Envoyer des requêtes aux modèles DeepSeek

Travailler avec DeepSeek en Python

James Chapman

AI Curriculum Lead, DataCamp

Récapitulatif

Un ordinateur envoie une requête à l'API de Together.ai, qui contient le modèle ainsi que les données et paramètres du modèle.

Créer une requête

from openai import OpenAI

# To send to DeepSeek API: base_url="https://api.deepseek.com"
client = OpenAI(api_key="<TogetherAI API Key>", base_url="https://api.together.xyz/v1")

Créez des requêtes avec la bibliothèque openai
OpenAI : développeurs de modèles et d'applis d'IA (p. ex., ChatGPT)
base_url : redirige la requête du fournisseur OpenAI vers DeepSeek
api_key : authentification à l'API (voir la doc du fournisseur)

¹ https://platform.deepseek.com/api_keys

Créer une requête

from openai import OpenAI
# To send to DeepSeek API: base_url="https://api.deepseek.com"
client = OpenAI(api_key="<TogetherAI API Key>", base_url="https://api.together.xyz/v1")


response = client.chat.completions.create(

    model="deepseek-ai/DeepSeek-V4-Pro",

    messages=[{"role": "user",
               "content": "In one sentence, what is hallucination in AI?"}]

)


print(response)

La réponse

ChatCompletion(id='onVWeEx-gBAtS-a09121fcae54578c', choices=[Choice(finish_reason='stop', index=0,
logprobs=None, message=ChatCompletionMessage(content='Hallucination in AI is when a model confidently
generates information that sounds plausible but is factually incorrect, misleading, or entirely
fabricated.', refusal=None, role='assistant', annotations=None, audio=None, function_call=None,
tool_calls=[]), seed=None)], created=1781018557, model='deepseek-ai/DeepSeek-V4-Pro',
object='chat.completion', moderation=None, service_tier=None, system_fingerprint='default',
usage=CompletionUsage(completion_tokens=28, prompt_tokens=15, total_tokens=43,
completion_tokens_details=None, prompt_tokens_details=PromptTokensDetails(audio_tokens=None,
cached_tokens=0)), prompt=[])

La réponse

ChatCompletion(id='onVWeEx-gBAtS-a09121fcae54578c',
               choices=[Choice(finish_reason='stop', index=0, logprobs=None,
                               message=ChatCompletionMessage(content='Hallucination in AI refers...',
                               refusal=None, role='assistant', annotations=None, audio=None,
                               function_call=None, tool_calls=[]), seed=None)],
               created=1781018557,
               model='deepseek-ai/DeepSeek-V4-Pro',
               object='chat.completion',
               moderation=None,
               service_tier=None,
               system_fingerprint='default',
               usage=CompletionUsage(completion_tokens=28, prompt_tokens=15, total_tokens=43,
                                     completion_tokens_details=None,
                                     prompt_tokens_details=PromptTokensDetails(audio_tokens=None,
                                          cached_tokens=0)),
               prompt=[])

Interpréter la réponse

print(response.choices)

[Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(
content='Hallucination in AI refers to the generation of false or nonsensical information
that is presented as factual, often due to limitations in training data, model biases, or
inference errors.', refusal=None, role='assistant', annotations=None, audio=None,
function_call=None, tool_calls=[]), seed=None)]

Interpréter la réponse

print(response.choices[0])

Choice(finish_reason='stop', index=0, logprobs=None, message=ChatCompletionMessage(
content='Hallucination in AI refers to the generation of false or nonsensical information
that is presented as factual, often due to limitations in training data, model biases, or
inference errors.', refusal=None, role='assistant', annotations=None, audio=None,
function_call=None, tool_calls=[]), seed=None)

Interpréter la réponse

print(response.choices[0].message)

ChatCompletionMessage(content='Hallucination in AI refers to the generation of false or
nonsensical information that is presented as factual, often due to limitations in training
data, model biases, or inference errors.', refusal=None, role='assistant', annotations=None,
audio=None, function_call=None, tool_calls=[])

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

Hallucination in AI refers to the generation of false or nonsensical information that is
presented as factual, often due to limitations in training data, model biases, or inference
errors.

Coûts d'utilisation de l'API

Pour DeepSeek :
- Fournisseur (p. ex., together.ai)
- Modèle
- Plus d'entrées + sorties = coût plus élevé

Une autoroute à péage.

¹ https://api-docs.deepseek.com/quick_start/pricing

Passons à la pratique !

Travailler avec DeepSeek en Python