text-generation-inference/main/en/installation_amd.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Using TGI with AMD GPUs&quot;,&quot;local&quot;:&quot;using-tgi-with-amd-gpus&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;TunableOp&quot;,&quot;local&quot;:&quot;tunableop&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Flash attention implementation&quot;,&quot;local&quot;:&quot;flash-attention-implementation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Unsupported features&quot;,&quot;local&quot;:&quot;unsupported-features&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
		<link href="/docs/text-generation-inference/main/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/entry/start.1810066f.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/scheduler.362310b7.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/singletons.fa2b0eb7.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/index.7f53ec41.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/paths.284aef40.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/entry/app.8cfc1931.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/index.57dfc70d.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/nodes/0.543c9bd9.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/nodes/25.cb494dda.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/CodeBlock.d3c47f83.js">
		<link rel="modulepreload" href="/docs/text-generation-inference/main/en/_app/immutable/chunks/EditOnGithub.9633c464.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Using TGI with AMD GPUs&quot;,&quot;local&quot;:&quot;using-tgi-with-amd-gpus&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;TunableOp&quot;,&quot;local&quot;:&quot;tunableop&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Flash attention implementation&quot;,&quot;local&quot;:&quot;flash-attention-implementation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Unsupported features&quot;,&quot;local&quot;:&quot;unsupported-features&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END -->      <p></p>  <h1 class="relative group"><a id="using-tgi-with-amd-gpus" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#using-tgi-with-amd-gpus"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Using TGI with AMD GPUs</span></h1> <p data-svelte-h="svelte-1oiskdb">TGI is supported and tested on <a href="https://www.amd.com/en/products/accelerators/instinct/mi200/mi210.html" rel="nofollow">AMD Instinct MI210</a>, <a href="https://www.amd.com/en/products/accelerators/instinct/mi200/mi250.html" rel="nofollow">MI250</a> and <a href="https://www.amd.com/en/products/accelerators/instinct/mi300.html" rel="nofollow">MI300</a> GPUs. The support may be extended in the future. The recommended usage is through Docker. Make sure to check the <a href="https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/docker.html" rel="nofollow">AMD documentation</a> on how to use Docker with AMD GPUs.</p> <p data-svelte-h="svelte-1culvup">On a server powered by AMD GPUs, TGI can be launched with the following command:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->model=teknium/OpenHermes-2.5-Mistral-7B
volume=<span class="hljs-variable">$PWD</span>/data <span class="hljs-comment"># share a volume with the Docker container to avoid downloading weights every run</span>

docker run --<span class="hljs-built_in">rm</span> -it --cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
    --device=/dev/kfd --device=/dev/dri --group-add video \
    --ipc=host --shm-size 256g --net host -v <span class="hljs-variable">$volume</span>:/data \
    ghcr.io/huggingface/text-generation-inference:2.2.0-rocm \
    --model-id <span class="hljs-variable">$model</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-uo4xf6">The launched TGI server can then be queried from clients, make sure to check out the <a href="./basic_tutorials/consuming_tgi">Consuming TGI</a> guide.</p>  <h2 class="relative group"><a id="tunableop" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#tunableop"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>TunableOp</span></h2> <p data-svelte-h="svelte-9xgviz">TGI’s docker image for AMD GPUs integrates <a href="https://github.com/pytorch/pytorch/tree/main/aten/src/ATen/cuda/tunable" rel="nofollow">PyTorch’s TunableOp</a>, which allows to do an additional warmup to select the best performing matrix multiplication (GEMM) kernel from rocBLAS or hipBLASLt.</p> <p data-svelte-h="svelte-fxh8t6">Experimentally, on MI300X, we noticed a 6-8% latency improvement when using TunableOp on top of ROCm 6.1 and PyTorch 2.3.</p> <p data-svelte-h="svelte-1pxq52n">TunableOp is enabled by default, the warmup may take 1-2 minutes. In case you would like to disable TunableOp, please pass <code>--env PYTORCH_TUNABLEOP_ENABLED=&quot;0&quot;</code> when launcher TGI’s docker container.</p>  <h2 class="relative group"><a id="flash-attention-implementation" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#flash-attention-implementation"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Flash attention implementation</span></h2> <p data-svelte-h="svelte-vrgeg6">Two implementations of Flash Attention are available for ROCm, the first is <a href="https://github.com/ROCm/flash-attention" rel="nofollow">ROCm/flash-attention</a> based on a <a href="https://github.com/ROCm/composable_kernel" rel="nofollow">Composable Kernel</a> (CK) implementation, and the second is a <a href="https://github.com/huggingface/text-generation-inference/blob/main/server/text_generation_server/layers/attention/flash_attn_triton.py" rel="nofollow">Triton implementation</a>.</p> <p data-svelte-h="svelte-16up6ld">By default, the Composable Kernel implementation is used. However, the Triton implementation has slightly lower latency on MI250 and MI300, but requires a warmup which can be prohibitive as it needs to be done again for each new prompt length. If needed, FA Triton impelmentation can be enabled with <code>--env ROCM_USE_FLASH_ATTN_V2_TRITON=&quot;0&quot;</code> when launching TGI’s docker container.</p>  <h2 class="relative group"><a id="unsupported-features" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#unsupported-features"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Unsupported features</span></h2> <p data-svelte-h="svelte-1ym5nk6">The following features are currently not supported in the ROCm version of TGI, and the supported may be extended in the future:</p> <ul data-svelte-h="svelte-xh8ses"><li>Loading <a href="https://huggingface.co/docs/transformers/quantization#awq" rel="nofollow">AWQ</a> checkpoints.</li> <li>Kernel for sliding window attention (Mistral)</li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/text-generation-inference/blob/main/docs/source/installation_amd.md" target="_blank"><span data-svelte-h="svelte-1kd6by1">&lt;</span> <span data-svelte-h="svelte-x0xyl0">&gt;</span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a>  <p></p> 
			
			<script>
				{
					__sveltekit_1dfb6m4 = {
						assets: "/docs/text-generation-inference/main/en",
						base: "/docs/text-generation-inference/main/en",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/text-generation-inference/main/en/_app/immutable/entry/start.1810066f.js"),
						import("/docs/text-generation-inference/main/en/_app/immutable/entry/app.8cfc1931.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 25],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>