Using Compressa API

Compressa has an integrated API layer compatible with OpenAI's Chat Completion API. This means developers can use the existing OpenAI client library (including the Langchain library) and adapt current code with minimal changes to work with Compressa.

The full REST API schema compatible with OpenAI is available here.

Here's how to replace OpenAI models with Compressa

Python (клиент OpenAI)

from openai import OpenAI

client = OpenAI(
  base_url="https://compressa-api.mil-team.ru/v1",
  api_key="Your_Compressa_API_key"
)

response = client.chat.completions.create(
  model="Compressa-LLM",
  messages=[
      {"role": "system", "content": "You can write funny jokes."},
      {"role": "user", "content": "Write a short and funny joke about a programmer."}
  ],
  stream=False
)

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

Python (OpenAI Langchain)

# pip install langchain langchain-openai openai

from langchain_openai import ChatOpenAI
from langchain_core.messages import SystemMessage, HumanMessage

llm = ChatOpenAI(
  base_url="https://compressa-api.mil-team.ru/v1",
  api_key="Your_Compressa_API_key",
  model="Compressa-LLM"
)

messages = [
  SystemMessage(content="You can write funny jokes."),
  HumanMessage(content="Write a short and funny joke about a programmer."),
]

ai_msg = llm.invoke(messages)
print(f"Model response: {ai_msg.content}")

You can also enable token streaming option available in the OpenAI client

#pip install openai - если у вас еще нет этого пакета

from openai import OpenAI

client = OpenAI(
    base_url="https://compressa-api.mil-team.ru/v1", 
    api_key="Your_Compressa_API_key" 
)

completion = client.chat.completions.create(
  model="Compressa-LLM",
  messages=[
    {"role": "system", "content": "You can write funny jokes."},
    {"role": "user", "content": "Write a short and funny joke about a programmer."}
  ],
  stream=True
)

for chunk in completion:
  print(chunk.choices[0].delta)

#ChoiceDelta(content='A', function_call=None, refusal=None, role='assistant', tool_calls=None)
#ChoiceDelta(content=' p', function_call=None, refusal=None, role='assistant', tool_calls=None)
#ChoiceDelta(content='r', function_call=None, refusal=None, role='assistant', tool_calls=None)
#ChoiceDelta(content='o', function_call=None, refusal=None, role='assistant', tool_calls=None)
#...