Making TGI deployment optimal # Text Generation Inference GitHub Repo stars Swagger API documentation A Rust, Python and gRPC server for text generation inference. Used in production at [HuggingFace](https://huggingface.co) to power Hugging Chat, the Inference API and Inference Endpoint.
## Table of contents - [Get Started](#get-started) - [API Documentation](#api-documentation) - [Using a private or gated model](#using-a-private-or-gated-model) - [A note on Shared Memory](#a-note-on-shared-memory-shm) - [Distributed Tracing](#distributed-tracing) - [Local Install](#local-install) - [CUDA Kernels](#cuda-kernels) - [Optimized architectures](#optimized-architectures) - [Run Falcon](#run-falcon) - [Run](#run) - [Quantization](#quantization) - [Develop](#develop) - [Testing](#testing) Text Generation Inference (TGI) is a toolkit for deploying and serving Large Language Models (LLMs). TGI enables high-performance text generation for the most popular open-source LLMs, including Llama, Falcon, StarCoder, BLOOM, GPT-NeoX, and [more](https://huggingface.co/docs/text-generation-inference/supported_models). TGI implements many features, such as: - Simple launcher to serve most popular LLMs - Production ready (distributed tracing with Open Telemetry, Prometheus metrics) - 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](https://github.com/HazyResearch/flash-attention) and [Paged Attention](https://github.com/vllm-project/vllm) on the most popular architectures - Quantization with [bitsandbytes](https://github.com/TimDettmers/bitsandbytes) and [GPT-Q](https://arxiv.org/abs/2210.17323) - [Safetensors](https://github.com/huggingface/safetensors) weight loading - 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)) - Stop sequences - Log probabilities - Custom Prompt Generation: Easily generate text by providing custom prompts to guide the model's output - Fine-tuning Support: Utilize fine-tuned models for specific tasks to achieve higher accuracy and performance ## Get Started ### Docker For a detailed starting guide, please see the [Quick Tour](https://huggingface.co/docs/text-generation-inference/quicktour). The easiest way of getting started is using the official Docker container: ```shell model=HuggingFaceH4/zephyr-7b-beta 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:1.2 --model-id $model ``` And then you can make requests like ```bash curl 127.0.0.1:8080/generate \ -X POST \ -d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":20}}' \ -H 'Content-Type: application/json' ``` **Note:** To use NVIDIA 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. For running the Docker container on a machine with no GPUs or CUDA support, it is enough to remove the `--gpus all` flag and add `--disable-custom-kernels`, please note CPU is not the intended platform for this project, so performance might be subpar. **Note:** TGI supports AMD Instinct MI210 and MI250 GPUs. Details can be found in the [Supported Hardware documentation](https://huggingface.co/docs/text-generation-inference/supported_models#supported-hardware). To use AMD GPUs, please use `docker run --device /dev/kfd --device /dev/dri --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:1.2-rocm --model-id $model` instead of the command above. To see all options to serve your models (in the [code](https://github.com/huggingface/text-generation-inference/blob/main/launcher/src/main.rs) or in the cli): ``` text-generation-launcher --help ``` ### API documentation You can consult the OpenAPI documentation of the `text-generation-inference` REST API using the `/docs` route. The Swagger UI is also available at: [https://huggingface.github.io/text-generation-inference](https://huggingface.github.io/text-generation-inference). ### Using a private or gated model You have the option to utilize the `HUGGING_FACE_HUB_TOKEN` environment variable for configuring the token employed by `text-generation-inference`. This allows you to gain access to protected resources. For example, if you want to serve the gated Llama V2 model variants: 1. Go to https://huggingface.co/settings/tokens 2. Copy your cli READ token 3. Export `HUGGING_FACE_HUB_TOKEN=` or with Docker: ```shell model=meta-llama/Llama-2-7b-chat-hf volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run token= docker run --gpus all --shm-size 1g -e HUGGING_FACE_HUB_TOKEN=$token -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:1.2 --model-id $model ``` ### A note on Shared Memory (shm) [`NCCL`](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/index.html) is a communication framework used by `PyTorch` to do distributed training/inference. `text-generation-inference` make use of `NCCL` to enable Tensor Parallelism to dramatically speed up inference for large language models. In order to share data between the different devices of a `NCCL` group, `NCCL` might fall back to using the host memory if peer-to-peer using NVLink or PCI is not possible. To allow the container to use 1G of Shared Memory and support SHM sharing, we add `--shm-size 1g` on the above command. If you are running `text-generation-inference` inside `Kubernetes`. You can also add Shared Memory to the container by creating a volume with: ```yaml - name: shm emptyDir: medium: Memory sizeLimit: 1Gi ``` and mounting it to `/dev/shm`. Finally, you can also disable SHM sharing by using the `NCCL_SHM_DISABLE=1` environment variable. However, note that this will impact performance. ### Distributed Tracing `text-generation-inference` is instrumented with distributed tracing using OpenTelemetry. You can use this feature by setting the address to an OTLP collector with the `--otlp-endpoint` argument. ### Architecture ![TGI architecture](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/TGI.png) ### Local install You can also opt to install `text-generation-inference` locally. First [install Rust](https://rustup.rs/) and create a Python virtual environment with at least Python 3.9, e.g. using `conda`: ```shell curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh conda create -n text-generation-inference python=3.9 conda activate text-generation-inference ``` You may also need to install Protoc. On Linux: ```shell PROTOC_ZIP=protoc-21.12-linux-x86_64.zip curl -OL https://github.com/protocolbuffers/protobuf/releases/download/v21.12/$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: ```shell BUILD_EXTENSIONS=True make install # Install repository and HF/transformer fork with CUDA kernels make run-falcon-7b-instruct ``` **Note:** on some machines, you may also need the OpenSSL libraries and gcc. On Linux machines, run: ```shell sudo apt-get install libssl-dev gcc -y ``` ### CUDA Kernels The custom CUDA kernels are only tested on NVIDIA A100, AMD MI210 and AMD MI250. If you have any installation or runtime issues, you can remove the kernels by using the `DISABLE_CUSTOM_KERNELS=True` environment variable. Be aware that the official Docker image has them enabled by default. ## Optimized architectures TGI works out of the box to serve optimized models in [this list](https://huggingface.co/docs/text-generation-inference/supported_models). Other architectures are supported on a best-effort basis using: `AutoModelForCausalLM.from_pretrained(, device_map="auto")` or `AutoModelForSeq2SeqLM.from_pretrained(, device_map="auto")` ## Run Falcon ### Run ```shell make run-falcon-7b-instruct ``` ### Quantization You can also quantize the weights with bitsandbytes to reduce the VRAM requirement: ```shell make run-falcon-7b-instruct-quantize ``` 4bit quantization is available using the [NF4 and FP4 data types from bitsandbytes](https://arxiv.org/pdf/2305.14314.pdf). It can be enabled by providing `--quantize bitsandbytes-nf4` or `--quantize bitsandbytes-fp4` as a command line argument to `text-generation-launcher`. ## Develop ```shell make server-dev make router-dev ``` ## Testing ```shell # 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 ```