Skip to content

Latest commit

 

History

History
170 lines (129 loc) · 5.07 KB

openai_api.md

File metadata and controls

170 lines (129 loc) · 5.07 KB
Raw

OpenAI-Compatible RESTful APIs

FastChat provides OpenAI-compatible APIs for its supported models, so you can use FastChat as a local drop-in replacement for OpenAI APIs. The FastChat server is compatible with both openai-python library and cURL commands.

The following OpenAI APIs are supported:

The REST API can be seamlessly operated from Google Colab, as demonstrated in the FastChat_API_GoogleColab.ipynb notebook, available in our repository. This notebook provides a practical example of how to utilize the API effectively within the Google Colab environment.

RESTful API Server

First, launch the controller

python3 -m fastchat.serve.controller

Then, launch the model worker(s)

python3 -m fastchat.serve.model_worker --model-path lmsys/vicuna-7b-v1.5

Finally, launch the RESTful API server

python3 -m fastchat.serve.openai_api_server --host localhost --port 8000

Now, let us test the API server.

OpenAI Official SDK

The goal of openai_api_server.py is to implement a fully OpenAI-compatible API server, so the models can be used directly with openai-python library.

First, install OpenAI python package >= 1.0:

pip install --upgrade openai

Then, interact with the Vicuna model:

import openai

openai.api_key = "EMPTY"
openai.base_url = "http://localhost:8000/v1/"

model = "vicuna-7b-v1.5"
prompt = "Once upon a time"

# create a completion
completion = openai.completions.create(model=model, prompt=prompt, max_tokens=64)
# print the completion
print(prompt + completion.choices[0].text)

# create a chat completion
completion = openai.chat.completions.create(
  model=model,
  messages=[{"role": "user", "content": "Hello! What is your name?"}]
)
# print the completion
print(completion.choices[0].message.content)

Logprobs

Logprobs are supported with the OpenAI API:

# create a chat completion
completion = openai.chat.completions.create(
  model=model,
  messages=[{"role": "user", "content": "Hello! What is your name?"}],
  log_probs=True
)
# print the completion
print(completion.choices[0].logprobs)

Streaming

Streaming is also supported. See test_openai_api.py. If your api server is behind a proxy you'll need to turn off buffering, you can do so in Nginx by setting proxy_buffering off; in the location block for the proxy.

cURL

cURL is another good tool for observing the output of the api.

List Models:

curl http://localhost:8000/v1/models

Chat Completions:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vicuna-7b-v1.5",
    "messages": [{"role": "user", "content": "Hello! What is your name?"}]
  }'

Text Completions:

curl http://localhost:8000/v1/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vicuna-7b-v1.5",
    "prompt": "Once upon a time",
    "max_tokens": 41,
    "temperature": 0.5
  }'

Embeddings:

curl http://localhost:8000/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vicuna-7b-v1.5",
    "input": "Hello world!"
  }'

Running multiple

If you want to run multiple models on the same machine and in the same process, you can replace the model_worker step above with a multi model variant:

python3 -m fastchat.serve.multi_model_worker \
    --model-path lmsys/vicuna-7b-v1.5 \
    --model-names vicuna-7b-v1.5 \
    --model-path lmsys/longchat-7b-16k \
    --model-names longchat-7b-16k

This loads both models into the same accelerator and in the same process. This works best when using a Peft model that triggers the PeftModelAdapter.

TODO: Base model weight optimization will be fixed once this Peft issue is resolved.

LangChain Support

This OpenAI-compatible API server supports LangChain. See LangChain Integration for details.

Adjusting Environment Variables

Timeout

By default, a timeout error will occur if a model worker does not response within 100 seconds. If your model/hardware is slower, you can change this timeout through an environment variable:

export FASTCHAT_WORKER_API_TIMEOUT=<larger timeout in seconds>

Batch size

If you meet the following OOM error while creating embeddings. You can use a smaller batch size by setting

export FASTCHAT_WORKER_API_EMBEDDING_BATCH_SIZE=1

Todos

Some features to be implemented:

  • Support more parameters like logit_bias, user, presence_penalty and frequency_penalty
  • Model details (permissions, owner and create time)
  • Edits API
  • Rate Limitation Settings