Getting Began with Claude 2 API

Picture by Writer

Anthropic’s conversational AI assistant, Claude 2, is the newest model that comes with vital enhancements in efficiency, response size, and availability in comparison with its earlier iteration. The newest model of the mannequin may be accessed by way of our API and a brand new public beta web site at claude.ai.

Claude 2 is thought for being simple to speak with, clearly explaining its reasoning, avoiding dangerous outputs, and having a sturdy reminiscence. It has enhanced reasoning capabilities. 

Claude 2 confirmed a major enchancment within the multiple-choice part of the Bar examination, scoring 76.5% in comparison with Claude 1.3’s 73.0%. Furthermore, Claude 2 outperforms over 90% of human check takers in studying and writing sections of the GRE. In coding evaluations akin to HumanEval, Claude 2 achieved 71.2% accuracy, which is a substantial enhance from 56.0% previously.

The Claude 2 API is being supplied to our hundreds of enterprise clients on the identical value as Claude 1.3. You possibly can simply use it by way of net API, in addition to Python and Typescript purchasers. This tutorial will information you thru the setup and utilization of the Claude 2 Python API, and provide help to be taught in regards to the varied functionalities it provides.

Earlier than we bounce into accessing the API, we have to first apply for API Early Access. You’ll fill out the shape and watch for affirmation. Be sure you are utilizing your small business electronic mail tackle. I used to be utilizing @kdnuggets.com. 

After receiving the affirmation electronic mail, you may be supplied with entry to the console. From there, you possibly can generate API keys by going to Account Settings.

Set up the Anthropic Python consumer utilizing PiP. Be sure you are utilizing the newest Python model.   

Setup the anthropic consumer utilizing the API key.

consumer = anthropic.Anthropic(api_key=os.environ["API_KEY"])

As a substitute of offering the API key for creating the consumer object, you possibly can set ANTHROPIC_API_KEY setting variable and supply it the important thing.

Right here is the essential sync model of producing responses utilizing the immediate. 

1. Import all essential modules.
2. Preliminary the consumer utilizing API key.
3. To generate a response it’s important to present a mannequin identify, max tokes, and immediate. 
4. Your immediate normally has HUMAN_PROMPT (‘nnHuman:’) and AI_PROMPT (‘nnAssistant:’).
5. Print the response. 

from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
import os


anthropic = Anthropic(
    api_key= os.environ["ANTHROPIC_API_KEY"],
)

completion = anthropic.completions.create(
    mannequin="claude-2",
    max_tokens_to_sample=300,
    immediate=f"{HUMAN_PROMPT} How do I discover soul mate?{AI_PROMPT}",
)
print(completion.completion)

Output:

As we are able to see, we obtained fairly good outcomes. I believe it’s even higher than GPT-4. 

Listed below are some ideas for locating your soulmate:

- Give attention to turning into your finest self. Pursue your passions and pursuits, develop as an individual, and work on growing your self into somebody you admire. If you end up dwelling your finest life, you'll entice the suitable individual for you.

- Put your self on the market and meet new individuals. Develop your social circles by attempting new actions, becoming a member of golf equipment, volunteering, or utilizing relationship apps. The extra individuals you meet, the extra doubtless you might be to come across somebody particular.........

You can too name Claude 2 API utilizing asynchronous requests. 

Synchronous APIs execute requests sequentially, blocking till a response is obtained earlier than invoking the following name, whereas asynchronous APIs enable a number of concurrent requests with out blocking, dealing with responses as they full by way of callbacks, guarantees or occasions; this gives asynchronous APIs better effectivity and scalability.

1. Import AsyncAnthropic as an alternative of Anthropic
2. Outline a operate with async syntax. 
3. Use await with every API name

from anthropic import AsyncAnthropic

anthropic = AsyncAnthropic()


async def essential():
    completion = await anthropic.completions.create(
        mannequin="claude-2",
        max_tokens_to_sample=300,
        immediate=f"{HUMAN_PROMPT} What share of nitrogen is current within the air?{AI_PROMPT}",
    )
    print(completion.completion)


await essential()

Output: 

We obtained correct outcomes. 

About 78% of the air is nitrogen. Particularly:
- Nitrogen makes up roughly 78.09% of the air by quantity.
- Oxygen makes up roughly 20.95% of air. 
- The remaining 0.96% is made up of different gases like argon, carbon dioxide, neon, helium, and hydrogen.

Notice: If you’re utilizing the async operate in Jupyter Pocket book, use await essential(). In any other case, use asyncio.run(essential()).

Streaming has turn into more and more fashionable for big language fashions. As a substitute of ready for the entire response, you can begin processing the output as quickly because it turns into out there. This strategy helps scale back the perceived latency by returning the output of the Language Mannequin token by token, versus all of sudden.

You simply must set a brand new argument stream to True in completion operate. Caude 2 makes use of Server Aspect Occasions (SSE) to assist the response streaming.

stream = anthropic.completions.create(
    immediate=f"{HUMAN_PROMPT} May you please write a Python code to investigate a mortgage dataset?{AI_PROMPT}",
    max_tokens_to_sample=300,
    mannequin="claude-2",
    stream=True,
)
for completion in stream:
    print(completion.completion, finish="", flush=True)

Output:

Billing is crucial facet of integrating the API in your utility. It’ll provide help to plan for the funds and cost your purchasers. Aloof the LLMs APIs are charged primarily based on token.  You possibly can examine the under desk to know the pricing construction. 

Picture from Anthropic

A simple method to rely the variety of tokens is by offering a immediate or response to the count_tokens operate.

consumer = Anthropic()
consumer.count_tokens('What share of nitrogen is current within the air?')

Other than primary response technology, you need to use the API to completely combine into your utility.

• Utilizing varieties: The requests and responses use TypedDicts and Pydantic fashions respectively for sort checking and autocomplete.
• Dealing with errors: Errors raised embody APIConnectionError for connection points and APIStatusError for HTTP errors.
• Default Headers: anthropic-version header is routinely added. This may be personalized.
• Logging: Logging may be enabled by setting the ANTHROPIC_LOG setting variable.
• Configuring the HTTP consumer: The HTTPx consumer may be personalized for proxies, transports and so on.
• Managing HTTP assets: The consumer may be manually closed or utilized in a context supervisor.
• Versioning: Follows Semantic Versioning conventions however some backwards-incompatible modifications could also be launched as minor variations.

The Anthropic Python API gives easy accessibility to Claude 2 state-of-the-art conversational AI mannequin, enabling builders to combine Claude’s superior pure language capabilities into their purposes. The API provides synchronous and asynchronous calls, streaming, billing primarily based on token utilization, and different options to completely leverage Claude 2’s enhancements over earlier variations. 

The Claude 2 is my favourite up to now, and I believe constructing purposes utilizing the Anthropic API will provide help to construct a product that outshines others.

Let me know if you need to learn a extra superior tutorial. Maybe I can create an utility utilizing Anthropic API.
 
 

Abid Ali Awan (@1abidaliawan) is an authorized information scientist skilled who loves constructing machine studying fashions. At present, he’s specializing in content material creation and writing technical blogs on machine studying and information science applied sciences. Abid holds a Grasp’s diploma in Know-how Administration and a bachelor’s diploma in Telecommunication Engineering. His imaginative and prescient is to construct an AI product utilizing a graph neural community for college students fighting psychological sickness.