hf_text-generation-inference/README.md

Text Generation Inference

A Rust, Python and gRPC server for text generation inference. Used in production at HuggingFace to power LLMs api-inference widgets.

Table of contents

• Features
• Optimized Architectures
• Get Started
  • Docker
  • API Documentation
  • A note on Shared Memory
  • Distributed Tracing
  • Local Install
  • CUDA Kernels
• Run BLOOM
  • Download
  • Run
  • Quantization
• Develop
• Testing

Features

• Serve the most popular Large Language Models with a simple launcher
• Tensor Parallelism for faster inference on multiple GPUs
• Token streaming using Server-Sent Events (SSE)
• Continuous batching of incoming requests for increased total throughput
• Optimized transformers code for inference using flash-attention and Paged Attention on the most popular architectures
• Quantization with bitsandbytes and GPT-Q
• Safetensors weight loading
• Watermarking with A Watermark for Large Language Models
• Logits warper (temperature scaling, top-p, top-k, repetition penalty, more details see transformers.LogitsProcessor)
• Stop sequences
• Log probabilities
• Production ready (distributed tracing with Open Telemetry, Prometheus metrics)

Optimized architectures

• BLOOM
• FLAN-T5
• Galactica
• GPT-Neox
• Llama
• OPT
• SantaCoder
• Starcoder
• Falcon 7B
• Falcon 40B

Other architectures are supported on a best effort basis using:

AutoModelForCausalLM.from_pretrained(<model>, device_map="auto")

or

AutoModelForSeq2SeqLM.from_pretrained(<model>, device_map="auto")

Get started

Docker

The easiest way of getting started is using the official Docker container:

model=bigscience/bloom-560m
num_shard=2
volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run

docker run --gpus all --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:0.9 --model-id $model --num-shard $num_shard

Note: To use GPUs, you need to install the NVIDIA Container Toolkit. We also recommend using NVIDIA drivers with CUDA version 11.8 or higher.

To see all options to serve your models (in the code or in the cli:

text-generation-launcher --help

You can then query the model using either the /generate or /generate_stream routes:

curl 127.0.0.1:8080/generate \
    -X POST \
    -d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17}}' \
    -H 'Content-Type: application/json'

curl 127.0.0.1:8080/generate_stream \
    -X POST \
    -d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17}}' \
    -H 'Content-Type: application/json'

or from Python:

pip install text-generation

from text_generation import Client

client = Client("http://127.0.0.1:8080")
print(client.generate("What is Deep Learning?", max_new_tokens=17).generated_text)

text = ""
for response in client.generate_stream("What is Deep Learning?", max_new_tokens=17):
    if not response.token.special:
        text += response.token.text
print(text)

API documentation

You can consult the OpenAPI documentation of the text-generation-inference REST API using the /docs route. The Swagger UI is also available at: https://huggingface.github.io/text-generation-inference.

Distributed Tracing

text-generation-inference is instrumented with distributed tracing using OpenTelemetry. You can use this feature by setting the address to an OTLP collector with the --otlp-endpoint argument.

A note on Shared Memory (shm)

NCCL is a communication framework used by PyTorch to do distributed training/inference. text-generation-inference make use of NCCL to enable Tensor Parallelism to dramatically speed up inference for large language models.

In order to share data between the different devices of a NCCL group, NCCL might fall back to using the host memory if peer-to-peer using NVLink or PCI is not possible.

To allow the container to use 1G of Shared Memory and support SHM sharing, we add --shm-size 1g on the above command.

If you are running text-generation-inference inside Kubernetes. You can also add Shared Memory to the container by creating a volume with:

- name: shm
  emptyDir:
   medium: Memory
   sizeLimit: 1Gi

and mounting it to /dev/shm.

Finally, you can also disable SHM sharing by using the NCCL_SHM_DISABLE=1 environment variable. However, note that this will impact performance.

Local install

You can also opt to install text-generation-inference locally.

First install Rust and create a Python virtual environment with at least Python 3.9, e.g. using conda:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

conda create -n text-generation-inference python=3.9 
conda activate text-generation-inference

You may also need to install Protoc.

On Linux:

PROTOC_ZIP=protoc-21.12-linux-x86_64.zip
curl -OL https://github.com/protocolbuffers/protobuf/releases/download/v21.12/$PROTOC_ZIP
sudo unzip -o $PROTOC_ZIP -d /usr/local bin/protoc
sudo unzip -o $PROTOC_ZIP -d /usr/local 'include/*'
rm -f $PROTOC_ZIP

On MacOS, using Homebrew:

brew install protobuf

Then run:

BUILD_EXTENSIONS=True make install # Install repository and HF/transformer fork with CUDA kernels
make run-bloom-560m

Note: on some machines, you may also need the OpenSSL libraries and gcc. On Linux machines, run:

sudo apt-get install libssl-dev gcc -y

CUDA Kernels

The custom CUDA kernels are only tested on NVIDIA A100s. If you have any installation or runtime issues, you can remove the kernels by using the BUILD_EXTENSIONS=False environment variable.

Be aware that the official Docker image has them enabled by default.

Run BLOOM

Download

It is advised to download the weights ahead of time with the following command:

make download-bloom

Run

make run-bloom # Requires 8xA100 80GB

Quantization

You can also quantize the weights with bitsandbytes to reduce the VRAM requirement:

make run-bloom-quantize # Requires 8xA100 40GB

Develop

make server-dev
make router-dev

Testing

# python
make python-server-tests
make python-client-tests
# or both server and client tests
make python-tests
# rust cargo tests
make rust-tests
# integration tests
make integration-tests