Cohereを試す

cohereというサービスについて教えてください。

OpenAIやAnthropicとの違いは何でしょうか？

LLMを使ったアプリ開発者の立場から見て、それぞれの違いはどういうところにあるでしょうか？

!pip install cohere

from google.colab import userdata
import cohere

co = cohere.Client(userdata.get('COHERE_API_KEY'))

response = co.generate(
  prompt='問題:富士山の高さを2倍すると何m？\n答え:',
)
print(response)

[cohere.Generation {
	id: 9dd56c9d-558b-44d6-8ab4-db7f2465e61b
	prompt: 問題:富士山の高さを2倍すると何m？
答え:
	text:  4000m
	likelihood: None
	finish_reason: COMPLETE
	token_likelihoods: None
}]

response = co.generate(
  prompt='問題:現在の日本の総理大臣は？\n答え:',
)
print(response)

[cohere.Generation {
	id: 4f2239de-8bbc-43e8-a6d1-9da50f84fd3e
	prompt: 問題:現在の日本の総理大臣は？
答え:
	text:  Fumio Kishida
	likelihood: None
	finish_reason: COMPLETE
	token_likelihoods: None
}]

response = co.generate(
  prompt='こんにちは。今日の',
  k=1,
  temperature=0
)
print(response)

[cohere.Generation {
	id: 806b3d9c-3500-435f-9f0b-c1f6c2befb0a
	prompt: こんにちは。今日の
	text:  こんにちは! You've caught me on a good day as today is Thursday, and thus it is Thursday in Japan as well! 今日は?
	likelihood: None
	finish_reason: COMPLETE
	token_likelihoods: None
}]

response = co.chat(
  chat_history=[
    {"role": "USER", "message": "重力を発見したのは誰？"},
    {"role": "CHATBOT", "message": "重力の発見者として広く認知されているのはアイザック・ニュートンです。"}
  ],
  message="彼が生まれたのは何年？",
)
print(response)

cohere.Chat {
	id: f808f4f4-d6c2-455a-a81a-f7eed20fbfab
	response_id: f808f4f4-d6c2-455a-a81a-f7eed20fbfab
	generation_id: 8016f7b3-6f7a-4ba6-8ef5-38909ffa7548
	message: 彼が生まれたのは何年？
	text: 西暦十一十三年十月archar(11)の日に、アイザック・ニュートンは世界に出産されました（1870年）。
	conversation_id: None
	prompt: None
	chat_history: None
	preamble: None
	client: <cohere.client.Client object at 0x7daa7a880700>
	token_count: {'prompt_tokens': 150, 'response_tokens': 60, 'total_tokens': 210, 'billed_tokens': 187}
	meta: {'api_version': {'version': '1'}}
	is_search_required: None
	citations: None
	documents: None
	search_results: None
	search_queries: None
}

response = co.chat(
  chat_history=[
    {"role": "USER", "message": "重力を発見したのは誰？"},
    {"role": "CHATBOT", "message": "重力の発見者として広く認知されているのはアイザック・ニュートンです。"}
  ],
  message="彼の妻の名前は？",
  connectors=[{"id": "web-search"}],
)
print(response)

cohere.Chat {
	id: 843d1fdd-6ef1-4cc1-b5ab-f82e5bac1718
	response_id: 843d1fdd-6ef1-4cc1-b5ab-f82e5bac1718
	generation_id: 1d5344ab-3b95-49a2-aec8-e44b1f15acbb
	message: 彼の妻の名前は？
	text: I couldn't find any information in my search, but i'll try to answer you anyway. ういドリフュンです。
	conversation_id: None
	prompt: None
	chat_history: None
	preamble: None
	client: <cohere.client.Client object at 0x7daa7a880700>
	token_count: {'prompt_tokens': 448, 'response_tokens': 32, 'total_tokens': 480, 'billed_tokens': 105}
	meta: {'api_version': {'version': '1'}}
	is_search_required: None
	citations: None
	documents: []
	search_results: [{'search_query': {}, 'document_ids': [], 'connector': {'id': 'web-search'}, 'error_message': 'web results not found in JSON', 'continue_on_failure': True}]
	search_queries: [{'text': '彼の妻の名前は？', 'generation_id': '5985f9e3-3661-4804-9835-612755dbcbeb'}]
}

response = co.chat(
  chat_history=[
    {"role": "USER", "message": "Who discovered gravity?"},
    {"role": "CHATBOT", "message": "The man who is widely credited with discovering gravity is Sir Isaac Newton"}
  ],
  message="What year was he born?",
  connectors=[{"id": "web-search"}]
)
print(response)

cohere.Chat {
	id: b13016c4-d131-4435-ac27-85f024d0494c
	response_id: b13016c4-d131-4435-ac27-85f024d0494c
	generation_id: bf1fe086-3bc7-4fb3-bc1b-1741c12f17f4
	message: What year was he born?
	text: Isaac Newton was born on January 4, 1643, but had two birth dates, as England was still using the Julian calendar, which was ten days behind the Gregorian calendar that the rest of the continent was using. This means his birthdate is recorded as December 25, 1642. He died on March 20, 1726.
	conversation_id: None
	prompt: None
	chat_history: None
	preamble: None
	client: <cohere.client.Client object at 0x7daa5e6ae8c0>
	token_count: {'prompt_tokens': 1412, 'response_tokens': 67, 'total_tokens': 1479, 'billed_tokens': 90}
	meta: {'api_version': {'version': '1'}}
	is_search_required: None
	citations: [{'start': 25, 'end': 40, 'text': 'January 4, 1643', 'document_ids': ['web-search_4:0', 'web-search_2:8']}, {'start': 50, 'end': 65, 'text': 'two birth dates', 'document_ids': ['web-search_2:8']}, {'start': 70, 'end': 113, 'text': 'England was still using the Julian calendar', 'document_ids': ['web-search_2:7']}, {'start': 125, 'end': 140, 'text': 'ten days behind', 'document_ids': ['web-search_0:69', 'web-search_2:7']}, {'start': 145, 'end': 163, 'text': 'Gregorian calendar', 'document_ids': ['web-search_0:69', 'web-search_2:7']}, {'start': 173, 'end': 205, 'text': 'rest of the continent was using.', 'document_ids': ['web-search_2:7']}, {'start': 221, 'end': 264, 'text': 'birthdate is recorded as December 25, 1642.', 'document_ids': ['web-search_0:54', 'web-search_0:69', 'web-search_2:7']}, {'start': 268, 'end': 291, 'text': 'died on March 20, 1726.', 'document_ids': ['web-search_0:54']}]
	documents: [{'id': 'web-search_4:0', 'snippet': 'Arts & Entertainment\n\nIsaac Newton Birthday\n\nIsaac Newton, born January 4, 1643, was an English physicist and mathematician. He developed the laws of motion and is credited as one of the great minds of the 17th-century Scientific Revolution. In 1687, Newton published his most acclaimed work, “Philosophiae Naturalis Principia Mathematica” (“Mathematical Principles of Natural Philosophy”), which is regarded as the most significant book on physics.\n\nMarch 31, 1727 (age 84)\n\nIsaac Newton was born on January 4, 1643, in Lincolnshire, England. Newton was the only child of a local farmer, also named Isaac, who died three months before his son was born.', 'title': 'Isaac Newton - Age, Bio, Birthday, Family, Net Worth | National Today', 'url': 'https://nationaltoday.com/birthday/isaac-newton/'}, {'id': 'web-search_2:8', 'snippet': ' Artist’s concept via NASA.\n\nBottom line: Isaac Newton could claim two birth dates, but now we celebrate his birthday as January 4, 1643. Newton’s work in gravity and the laws of motion form the basis of much of today’s understanding of physics and astronomy.\n\n34X885Facebook6Pinterest23BufferShare\n\nDeborah Byrd View Articles\n\nDeborah Byrd created the EarthSky radio series in 1991 and founded EarthSky.org in 1994. Today, she serves as Editor-in-Chief of this website. She has won a galaxy of awards from the broadcasting and science communities, including having an asteroid named 3505 Byrd in her honor. In 2020, she was the Education Prize from the American Astronomical Society, the largest organization of professional astronomers in North America.', 'title': 'EarthSky | Isaac Newton born today in 1643', 'url': 'https://earthsky.org/human-world/this-date-in-science-isaac-newtons-birthday/'}, {'id': 'web-search_2:7', 'snippet': '\n\nThe rest of the continent had already adopted the Gregorian calendar, which is the same calendar we use today. However, at the time of Newton’s birth, the English were still using the Julian calendar, which lagged ten days behind because of a faulty method of accounting for leap years. (Coincidentally, 1642 was the year that Galileo died).\n\nSo Newton himself would have said his birthday was December 25. But everywhere outside of England he was born on January 4. Read more about Newton’s birthday discrepancy.\n\nEinstein’s 1916 theory of general relativity didn’t replace Newton’s theory of gravity. But it did change our understanding of gravity so that now we know massive objects cause a distortion in space-time, which passing objects feel as gravity.', 'title': 'EarthSky | Isaac Newton born today in 1643', 'url': 'https://earthsky.org/human-world/this-date-in-science-isaac-newtons-birthday/'}, {'id': 'web-search_0:69', 'snippet': " At Newton's birth, Gregorian dates were ten days ahead of Julian dates; thus, his birth is recorded as taking place on 25 December 1642 Old Style, but it can be converted to a New Style (modern) date of 4 January 1643. By the time of his death, the difference between the calendars had increased to eleven days. Moreover, he died in the period after the start of the New Style year on 1 January but before that of the Old Style new year on 25 March. His death occurred on 20 March 1726, according to the Old Style calendar, but the year is usually adjusted to 1727. A full conversion to New Style gives the date 31 March 1727.\n\n^ This claim was made by William Stukeley in 1727, in a letter about Newton written to Richard Mead.", 'title': 'Isaac Newton - Wikipedia', 'url': 'https://en.wikipedia.org/wiki/Isaac_Newton'}, {'id': 'web-search_0:54', 'snippet': " Diligent, sagacious and faithful, in his expositions of nature, antiquity and the holy Scriptures, he vindicated by his philosophy the majesty of God mighty and good, and expressed the simplicity of the Gospel in his manners. Mortals rejoice that there has existed such and so great an ornament of the human race! He was born on 25th December 1642, and died on 20th March 1726.\n\nIn a 2005 survey of members of Britain's Royal Society (formerly headed by Newton) asking who had the greater effect on the history of science, Newton or Albert Einstein, the members deemed Newton to have made the greater overall contribution.", 'title': 'Isaac Newton - Wikipedia', 'url': 'https://en.wikipedia.org/wiki/Isaac_Newton'}]
	search_results: [{'search_query': {'text': 'what year was isaac newton born?', 'generation_id': '3d5dc5c5-d8f8-43a6-ad3f-4065a4dc8541'}, 'document_ids': ['web-search_0:69', 'web-search_0:54', 'web-search_2:8', 'web-search_2:7', 'web-search_4:0'], 'connector': {'id': 'web-search'}}]
	search_queries: [{'text': 'what year was isaac newton born?', 'generation_id': '3d5dc5c5-d8f8-43a6-ad3f-4065a4dc8541'}]
}

text=(
  "Ice cream is a sweetened frozen food typically eaten as a snack or dessert. "
  "It may be made from milk or cream and is flavoured with a sweetener, "
  "either sugar or an alternative, and a spice, such as cocoa or vanilla, "
  "or with fruit such as strawberries or peaches. "
  "It can also be made by whisking a flavored cream base and liquid nitrogen together. "
  "Food coloring is sometimes added, in addition to stabilizers. "
  "The mixture is cooled below the freezing point of water and stirred to incorporate air spaces "
  "and to prevent detectable ice crystals from forming. The result is a smooth, "
  "semi-solid foam that is solid at very low temperatures (below 2 °C or 35 °F). "
  "It becomes more malleable as its temperature increases.\n\n"
  "The meaning of the name \"ice cream\" varies from one country to another. "
  "In some countries, such as the United States, \"ice cream\" applies only to a specific variety, "
  "and most governments regulate the commercial use of the various terms according to the "
  "relative quantities of the main ingredients, notably the amount of cream. "
  "Products that do not meet the criteria to be called ice cream are sometimes labelled "
  "\"frozen dairy dessert\" instead. In other countries, such as Italy and Argentina, "
  "one word is used fo\r all variants. Analogues made from dairy alternatives, "
  "such as goat's or sheep's milk, or milk substitutes "
  "(e.g., soy, cashew, coconut, almond milk or tofu), are available for those who are "
  "lactose intolerant, allergic to dairy protein or vegan."
)

response = co.summarize(
  text=text,
)
print(response)

SummarizeResponse(id='77bfa2f5-37a3-4ca4-b789-65a269f64c76',
  summary='Ice cream is a sweetened frozen food made from milk or cream and flavoured with a sweetener and spice. The name varies between countries and ice cream made from dairy alternatives is available.',
  meta={'api_version': {'version': '1'}}
)

from pprint import pprint

text=(
  "アイスクリームは、一般的にスナックやデザートとして食べられる甘味のある冷凍食品である。" 
  "ミルクやクリームから作られ、甘味料で味付けされる。" 
  "砂糖やその代替品、ココアやバニラなどのスパイスで味付けされる。" 
  "イチゴや桃のような果物で味付けされている。" 
  "フレーバークリームのベースと液体窒素を一緒に泡立てて作ることもできる。"
  "安定剤に加えて着色料が加えられることもある。"
  "混合物は水の凝固点以下に冷却され、空隙を取り込むために撹拌される。"
  "氷の結晶ができないようにする。その結果、滑らかな、"
  "非常に低い温度（2℃以下）で固体である半固体の泡である。"
)

response = co.summarize(
  text=text,
)
print(response)

SummarizeResponse(id='1861a112-6d0a-48e9-8267-0b8c73ed312f', 
  summary='Ice cream is a frozen dessert that is typically made from dairy products, such as milk and cream, and often combined with a sweetener, either sugar or an alternative, and a spice, such as cocoa or vanilla, or a fruit, such as strawberries or peaches. The fluid is agitated while it is cooled, usually by electric mixer. Disodium phosphate, dextrose and carrageenan can be added as stabilizers. The mixture is cooled below the freezing point of water and agitated to incorporate air spaces and to prevent detectable ice crystals from forming. The result is a smooth, semi-solid foam that is solid at very low temperatures (below 2 °C).',
  meta={'api_version': {'version': '1'}}
)

response = co.tokenize(
  text='tokenize me! :D',
  model='command' # optional
)
print(response)

cohere.Tokens {
	tokens: [10002, 2261, 2012, 8, 2792, 43]
	token_strings: ['token', 'ize', ' me', '!', ' :', 'D']
	meta: {'api_version': {'version': '1'}}
}

response = co.tokenize(
  text='トークン化してください',
  model='command'
)
print(response)

cohere.Tokens {
	tokens: [33794, 20181, 37122, 23925, 54464, 38016, 51512, 66083, 6758, 262, 6758, 251, 36442]
	token_strings: ['ト', 'ー', 'ク', 'ン', '化', 'し', 'て', 'く', '�', '�', '�', '�', 'い']
	meta: {'api_version': {'version': '1'}}
}

response = co.tokenize(
    text="明日の天気は晴れでしょう",
    model="embed-multilingual-v3.0"
)
print(response)

cohere.Tokens {
	tokens: [6, 81766, 154, 171391, 342, 237785, 12547]
	token_strings: ['', '明日', 'の', '天気', 'は', '晴れ', 'でしょう']
	meta: {'api_version': {'version': '1'}}
}

import time

models = [
    "command",
    "command-light",
    "command-nightly",
    "embed-english-v3.0",
    "embed-multilingual-v3.0",
    "embed-multilingual-light-v3.0",
    "embed-multilingual-v2.0",
]

for m in models:
    print(f"##### {m} #####\n")
    response = co.tokenize(
        text="明日の天気は晴れでしょう",
        model=m
    )
    print(response)
    print()
    time.sleep(10)     # トライアルのAPIキーはrate limitが10APIコール/分なので

##### command #####

cohere.Tokens {
	tokens: [55299, 38094, 21455, 49413, 36730, 253, 51183, 50972, 120, 74946, 50725, 38016, 6966, 237, 57354]
	token_strings: ['明', '日', 'の', '天', '�', '�', 'は', '�', '�', 'れ', 'で', 'し', '�', '�', 'う']
	meta: {'api_version': {'version': '1'}}
}

##### command-light #####

cohere.Tokens {
	tokens: [55299, 38094, 21455, 49413, 36730, 253, 51183, 50972, 120, 74946, 50725, 38016, 6966, 237, 57354]
	token_strings: ['明', '日', 'の', '天', '�', '�', 'は', '�', '�', 'れ', 'で', 'し', '�', '�', 'う']
	meta: {'api_version': {'version': '1'}}
}

##### command-nightly #####

cohere.Tokens {
	tokens: [55299, 38094, 21455, 49413, 36730, 253, 51183, 50972, 120, 74946, 50725, 38016, 6966, 237, 57354]
	token_strings: ['明', '日', 'の', '天', '�', '�', 'は', '�', '�', 'れ', 'で', 'し', '�', '�', 'う']
	meta: {'api_version': {'version': '1'}}
}

##### embed-english-v3.0 #####

cohere.Tokens {
	tokens: [1865, 1864, 1671, 1811, 100, 1672, 100, 100]
	token_strings: ['明', '日', 'の', '天', '', 'は', '', '']
	meta: {'api_version': {'version': '1'}}
}

##### embed-multilingual-v3.0 #####

cohere.Tokens {
	tokens: [6, 81766, 154, 171391, 342, 237785, 12547]
	token_strings: ['', '明日', 'の', '天気', 'は', '晴れ', 'でしょう']
	meta: {'api_version': {'version': '1'}}
}

##### embed-multilingual-light-v3.0 #####

cohere.Tokens {
	tokens: [6, 81766, 154, 171391, 342, 237785, 12547]
	token_strings: ['', '明日', 'の', '天気', 'は', '晴れ', 'でしょう']
	meta: {'api_version': {'version': '1'}}
}

##### embed-multilingual-v2.0 #####

cohere.Tokens {
	tokens: [6664, 6640, 3714, 5262, 7224, 3715, 6708, 320916, 245911]
	token_strings: ['明', '日', 'の', '天', '気', 'は', '晴', 'れで', '##しょう']
	meta: {'api_version': {'version': '1'}}
}

first_response = co.tokenize(
    text="明日の天気は晴れでしょう、そして気温も高いでしょう",
    model="command"
)

response = co.detokenize(
  tokens=first_response.tokens,
  model="command"
)
print(response)

明日の天気は晴れでしょう、そして気温も高いでしょう

first_response = co.tokenize(
    text="明日の天気は晴れでしょう、そして気温も高いでしょう",
    model="embed-multilingual-v3.0"
)

response = co.detokenize(
  tokens=first_response.tokens,
  model="embed-multilingual-v3.0"
)
print(response)

明日の天気は晴れでしょう、そして気温も高いでしょう

response = co.detect_language(
  texts=['Hello world', "'Здравствуй, Мир'", "こんにちは、世界", "안녕하세요", "世界你好"]
)
print(response)

cohere.DetectLanguageResponse {
    results: [
        Language<language_code: "en", language_name: "English">,
        Language<language_code: "ru", language_name: "Russian">,
        Language<language_code: "ja", language_name: "Japanese">,
        Language<language_code: "ko", language_name: "Korean">,
        Language<language_code: "zh", language_name: "Chinese">
    ]
     meta: {'api_version': {'version': '1'}}
}

response = co.embed(
  texts=['hello', 'goodbye'],
  model='embed-english-v3.0',
  input_type='classification'
)
print(response)

cohere.Embeddings {
    embeddings: [
        [
            0.016296387,
            -0.008354187,
            -0.04699707, 
            ...
            0.047332764,
            0.0023212433,
            0.0052719116
        ]
    ]
    compressed_embeddings: []
    meta: {'api_version': {'version': '1'}}
}

response = co.embed(
  texts=['こんにちは', '世界'],
  model='embed-multilingual-v3.0',
  input_type='classification'
)
print(response)

cohere.Embeddings {
    embeddings: [
        [
            0.0072784424,
            0.048553467,
            -0.0056419373,
            0.062683105,
            -0.03982544,
            0.0051574707
        ]
    ]
    compressed_embeddings: []
    meta: {'api_version': {'version': '1'}}
}

text = "明日の天気は晴れでしょう、そして気温も高いでしょう"

tokenize_response = co.tokenize(
    text=text,
    model="embed-multilingual-light-v3.0",
)

tokenized_embedding_response = co.embed(
  texts=tokenize_response.token_strings,
  model='embed-multilingual-v3.0',
  input_type='classification'
)

simplefilter_embedding_response = co.embed(
  texts=[text],
  model='embed-multilingual-v3.0',
  input_type='classification'
)

print(len(tokenized_embedding_response.embeddings))
print(len(simplefilter_embedding_response.embeddings))

13
1

import tiktoken

text = "東京都と周辺7県で首都圏を構成している。特に東京圏（東京都・神奈川県・千葉県・埼玉県）の総人口は約3500万人に達し、日本の人口の約30%が集中している。東京都市圏としては世界最大級の人口を有する国際的大都市である。"

encoding = tiktoken.get_encoding("cl100k_base")
print(len(encoding.encode(text)))

response = co.tokenize(
    text=text,
    model="embed-multilingual-v3.0"
)
print(response.length)

127
63

print(response.token_strings)

['', '東京都', 'と', '周辺', '7', '県', 'で', '首都', '圏', 'を', '構成', 'している', '。', '特に', '東京', '圏', '(', '東京都', '・', '神奈川県', '・', '千葉県', '・', '埼玉県', ')', 'の', '総', '人口', 'は', '約', '3', '500', '万人', 'に', '達', 'し', '、', '日本の', '人口', 'の', '約', '30%', 'が', '集中', 'している', '。', '東京都', '市', '圏', 'としては', '世界', '最大', '級', 'の', '人口', 'を', '有', 'する', '国際', '的大', '都市', 'である', '。']

from cohere.responses.classify import Example

examples=[
  Example("見たらすぐにお返事ください", "Spam"),
  Example("プレゼントの受け取り場所はコチラ", "Spam"),
  Example("1億円に当選しました！", "Spam"),
  Example("あなたのアカウントは停止されました!", "Spam"),
  Example("Amazonプライム会費のお支払い方法に問題があります。", "Spam"),
  Example("キャリアを拝見して連絡をしております！", "Spam"),
  Example("Amazonを装った不審なメール等にご注意ください", "Not spam"),
  Example("Apple からの領収書です", "Not spam"),
  Example("配達完了:ご注文商品の配達が完了しました。'", "Not spam"),
  Example("定期購入はキャンセルされました。別の商品をおすすめします", "Not spam"),
]

inputs=[
  "明日の会議について",
  "Re: ご注文にういてのお問い合わせ",
]

response = co.classify(
  inputs=inputs,
  examples=examples,
  model="embed-multilingual-v3.0"
)
print(response)

[
    Classification<prediction: "Spam", confidence: 0.6115304, labels: {'Not spam': LabelPrediction(confidence=0.3884696), 'Spam': LabelPrediction(confidence=0.6115304)}>,
    Classification<prediction: "Spam", confidence: 0.5856705, labels: {'Not spam': LabelPrediction(confidence=0.4143295), 'Spam': LabelPrediction(confidence=0.5856705)}>]
]

docs = [
    "カーソンシティはアメリカのネバダ州の州都である。",
    "北マリアナ諸島連邦は太平洋に浮かぶ島々である。首都はサイパンである。",
    "ワシントンD.C.（単にワシントンまたはD.C.とも呼ばれ、正式にはコロンビア特別区）は、アメリカ合衆国の首都である。連邦区である。",
    "死刑制度はアメリカ合衆国が国である以前から存在する。2017年現在、50州のうち30州で死刑が合法である。"
]

response = co.rerank(
  model = "rerank-multilingual-v2.0",
  query = 'アメリカ合衆国の首都は？',
  documents = docs,
  top_n = 3,
)
print(response)

[
    RerankResult<document['text']: ワシントンD.C.（単にワシントンまたはD.C.とも呼ばれ、正式にはコロンビア特別区）は、アメリカ合衆国の首都である。連邦区である。, index: 2, relevance_score: 0.99992096>,
    RerankResult<document['text']: 北マリアナ諸島連邦は太平洋に浮かぶ島々である。首都はサイパンである。, index: 1, relevance_score: 0.170508>,
    RerankResult<document['text']: カーソンシティはアメリカのネバダ州の州都である。, index: 0, relevance_score: 0.13753247>
]