hf_text-generation-inference/README.md

247 lines
7.6 KiB
Markdown
Raw Normal View History

<div align="center">
# Text Generation Inference
2022-10-08 04:30:12 -06:00
<a href="https://github.com/huggingface/text-generation-inference">
<img alt="GitHub Repo stars" src="https://img.shields.io/github/stars/huggingface/text-generation-inference?style=social">
</a>
<a href="https://github.com/huggingface/text-generation-inference/blob/main/LICENSE">
<img alt="License" src="https://img.shields.io/github/license/huggingface/text-generation-inference">
</a>
<a href="https://huggingface.github.io/text-generation-inference">
<img alt="Swagger API documentation" src="https://img.shields.io/badge/API-Swagger-informational">
</a>
2022-10-18 07:19:03 -06:00
![architecture](assets/architecture.jpg)
</div>
A Rust, Python and gRPC server for text generation inference. Used in production at [HuggingFace](https://huggingface.co)
to power LLMs api-inference widgets.
## Table of contents
- [Features](#features)
- [Optimized Architectures](#optimized-architectures)
- [Get Started](#get-started)
- [Docker](#docker)
2023-02-08 09:53:33 -07:00
- [API Documentation](#api-documentation)
- [A note on Shared Memory](#a-note-on-shared-memory-shm)
2023-02-13 05:02:45 -07:00
- [Distributed Tracing](#distributed-tracing)
- [Local Install](#local-install)
- [CUDA Kernels](#cuda-kernels)
- [Run BLOOM](#run-bloom)
- [Download](#download)
- [Run](#run)
- [Quantization](#quantization)
- [Develop](#develop)
- [Testing](#testing)
2022-10-27 06:25:29 -06:00
## Features
2023-03-03 10:42:20 -07:00
- 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)
2022-11-14 08:22:10 -07:00
- [Dynamic batching of incoming requests](https://github.com/huggingface/text-generation-inference/blob/main/router/src/batcher.rs#L88) for increased total throughput
- Quantization with [bitsandbytes](https://github.com/TimDettmers/bitsandbytes)
2022-10-27 06:25:29 -06:00
- [Safetensors](https://github.com/huggingface/safetensors) weight loading
2023-03-03 10:42:20 -07:00
- Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226)
- Logits warper (temperature scaling, top-p, top-k, repetition penalty, more details see [transformers.LogitsProcessor](https://huggingface.co/docs/transformers/internal/generation_utils#transformers.LogitsProcessor))
2022-12-12 10:25:22 -07:00
- Stop sequences
2022-12-15 09:03:56 -07:00
- Log probabilities
2023-03-03 10:42:20 -07:00
- Production ready (distributed tracing with Open Telemetry, Prometheus metrics)
2022-10-27 06:25:29 -06:00
## Optimized architectures
2022-11-07 04:53:56 -07:00
- [BLOOM](https://huggingface.co/bigscience/bloom)
- [FLAN-T5](https://huggingface.co/google/flan-t5-xxl)
2023-02-14 05:02:16 -07:00
- [Galactica](https://huggingface.co/facebook/galactica-120b)
- [GPT-Neox](https://huggingface.co/EleutherAI/gpt-neox-20b)
- [Llama](https://github.com/facebookresearch/llama)
- [OPT](https://huggingface.co/facebook/opt-66b)
- [SantaCoder](https://huggingface.co/bigcode/santacoder)
2022-10-27 06:25:29 -06:00
2023-02-14 05:02:16 -07:00
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:
```shell
model=bigscience/bloom-560m
num_shard=2
volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run
2023-02-08 09:53:33 -07:00
docker run --gpus all --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:latest --model-id $model --num-shard $num_shard
```
**Note:** To use GPUs, you need to install the [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html). We also recommend using NVIDIA drivers with CUDA version 11.8 or higher.
You can then query the model using either the `/generate` or `/generate_stream` routes:
2022-10-08 04:30:12 -06:00
```shell
curl 127.0.0.1:8080/generate \
-X POST \
2023-03-07 10:52:22 -07:00
-d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17}}' \
-H 'Content-Type: application/json'
```
2022-10-08 04:30:12 -06:00
```shell
curl 127.0.0.1:8080/generate_stream \
-X POST \
2023-03-07 10:52:22 -07:00
-d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17}}' \
-H 'Content-Type: application/json'
2022-10-08 04:30:12 -06:00
```
2023-03-03 10:42:20 -07:00
or from Python:
```shell
2023-03-07 10:52:22 -07:00
pip install text-generation
2023-03-03 10:42:20 -07:00
```
2023-03-07 10:52:22 -07:00
```python
from text_generation import Client
2023-03-03 10:42:20 -07:00
2023-03-07 10:52:22 -07:00
client = Client("http://127.0.0.1:8080")
print(client.generate("What is Deep Learning?", max_new_tokens=17).generated_text)
2023-03-03 10:42:20 -07:00
2023-03-07 10:52:22 -07:00
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)
```
2023-03-03 10:42:20 -07:00
### API documentation
2022-10-08 04:30:12 -06:00
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](https://huggingface.github.io/text-generation-inference).
2023-02-13 05:02:45 -07:00
### 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.
2023-02-08 09:53:33 -07:00
### A note on Shared Memory (shm)
[`NCCL`](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/index.html) 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:
```yaml
- 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](https://rustup.rs/) and create a Python virtual environment with at least
Python 3.9, e.g. using `conda`:
```shell
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
```
2023-02-13 05:02:45 -07:00
You may also need to install Protoc.
On Linux:
```shell
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:
```shell
brew install protobuf
```
Then run:
2022-10-27 06:25:29 -06:00
2022-10-08 04:30:12 -06:00
```shell
BUILD_EXTENSIONS=True make install # Install repository and HF/transformer fork with CUDA kernels
2022-10-18 07:19:03 -06:00
make run-bloom-560m
2022-10-08 04:30:12 -06:00
```
2023-02-08 09:53:33 -07:00
**Note:** on some machines, you may also need the OpenSSL libraries and gcc. On Linux machines, run:
```shell
2023-02-08 09:53:33 -07:00
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
2022-10-27 06:25:29 -06:00
2023-02-14 05:02:16 -07:00
It is advised to download the weights ahead of time with the following command:
2022-10-27 06:25:29 -06:00
```shell
make download-bloom
```
### Run
2022-10-27 06:25:29 -06:00
```shell
make run-bloom # Requires 8xA100 80GB
```
### Quantization
2022-10-27 06:25:29 -06:00
You can also quantize the weights with bitsandbytes to reduce the VRAM requirement:
```shell
make run-bloom-quantize # Requires 8xA100 40GB
```
## Develop
2022-10-18 07:19:03 -06:00
2022-10-08 04:30:12 -06:00
```shell
make server-dev
make router-dev
2022-10-08 04:30:12 -06:00
```
## Testing
```shell
make python-tests
make integration-tests
```