Resumindo e editando texto

Trabalhando com DeepSeek em Python

James Chapman

AI Curriculum Lead, DataCamp

Recapitulando...

Perguntas e respostas

response = client.chat.completions.create(
  model="deepseek-ai/DeepSeek-V4-Pro",
  messages=[{"role": "user", "content": "How many days are in October?"}]
)

print(response.choices[0].message.content)

October has **31 days**.  

It's one of the seven months in the Gregorian calendar with 31 days...

Edição de texto

Exemplo: atualizar nome, pronomes e cargo

prompt = """
Update name to Maarten, pronouns to he/him, and job title to Senior Content Developer
in the following text:

Joanne is a Content Developer at DataCamp. Her favorite programming language is R,
which she uses for her statistical analyses.
"""

Edição de texto

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4-Pro",

    messages=[{"role": "user", "content": prompt}]

)

print(response.choices[0].message.content)

Here's the updated text with the requested changes:

Maarten is a Senior Content Developer at DataCamp. His favorite programming language
is R, which he uses for his statistical analyses.

Let me know if you'd like any further adjustments!

Resumo de texto

Exemplo: resumo de conversas do chat com clientes

Equipe de suporte ao cliente

text = """
Customer: Hi, I'm trying to log into 
my account, but it keeps saying 
my password is incorrect. I'm sure 
I'm entering the right one.  

Support: I'm sorry to hear that! 
Have you tried resetting your password?  
...
"""

Resumo de texto

prompt = f"""Summarize the customer support chat 
             in three concise key points: {text}"""


response = client.chat.completions.create(
  model="deepseek-ai/DeepSeek-V4-Pro",
  messages=[{"role": "user", "content": prompt}]
)
print(response.choices[0].message.content)

1. **Login Issue**: Customer couldn't log in due to a password issue and missing reset link.  
2. **Password Reset Suggestion**: Support resent the reset email after confirming it was sent.  
3. **Prompt Assistance**: Customer resolved the issue by using Google sign-in.

Controlando o tamanho da resposta

max_tokens = 5

response = client.chat.completions.create(
  model="deepseek-ai/DeepSeek-V4-Pro",
  messages=[{"role":"user",
      "content":"Write a haiku about AI."}],
  max_tokens=5
)

**Silent circuits hum

max_tokens = 30

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4-Pro",
    messages=[{"role":"user",
      "content":"Write a haiku about AI."}],
    max_tokens=30
)

**Silent circuits hum,**  
**thoughts of light and logic bloom—**  
**minds beyond our own.**

Entendendo tokens

Tokens: unidades de texto que ajudam a IA a entender e interpretar o texto

A frase, "How can the OpenAI API deliver business value?" com cada token destacado em uma cor diferente.

¹ https://lunary.ai/deepseek-tokenizer

Calculando o custo

Os custos de uso da API dependem da plataforma, do modelo e do número de tokens 💰
- Modelos têm preço por custo/tokens
- Tokens de entrada e saída podem ter custos diferentes
Aumentar max_tokens eleva o custo 📈

Screenshot 2025-03-05 at 11.46.54.png

Calculando o custo

prompt = f"""Summarize the customer support chat 
             in three concise key points: {text}"""

max_tokens = 500

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4-Pro",
    messages=[{"role": "user", "content": prompt}], 
    max_tokens=max_tokens
)

Calculando o custo

# Define price per token
input_token_price = 2.1 / 1_000_000
output_token_price = 4.4 / 1_000_000

# Extract token usage
input_tokens = response.usage.prompt_tokens

output_tokens = max_tokens

# Calculate cost
cost = (input_tokens * input_token_price + output_tokens * output_token_price)
print(f"Estimated cost: ${cost}")

Estimated cost: $0.0153964

Vamos praticar!

Trabalhando com DeepSeek em Python