01. LLM models available

LLM models available

Copy

# Configuration file for managing API KEY as environment variable
from dotenv import load_dotenv

# Load API KEY information
load_dotenv()

Copy

True

Copy

# Set up LangSmith tracking. https://smith.langchain.com
# !pip install langchain-teddynote
from langchain_teddynote import logging
from langchain_teddynote.messages import stream_response

# Enter a project name.
logging.langsmith("CH04-Models")

Copy

 Start tracking LangSmith. 
[Project name] 
CH04-Models 

OpenAI

Summary

OpenAI offers a Large Language Model (LLM) dedicated to chatting. When creating this model, you can specify various options, and these options affect how the model works.

Option details

temperature

• Option to set sampling temperature. Values can be selected between 0 and 2. High values (e.g. 0.8) make the output more random, and lower values (e.g. 0.2) make the output more focused and deterministic.

max_tokens

• Specifies the maximum number of tokens to be created in the chat completion. This option controls the length of the text that the model can generate at once.

model_name

• Option to select an applicable model. For more information OpenAI model documentation You can check in.

Model specification

• Link: https://platform.openai.com/docs/models/gpt-4o

Model name

Explanation

Context length

Max output token

Learning data

gpt-4o

The latest multi-mode flagship model cheaper and faster than the GPT-4 turbo

128,000 tokens

16,384 tokens

Until October 2023

gpt-4-turbo

The latest GPT-4 turbo model. Vision function, JSON mode, function call support

128,000 tokens

4,096 tokens

Until December 2023

gpt-4o-mini

Small model with better performance than the GPT-3.5 turbo

128,000 tokens

16,384 tokens

Until October 2023

o1-preview

Reasoning model for solving difficult problems in various domains

128,000 tokens

32,768 tokens

Until October 2023

o1-mini

Quick reasoning model specialized in coding, math and science

128,000 tokens

65,536 tokens

Until October 2023

gpt-4o-realtime

Audio and text input processing model for real-time API (beta)

128,000 tokens

4,096 tokens

Until October 2023

Copy

from langchain_openai import ChatOpenAI

# ChatOpenAI create an object.
gpt = ChatOpenAI(
    temperature=0,
    model_name="gpt-4o",  # model name
)

# For streaming output invoke() instead stream()I use.
answer = gpt.stream("What is love?")

# print answer
stream_response(answer)

Copy

Love is a very complex and concept that manifests itself in the form of various emotions, actions, and relationships. In general, love includes deep affection and interest, and dedication to others. Love can appear in many forms, some of which are: 

One. **Romantic love**: A love that includes deep emotional and physical attraction between two people. It appears mainly among lovers or spouses. 

2. **Family love**: Love between parents, children and family members such as siblings. This love includes protection, support, and unconditional affection. 

3. **Friendship**: With love between friends, based on mutual trust, respect and pleasure. 

4. **Self-love**: A heart that respects and cares for itself. This is important to maintaining a healthy self-respect and mental and physical well-being. 

5. ** Altruistic love**: A love dedicated to the happiness and well-being of others. This can appear in the form of charity or service. 

Love can be experienced and expressed differently from person to person, and its meaning may vary depending on cultural, social and personal factors. 

Anthropic

Anthropic is an American startup focused on artificial (AI) safety and research. The main information is:

• Year established : 2021 years

• location : San Francisco, USA

• Founder : OpenAI employees (Daniela Amodei and Dario Amodei, etc.)

• Corporate form : Registered as a Public Benefit Corporation

Claude

Claude is Anthropic's representative large-scale language model (LLM) family.

• API key issuance : https://console.anthropic.com/settings/keys

• Model list : https://docs.anthropic.com/en/docs/about-claude/models

model_name

model

Anthropic API

AWS Bedrock

GCP Vertex AI

Claude 3.5 Opus

Coming soon

Coming soon

Coming soon

Coming soon

Claude 3.5 Sonnet

claude-3-5-sonnet-20241022

claude-3-5-sonnet-20241022

anthropic.claude-3-5-sonnet-20241022-v2:0

claude-3-5-sonnet-v2@20241022

Claude 3.5 Haiku

Coming soon

Coming soon

Coming soon

Coming soon

Claude 3 Opus

claude-3-opus-20240229

claude-3-opus-20240229

anthropic.claude-3-opus-20240229-v1:0

claude-3-opus@20240229

Claude 3 Sonnet

claude-3-sonnet-20240229

claude-3-sonnet-20240229

anthropic.claude-3-sonnet-20240229-v1:0

claude-3-sonnet@20240229

Claude 3 Haiku

claude-3-haiku-20240307

claude-3-haiku-20240307

anthropic.claude-3-haiku-20240307-v1:0

claude-3-haiku@20240307

Copy

from langchain_anthropic import ChatAnthropic

# ChatAnthropic create an object.
anthropic = ChatAnthropic(model_name="claude-3-5-sonnet-20241022")

# For streaming output invoke() instead stream()I use.
answer = anthropic.stream("What is love?")

# Print answer
stream_response(answer)

Copy

Love is a very complex and subjective emotion. The following aspects may be included: 

One. Emotional side 
-Deep affection for the other person 
-Happy and Sullem 
-Sacrifice and consideration for opponents 

2. Behavioral aspects 
-The desire to be together 
-Find to protect and help the other person 
-Efforts for opponent happiness 

3. Biological aspects 
-Hormonal changes 
-Physical reaction 

4. Socio-cultural aspects 
-Different expressions depending on culture and age 
-Relationship with social norms 

Love is a very personal feeling that is experienced and expressed differently for each individual, and can change and grow over time. 

Perplexity

Link: https://www.perplexity.ai/

• Seonhang : 2022 years

• Main investors : Jeff Bezos, Nvidia, Databricks, Bessemer Venture Partners, IVP, Wayra, etc.

• Recent funding : $500 million (October 2024)

• Corporate value : About 90 billion dollars (as of November 2024)

• Active monthly users : 15 million people

Perplexity Pro exact features

• Daily Pro Search : 300 rounds

• AI model selection : GPT-4 Omni, Claude 3 Sonnet/Haiku, Sonar Large 32k[5]

• File analysis : PDF, CSV, image file support

• price : month In additionhasyear20In additionhasyear 200

How to use Perplexity API

price

• API credit acquisition

• Perplexity Pro subscription API credits equivalent to $5 per month

• API model options

• Llama 3 based model

• Perplexity Online LLM

• Citation function included

• API key issuance: API console

After issuing API key .env Save key to file

Copy

PPLX_API_KEY=here API Enter your key.

hump

Copy

import os

os.environ["PPLX_API_KEY"] = "Here API Enter your key."

Reference

• API documentation

Support model

Model

Parameter Count

Context Length

Model Type

llama-3.1-sonar-small-128k-online

8B

127,072

Chat Completion

llama-3.1-sonar-large-128k-online

70B

127,072

Chat Completion

llama-3.1-sonar-huge-128k-online

405B

127,072

Chat Completion

ChatPerplexity parameter

model Specify the language model to use (eg "llama-3.1-sonar-small-128k-online")-determine basic performance and abilities.

temperature Adjust the randomness of the responses (0.0-1.0), 0 is decisive, 1 is the most unlawful response generation.

top_p Setting the probability threshold for token sampling (0.0-1.0), the higher the more various outputs allowed.

search_domain_filter Limit search results to specified domains, provide in list form (eg ["perplexity.ai"]).

return_images A disadvantage flag that determines whether an image is included in the response.

return_related_questions Bulle flag to enable/disable related question suggestion function.

top_k Limit the number of search results to use (meaning 0 is unlimited).

streaming A disadvantage flag that determines whether a response will be received by streaming or in a finished form.

presence_penalty Penalty for token repetition (-2.0 to 2.0), higher to suppress reuse.

frequency_penalty General/rare token preference adjustment (-2.0 to 2.0), higher, rare token preference.

Copy

from langchain_teddynote.models import ChatPerplexity

perplexity = ChatPerplexity(
    model="llama-3.1-sonar-large-128k-online",
    temperature=0.2,
    top_p=0.9,
    search_domain_filter=["perplexity.ai"],
    return_images=False,
    return_related_questions=True,
    # search_recency_filter="month",
    top_k=0,
    streaming=False,
    presence_penalty=0,
    frequency_penalty=1,
)

Output the response. ChatPerplexity The source of knowledge information citations Save to property.

Copy

# Response Output
response = perplexity.invoke("2024Please research the Nobel Prize in Literature winners of the year")
print(response.content)

print()
for i, citation in enumerate(response.citations):
    print(f"[{i+1}] {citation}")

Copy

In 2024, the Nobel Prize was awarded by Korean novelist Han Kang. 

Han Kang was recognized for his work showing the innovation of modern prose through the intense poetic and experimental stylistic “that confronted the historical trauma and revealed the fragility of human life. Of his main works, 『vegetarians 』, 『boys come 』, 『do not say goodbye 』, etc. were specifically mentioned [1][3][4]. 

Han Kang is the first Asian female writer to win the Nobel Prize in literature, which is a major event in Korean literature and Asian literature. In addition, he previously won the 2016 Manbucker Award and the 2023 Medici Foreign Literature Award [1][4]. 

[1] https://times.postech.ac.kr/news/articleView.html?idxno=23342 
[2] https://www.segye.com/newsView/20241115508113 
[3] https://www.hani.co.kr/arti/science/science_general/1165375.html 
[4] https://www.umnews.org/ko/news/about-han-kangs-nobel-prize-in-literature 
[5] https://news.kbs.co.kr/news/pc/view/view.do?ncd=8097555 

Streaming output

Copy

response = perplexity.stream("2024Please research the Nobel Prize in Literature winners of the year")

for token in response:
    print(token.content, end="", flush=True)

print("\n")
for i, citation in enumerate(token.citations):
    print(f"[{i+1}] {citation}")

Copy

In 2024, the Nobel Prize was awarded by Korean novelist Han Kang. 

Han Kang was recognized for his work showing the innovation of modern prose through the intense poetic and experimental stylistic “that confronted the historical trauma and revealed the fragility of human life. The Swedish Korean Forest appreciated his works, especially 『vegetarian 』, 『boy coming 』, 『not saying goodbye 』. Han Kang is the first Asian female writer and the first Korean Nobel Prize-winning writer [1][3][4]. 

[1] https://times.postech.ac.kr/news/articleView.html?idxno=23342 
[2] https://www.segye.com/newsView/20241115508113 
[3] https://www.hani.co.kr/arti/science/science_general/1165375.html 
[4] https://www.umnews.org/ko/news/about-han-kangs-nobel-prize-in-literature 
[5] https://news.kbs.co.kr/news/pc/view/view.do?ncd=8097555 

Cohere

Cohere is a leading AI company that provides enterprise-wide artificial intelligence solutions, developing a large language model (LLM) to help companies easily introduce and leverage AI technology.

Cohere overview

• Seonhang : 2020 years

• Main investors : Inovia Capital, NVIDIA, Oracle, Salesforce Ventures

• Series C funding : $270 million childish

• Corporate mission : Providing an AI platform for businesses

Main products

Command R+

Command R+ is Cohere's latest LLM optimized for business use.

Main features

• Long context window : 128k token support

• Advanced RAG function : Provides search enhancement creation function

• Multilingual support : Support for 10 major business languages

• Automation tool usage function : Automating complex business processes

Aya

Aya is an open source multilingual LLM developed by Cohere for AI, a non-profit research institute in Cohere.

Main features

• Language support : 101 languages supported (more than twice the existing open source model)

• Training dataset : Revealing a large multilingual training dataset containing 500 million data points

Copy

from langchain_cohere import ChatCohere

# ChatCohere creaet an object.
cohere = ChatCohere(temperature=0)

# For streaming output invoke() instead stream()I use.
answer = cohere.stream("What is love?")

# print answer
stream_response(answer)

Copy

Love is a complex and intense emotion, which can be defined in a variety of ways. In general, love means an emotional and mental state that includes deep affection, interest, warmth, intimacy, devotion, consideration, respect, and strong bond with each other. 

Love can appear in various forms: 
-Romantic love: usually includes deep affection, passion, intimacy between the two, and devotion to each other. 
-Family love: refers to feelings such as attachment, protection, and care between parents and children, siblings, or family members. 
-Friendship: Emotions that include trust, respect, mutual understanding between friends, and support for each other. 
-Self-love: means acceptance, respect, care, and self-worth for yourself. 

Love can be interpreted differently depending on the context of culture, personal experience, and relationships. Some people define love as an emotional connection, while others believe that love is expressed through action and devotion. Love is often considered to bring happiness, joy, comfort, and meaning to life, but sometimes it is also a complex and challenging feeling. 

Love is one of the key elements of human relations and is a subject that is widely explored and discussed in literature, art, philosophy, and psychology. Understanding of the nature and meaning of love can vary from person to person and is shaped according to their experiences and perspectives. 

Upstage

Upstage is a domestic startup specialized in artificial intelligence (AI) technology, especially large language models (LLM) and document AI.

Main products and technologies

Solar LLM

• Main features : Upstage's flagship large-scale language model, noted for its fast performance and cost effectiveness.

• Technical approach : Maximize performance by applying Depth-Up Scaling (DUS) technology.

• Platform integration : Integrated into API through various platforms such as Amazon SageMaker JumpStart.

Document AI Pack

• function : A document processing solution based on OCR technology, which accurately extracts and digitizes what is needed from complex documents.

AskUp Seargest

• Features : Provides personalized search and referral services, and is an upgraded version of the existing ChatGPT integrated free chatbot AskUp.

API key issuance

API key issuance here Available in.

Copy

# UPSTAGE API KEY set
# import os

# os.environ["UPSTAGE_API_KEY"] = "here API KEY enter."

Copy

from langchain_upstage import ChatUpstage

# ChatUpstage create an object.
upstage = ChatUpstage(model="solar-pro")

# For streaming output invoke() instead stream() insted.
answer = upstage.stream("What is love?")

# print answer
stream_response(answer)

Copy

Love is a feeling of deep affection and strong attachment. It can appear in a variety of relationships, including family, friends, and lovers. Love consists of various elements such as consideration, respect, trust, understanding, and devotion. Love makes us happy, gives us purpose and meaning, and strengthens our relationship with others. 

Xionic

Sionic AI is a promising geothermal startup in South Korea, developing enterprise-generated AI solutions. Here are the main information about this company:

Main products

1. STORM Platform : A platform that allows companies to apply generated AI right away without technical concerns

2. STORM Answer : Aiming to improve business productivity with enterprise-optimized, productive AI solutions

3. Xionic : Korean AI model with commercial license

Korean model of commercial license

• Link: https://github.com/sionic-ai/xionic (Note) 2024.11.29 Standard Usage Value (API server error)

Copy

from langchain_openai import ChatOpenAI
from langchain_teddynote.messages import stream_response

# 2024. 11. 21 update code
xionic = ChatOpenAI(
    model_name="xionic-1-72b-20240919",
    base_url="https://sionic.chat/v1/",
    api_key="934c4bbc-c384-4bea-af82-1450d7f8128d",
)

# For streaming output invoke() instead stream()I use.
answer = xionic.stream("What is love?")

# print answer
stream_response(answer)

LogicKor

LogicKor is a benchmark leaderboard created to evaluate the multi-sector thinking power of Korean language models.

purpose

Benchmarks for measuring thinking in various fields of Korean language model

Evaluation area

• Korean reasoning

• Math

• writing

• Coding

• comprehension

Main features

1. Various model evaluation : You can compare the performance of various language models at home and abroad

2. Objective performance measurement : Provide objective indicators by evaluating the actual performance of the model in varying degrees

3. Open source : An open platform that anyone can access and check the results

LogicKor Leaderboard is an important tool for the development of the Korean AI model, and continuous improvement and development are expected.

• link: LogicKor leaderboard