pcuenca / candle-vllm Goto Github PK

Efficient, easy-to-use platform for inference and serving local LLMs including an OpenAI compatible API server.

candle-vllm is in active, breaking development and as such is currently unstable.

• OpenAI compatible API server provided for serving LLMs.
• Highly extensible trait-based system to allow rapid implementation of new module pipelines,
• Streaming support in generation.
• Efficient management of key-value cache with PagedAttention.
• Continuous batching.

• Llama
  • 7b
  • 13b
  • 70b
• Mistral
  • 7b

See this folder for some examples.

In your terminal, install the openai Python package by running pip install openai. I use version 1.3.5.

Then, create a new Python file and write the following code:

import openai

openai.api_key = "EMPTY"

openai.base_url = "http://localhost:2000/v1/"

completion = openai.chat.completions.create(
    model="llama7b",
    messages=[
        {
            "role": "user",
            "content": "Explain how to best learn Rust.",
        },
    ],
    max_tokens = 64,
)
print(completion.choices[0].message.content)

Next, launch a candle-vllm instance by running HF_TOKEN=... cargo run --release -- --hf-token HF_TOKEN --port 2000 llama7b --repeat-last-n 64.

After the candle-vllm instance is running, run the Python script and enjoy efficient inference with an OpenAI compatible API server!

Installing candle-vllm is as simple as the following steps. If you have any problems, please create an issue.

0. Be sure to install Rust here: https://www.rust-lang.org/tools/install
1. Run sudo apt install libssl-dev or equivalent install command
2. Run sudo apt install pkg-config or equivalent install command

The following features are planned to be implemented, but contributions are especially welcome:

• Sampling methods:
  • Beam search (huggingface/candle#1319)
• More pipelines (from candle-transformers)

• Python implementation: vllm-project
• vllm paper