Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Training customization","local":"training-customization","sections":[{"title":"Use different optimizers and schedulers","local":"use-different-optimizers-and-schedulers","sections":[{"title":"Add a learning rate scheduler","local":"add-a-learning-rate-scheduler","sections":[],"depth":3}],"depth":2},{"title":"Memory efficient fine-tuning by sharing layers","local":"memory-efficient-fine-tuning-by-sharing-layers","sections":[],"depth":2},{"title":"Pass 8-bit reference models","local":"pass-8-bit-reference-models","sections":[],"depth":2},{"title":"Use the accelerator cache optimizer","local":"use-the-accelerator-cache-optimizer","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/trl/pr_4331/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/entry/start.6bbbc54b.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/scheduler.7b731bd4.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/singletons.55eb59f9.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/index.ac28c20f.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/paths.677b038d.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/entry/app.b003256e.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/preload-helper.71df5523.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/index.cc268345.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/nodes/0.c996cd3a.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/nodes/8.f5bd09e9.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.d403d039.js"> | |
| <link rel="modulepreload" href="/docs/trl/pr_4331/en/_app/immutable/chunks/CodeBlock.17bc4142.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Training customization","local":"training-customization","sections":[{"title":"Use different optimizers and schedulers","local":"use-different-optimizers-and-schedulers","sections":[{"title":"Add a learning rate scheduler","local":"add-a-learning-rate-scheduler","sections":[],"depth":3}],"depth":2},{"title":"Memory efficient fine-tuning by sharing layers","local":"memory-efficient-fine-tuning-by-sharing-layers","sections":[],"depth":2},{"title":"Pass 8-bit reference models","local":"pass-8-bit-reference-models","sections":[],"depth":2},{"title":"Use the accelerator cache optimizer","local":"use-the-accelerator-cache-optimizer","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 max-sm:gap-0.5 h-6 max-sm:h-5 px-2 max-sm:px-1.5 text-[11px] max-sm:text-[9px] font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0"><svg class="w-3 h-3 max-sm:w-2.5 max-sm:h-2.5" 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></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-6 max-sm:h-5 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible w-3 h-3 max-sm:w-2.5 max-sm:h-2.5 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg></button></div> </div> <h1 class="relative group"><a id="training-customization" 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="#training-customization"><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>Training customization</span></h1> <p data-svelte-h="svelte-9xca1p">TRL is designed with modularity in mind so that users are able to efficiently customize the training loop for their needs. Below are some examples on how you can apply and test different techniques. Note: Although these examples use the DPOTrainer, the customization applies to most (if not all) trainers.</p> <h2 class="relative group"><a id="use-different-optimizers-and-schedulers" 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="#use-different-optimizers-and-schedulers"><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>Use different optimizers and schedulers</span></h2> <p data-svelte-h="svelte-mc250">By default, the <code>DPOTrainer</code> creates a <code>torch.optim.AdamW</code> optimizer. You can create and define a different optimizer and pass it to <code>DPOTrainer</code> as follows:</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 --><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| <span class="hljs-keyword">from</span> torch <span class="hljs-keyword">import</span> optim | |
| <span class="hljs-keyword">from</span> trl <span class="hljs-keyword">import</span> DPOConfig, DPOTrainer | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| dataset = load_dataset(<span class="hljs-string">"trl-lib/ultrafeedback_binarized"</span>, split=<span class="hljs-string">"train"</span>) | |
| training_args = DPOConfig(output_dir=<span class="hljs-string">"Qwen2.5-0.5B-DPO"</span>) | |
| optimizer = optim.SGD(model.parameters(), lr=training_args.learning_rate) | |
| trainer = DPOTrainer( | |
| model=model, | |
| args=training_args, | |
| train_dataset=dataset, | |
| tokenizer=tokenizer, | |
| optimizers=(optimizer, <span class="hljs-literal">None</span>), | |
| ) | |
| trainer.train()<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="add-a-learning-rate-scheduler" 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="#add-a-learning-rate-scheduler"><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>Add a learning rate scheduler</span></h3> <p data-svelte-h="svelte-gsqx31">You can also play with your training by adding learning rate schedulers.</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 --><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| <span class="hljs-keyword">from</span> torch <span class="hljs-keyword">import</span> optim | |
| <span class="hljs-keyword">from</span> trl <span class="hljs-keyword">import</span> DPOConfig, DPOTrainer | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| dataset = load_dataset(<span class="hljs-string">"trl-lib/ultrafeedback_binarized"</span>, split=<span class="hljs-string">"train"</span>) | |
| training_args = DPOConfig(output_dir=<span class="hljs-string">"Qwen2.5-0.5B-DPO"</span>) | |
| optimizer = optim.AdamW(model.parameters(), lr=training_args.learning_rate) | |
| lr_scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=<span class="hljs-number">30</span>, gamma=<span class="hljs-number">0.1</span>) | |
| trainer = DPOTrainer( | |
| model=model, | |
| args=training_args, | |
| train_dataset=dataset, | |
| tokenizer=tokenizer, | |
| optimizers=(optimizer, lr_scheduler), | |
| ) | |
| trainer.train()<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="memory-efficient-fine-tuning-by-sharing-layers" 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="#memory-efficient-fine-tuning-by-sharing-layers"><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>Memory efficient fine-tuning by sharing layers</span></h2> <p data-svelte-h="svelte-hswwhh">Another tool you can use for more memory efficient fine-tuning is to share layers between the reference model and the model you want to train.</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 --><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer | |
| <span class="hljs-keyword">from</span> trl <span class="hljs-keyword">import</span> create_reference_model, DPOConfig, DPOTrainer | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| ref_model = create_reference_model(model, num_shared_layers=<span class="hljs-number">6</span>) | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| dataset = load_dataset(<span class="hljs-string">"trl-lib/ultrafeedback_binarized"</span>, split=<span class="hljs-string">"train[:1%]"</span>) | |
| training_args = DPOConfig(output_dir=<span class="hljs-string">"Qwen2.5-0.5B-DPO"</span>) | |
| trainer = DPOTrainer( | |
| model=model, | |
| ref_model=ref_model, | |
| args=training_args, | |
| train_dataset=dataset, | |
| tokenizer=tokenizer, | |
| ) | |
| trainer.train()<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="pass-8-bit-reference-models" 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="#pass-8-bit-reference-models"><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>Pass 8-bit reference models</span></h2> <p data-svelte-h="svelte-orby5n">Since <code>trl</code> supports all keyword arguments when loading a model from <code>transformers</code> using <code>from_pretrained</code>, you can also leverage <code>load_in_8bit</code> from <code>transformers</code> for more memory efficient fine-tuning.</p> <p data-svelte-h="svelte-9orwvr">Read more about 8-bit model loading in <code>transformers</code> <a href="https://huggingface.co/docs/transformers/en/peft#load-in-8bit-or-4bit" rel="nofollow">Load in 8bit or 4bit</a>.</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 --><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig | |
| <span class="hljs-keyword">from</span> trl <span class="hljs-keyword">import</span> DPOConfig, DPOTrainer | |
| model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| quantization_config = BitsAndBytesConfig(load_in_8bit=<span class="hljs-literal">True</span>) | |
| ref_model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>, quantization_config= quantization_config) | |
| tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"Qwen/Qwen2.5-0.5B-Instruct"</span>) | |
| dataset = load_dataset(<span class="hljs-string">"trl-lib/ultrafeedback_binarized"</span>, split=<span class="hljs-string">"train"</span>) | |
| training_args = DPOConfig(output_dir=<span class="hljs-string">"Qwen2.5-0.5B-DPO"</span>) | |
| trainer = DPOTrainer( | |
| model=model, | |
| ref_model=ref_model, | |
| args=training_args, | |
| train_dataset=dataset, | |
| tokenizer=tokenizer, | |
| ) | |
| trainer.train()<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="use-the-accelerator-cache-optimizer" 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="#use-the-accelerator-cache-optimizer"><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>Use the accelerator cache optimizer</span></h2> <p data-svelte-h="svelte-4o5ne7">When training large models, you should better handle the accelerator cache by iteratively clearing it. To do so, simply pass <code>optimize_device_cache=True</code> to <a href="/docs/trl/pr_4331/en/dpo_trainer#trl.DPOConfig">DPOConfig</a>:</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 -->training_args = DPOConfig(..., optimize_device_cache=<span class="hljs-literal">True</span>)<!-- HTML_TAG_END --></pre></div> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/trl/blob/main/docs/source/customization.md" target="_blank"><svg class="mr-1" 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="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg> <span data-svelte-h="svelte-zjs2n5"><span class="underline">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_l5mord = { | |
| assets: "/docs/trl/pr_4331/en", | |
| base: "/docs/trl/pr_4331/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/trl/pr_4331/en/_app/immutable/entry/start.6bbbc54b.js"), | |
| import("/docs/trl/pr_4331/en/_app/immutable/entry/app.b003256e.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 8], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 26.4 kB
- Xet hash:
- 056773b65baa0b82c8d7b4c3dd12e077cde7f725dc40a778aa6bfe343e4b9402
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.