

key = 'user_question_1'
result = cache.get_or_set(key, lambda: llm.generate('What is AI?'))

This uses Redis to cache results, which works well for multiple app instances sharing the cache.

LangChain

from langchain.cache import RedisCache

cache = RedisCache(redis_url='redis://localhost:6379')

key = 'expensive_call'
result = cache.get_or_set(key, lambda: expensive_api_call())

Sample Program

This example shows caching an expensive calculation. The first call runs the function and caches the result. The second call returns the cached result instantly without running the function again.

LangChain

from langchain.cache import InMemoryCache

cache = InMemoryCache()

# Simulate an expensive function

def expensive_function(x):
    print('Running expensive function...')
    return x * x

key = 'square_4'

# First call runs the function
result1 = cache.get_or_set(key, lambda: expensive_function(4))
print('First call result:', result1)

# Second call uses cache, no print from function
result2 = cache.get_or_set(key, lambda: expensive_function(4))
print('Second call result:', result2)

Important Notes

Cache keys should be unique and descriptive to avoid collisions.

Cached data may become stale; consider cache expiration if needed.

Choose cache backend based on your app scale and persistence needs.

Summary

Caching stores results to avoid repeated work and reduce costs.

Use get_or_set to check cache before running expensive calls.

Pick the right cache type for your app, like in-memory or Redis.

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