trl/pr_5607/en/_app/immutable/nodes/53.7a815ae7.js

import{s as Ct,n as Ut,o as jt}from"../chunks/scheduler.7b731bd4.js";import{S as It,i as $t,e as r,s as l,c,h as St,a as s,d as a,b as o,f as K,g as p,j as d,k as ee,l as i,m as n,n as _,t as u,o as f,p as v}from"../chunks/index.cc268345.js";import{C as Dt,H as te,E as Pt}from"../chunks/MermaidChart.svelte_svelte_type_style_lang.f0d99f98.js";import{D as ue}from"../chunks/Docstring.03f7b462.js";import{C as kt}from"../chunks/CodeBlock.169a125f.js";function Ft(lt){let y,fe,pe,ve,w,he,J,ge,N,ot='Self-Distillation Policy Optimization (SDPO) was introduced in <a href="https://huggingface.co/papers/2601.20802" rel="nofollow">Reinforcement Learning via Self-Distillation</a> by <a href="https://huggingface.co/jonhue" rel="nofollow">Jonas Hübotter</a>, Frederike Lübeck, Lejs Behric, <a href="https://huggingface.co/antonbaumann" rel="nofollow">Anton Baumann</a>, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, and Andreas Krause.',be,x,rt="<p>Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math. Yet, current methods for reinforcement learning with verifiable rewards (RLVR) learn only from a scalar outcome reward per attempt, creating a severe credit-assignment bottleneck. Many verifiable environments actually provide rich textual feedback, such as runtime errors or judge evaluations, that explain why an attempt failed. We formalize this setting as reinforcement learning with rich feedback and introduce Self-Distillation Policy Optimization (SDPO), which converts tokenized feedback into a dense learning signal without any external teacher or explicit reward model. SDPO treats the current model conditioned on feedback as a self-teacher and distills its feedback-informed next-token predictions back into the policy. In this way, SDPO leverages the model’s ability to retrospectively identify its own mistakes in-context. Across scientific reasoning, tool use, and competitive programming on LiveCodeBench v6, SDPO improves sample efficiency and final accuracy over strong RLVR baselines. Notably, SDPO also outperforms baselines in standard RLVR environments that only return scalar feedback by using successful rollouts as implicit feedback for failed attempts. Finally, applying SDPO to individual questions at test time accelerates discovery on difficult binary-reward tasks, achieving the same discovery probability as best-of-k sampling or multi-turn conversations with 3x fewer attempts.</p>",ye,k,st="The SDPO trainer is built on TRL’s experimental shared self-distillation stack. It keeps the online rollout-and-reward training flow, then builds a teacher-conditioned view of the same completions from successful rollouts and optional environment feedback.",Me,C,it="In the current TRL implementation:",Te,U,dt="<li>the default SDPO policy loss mode is <code>distillation_only</code></li> <li><code>hybrid</code> mode is also available to combine the base policy loss with the self-distillation loss</li> <li>supported teacher regularization modes are <code>ema</code> and <code>none</code></li> <li><code>distillation_topk</code> is only valid when <code>full_logit_distillation=True</code></li> <li>when <code>full_logit_distillation=False</code>, SDPO uses token-level reverse KL and requires <code>distillation_alpha=1.0</code></li> <li>environment feedback can be injected into teacher reprompts when the dataset exposes a <code>privileged_context</code> column</li>",we,j,Je,I,mt="Each example must provide:",Ne,$,ct="<li><code>prompt</code>: the student-facing prompt</li> <li><code>privileged_context</code>: optional privileged text, such as environment feedback, used when <code>include_environment_feedback=True</code></li>",xe,S,ke,D,Ce,P,pt="SDPO always requires a <code>prompt</code> column. To use environment feedback, also include a <code>privileged_context</code> column and set <code>include_environment_feedback=True</code>. SDPO will use successful rollouts and, when enabled, that text to build teacher reprompts for self-distillation.",Ue,F,je,A,_t="The trainer emits a small set of callback hooks that are useful for debugging, observability, and tests. These hooks are intended as practical integration points for experimental self-distillation workflows.",Ie,O,ut="Shared self-distillation hooks:",$e,B,ft="<li><code>on_self_distillation_batch_prepared</code>: fired when a self-distillation batch is ready. The payload includes <code>prompt_ids</code>, <code>completion_ids</code>, and <code>old_per_token_logps</code> when importance-sampling clipping inputs are available.</li> <li><code>on_generation_batch_built</code>: fired when a new buffered generation batch is created. The payload includes <code>generate_every</code> and <code>steps_per_generation</code>.</li>",Se,L,vt="SDPO-specific hook:",De,R,ht="<li><code>on_teacher_context_built</code>: fired after SDPO constructs the teacher-conditioned inputs. The payload includes <code>teacher_input_ids</code>, <code>teacher_attention_mask</code>, <code>completion_mask</code>, and <code>self_distillation_mask</code>.</li>",Pe,Z,Fe,V,gt='Use <a href="https://github.com/huggingface/trl/blob/main/trl/experimental/sdpo/sdpo.py" rel="nofollow"><code>trl/experimental/sdpo/sdpo.py</code></a> to launch SDPO training from the command line. The script supports verifiable math rewards, environment feedback via <code>--feedback_column</code>, and PEFT/LoRA via the standard <code>ModelConfig</code> flags.',Ae,X,Oe,z,Be,g,W,Ee,ae,bt="Configuration class for the <code>SDPOTrainer</code>.",He,ne,yt=`This class extends <code>experimental.self_distillation.SelfDistillationConfig</code> with the online teacher-construction
parameters used by Self-Distillation Policy Optimization (SDPO).`,Le,E,Re,m,H,Qe,le,Mt="Trainer for Self-Distillation Policy Optimization (SDPO).",qe,oe,Tt=`SDPO augments on-policy optimization with self-distillation from the model’s own high-reward trajectories. It
converts tokenized feedback into a dense learning signal without any external teacher or explicit reward model.
SDPO treats the current model conditioned on feedback as a self-teacher and distills its feedback-informed
next-token predictions back into the policy.`,Ge,M,Q,Ye,re,wt="Main training entry point.",Ke,b,q,et,se,Jt="Will save the model, so you can reload it using <code>from_pretrained()</code>.",tt,ie,Nt="Will only save from the main process.",at,T,G,nt,de,xt="Upload <code>self.model</code> and <code>self.processing_class</code> to the 🤗 model hub on the repo <code>self.args.hub_model_id</code>.",Ze,Y,Ve,_e,Xe;return w=new Dt({props:{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"}}),J=new te({props:{title:"SDPO",local:"sdpo",headingTag:"h1"}}),j=new te({props:{title:"Expected dataset columns",local:"expected-dataset-columns",headingTag:"h2"}}),S=new te({props:{title:"Usage",local:"usage",headingTag:"h2"}}),D=new kt({props:{code:"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",highlighted:`<span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> Dataset

<span class="hljs-keyword">from</span> trl.experimental.sdpo <span class="hljs-keyword">import</span> SDPOConfig, SDPOTrainer

dataset = Dataset.from_dict(
    {
        <span class="hljs-string">&quot;prompt&quot;</span>: [[{<span class="hljs-string">&quot;role&quot;</span>: <span class="hljs-string">&quot;user&quot;</span>, <span class="hljs-string">&quot;content&quot;</span>: <span class="hljs-string">&quot;Solve 2+2.&quot;</span>}]],
        <span class="hljs-string">&quot;privileged_context&quot;</span>: [<span class="hljs-string">&quot;Your earlier answer used the wrong format.&quot;</span>],
    }
)

training_args = SDPOConfig(
    output_dir=<span class="hljs-string">&quot;sdpo-model&quot;</span>,
    distillation_topk=<span class="hljs-number">100</span>,                 <span class="hljs-comment"># Top-K logit distillation approximation</span>
    full_logit_distillation=<span class="hljs-literal">True</span>,          <span class="hljs-comment"># Required for top-K; enables non-reverse divergences</span>
    include_environment_feedback=<span class="hljs-literal">True</span>,     <span class="hljs-comment"># Use dataset privileged_context for teacher reprompts</span>
)

trainer = SDPOTrainer(
    model=<span class="hljs-string">&quot;Qwen/Qwen2.5-1.5B-Instruct&quot;</span>,
    reward_funcs=reward_func,
    args=training_args,
    train_dataset=dataset,
)
trainer.train()`,wrap:!1}}),F=new te({props:{title:"Callbacks",local:"callbacks",headingTag:"h2"}}),Z=new te({props:{title:"Example script",local:"example-script",headingTag:"h2"}}),X=new kt({props:{code:"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",highlighted:`python trl/experimental/sdpo/sdpo.py \\
    --model_name_or_path Qwen/Qwen2.5-Math-1.5B-Instruct \\
    --dataset_name openai/gsm8k \\
    --dataset_config main \\
    --output_dir outputs/sdpo-qwen35-2b-gsm8k \\
    --learning_rate 5e-5 \\
    --dtype bfloat16 \\
    --bf16 <span class="hljs-literal">true</span> \\
    --max_completion_length 128 \\
    --use_peft \\
    --lora_target_modules q_proj k_proj v_proj o_proj gate_proj up_proj down_proj \\
    --per_device_train_batch_size 1 \\
    --gradient_accumulation_steps 2 \\
    --num_generations 8 \\
    --generation_batch_size 32 \\
    --distillation_alpha 1.0 \\
    --full_logit_distillation <span class="hljs-literal">false</span> \\
    --sdpo_policy_loss_mode hybrid \\
    --report_to none \\
    --eval_strategy steps \\
    --eval_steps 1000 \\
    --save_strategy no`,wrap:!1}}),z=new te({props:{title:"SDPOConfig",local:"trl.experimental.sdpo.SDPOConfig",headingTag:"h2"}}),W=new ue({props:{name:"class trl.experimental.sdpo.SDPOConfig",anchor:"trl.experimental.sdpo.SDPOConfig",parameters:[{name:"output_dir",val:": str | None = None"},{name:"per_device_train_batch_size",val:": int = 8"},{name:"num_train_epochs",val:": float = 3.0"},{name:"max_steps",val:": int = -1"},{name:"learning_rate",val:": float = 5e-05"},{name:"lr_scheduler_type",val:": transformers.trainer_utils.SchedulerType | str = 'linear'"},{name:"lr_scheduler_kwargs",val:": dict | str | None = None"},{name:"warmup_steps",val:": float = 0"},{name:"optim",val:": transformers.training_args.OptimizerNames | str = 'adamw_torch_fused'"},{name:"optim_args",val:": str | None = None"},{name:"weight_decay",val:": float = 0.0"},{name:"adam_beta1",val:": float = 0.9"},{name:"adam_beta2",val:": float = 0.999"},{name:"adam_epsilon",val:": float = 1e-08"},{name:"optim_target_modules",val:": None | str | list[str] = None"},{name:"gradient_accumulation_steps",val:": int = 1"},{name:"average_tokens_across_devices",val:": bool = True"},{name:"max_grad_norm",val:": float = 1.0"},{name:"label_smoothing_factor",val:": float = 0.0"},{name:"bf16",val:": bool | None = None"},{name:"fp16",val:": bool = False"},{name:"bf16_full_eval",val:": bool = False"},{name:"fp16_full_eval",val:": bool = False"},{name:"tf32",val:": bool | None = None"},{name:"gradient_checkpointing",val:": bool = True"},{name:"gradient_checkpointing_kwargs",val:": dict[str, typing.Any] | str | None = None"},{name:"torch_compile",val:": bool = False"},{name:"torch_compile_backend",val:": str | None = None"},{name:"torch_compile_mode",val:": str | None = None"},{name:"use_liger_kernel",val:": bool = False"},{name:"liger_kernel_config",val:": dict[str, bool] | None = None"},{name:"use_cache",val:": bool = False"},{name:"neftune_noise_alpha",val:": float | None = None"},{name:"torch_empty_cache_steps",val:": int | None = None"},{name:"auto_find_batch_size",val:": bool = False"},{name:"logging_strategy",val:": transformers.trainer_utils.IntervalStrategy | str = 'steps'"},{name:"logging_steps",val:": float = 10"},{name:"logging_first_step",val:": bool = False"},{name:"log_on_each_node",val:": bool = True"},{name:"logging_nan_inf_filter",val:": bool = True"},{name:"include_num_input_tokens_seen",val:": str | bool = 'no'"},{name:"log_level",val:": str = 'passive'"},{name:"log_level_replica",val:": str = 'warning'"},{name:"disable_tqdm",val:": bool | None = None"},{name:"report_to",val:": None | str | list[str] = 'none'"},{name:"run_name",val:": str | None = None"},{name:"project",val:": str = 'huggingface'"},{name:"trackio_space_id",val:": str | None = 'trackio'"},{name:"eval_strategy",val:": transformers.trainer_utils.IntervalStrategy | str = 'no'"},{name:"eval_steps",val:": float | None = None"},{name:"eval_delay",val:": float = 0"},{name:"per_device_eval_batch_size",val:": int = 8"},{name:"prediction_loss_only",val:": bool = False"},{name:"eval_on_start",val:": bool = False"},{name:"eval_do_concat_batches",val:": bool = True"},{name:"eval_use_gather_object",val:": bool = False"},{name:"eval_accumulation_steps",val:": int | None = None"},{name:"include_for_metrics",val:": list = <factory>"},{name:"batch_eval_metrics",val:": bool = False"},{name:"save_only_model",val:": bool = False"},{name:"save_strategy",val:": transformers.trainer_utils.SaveStrategy | str = 'steps'"},{name:"save_steps",val:": float = 500"},{name:"save_on_each_node",val:": bool = False"},{name:"save_total_limit",val:": int | None = None"},{name:"enable_jit_checkpoint",val:": bool = False"},{name:"push_to_hub",val:": bool = False"},{name:"hub_token",val:": str | None = None"},{name:"hub_private_repo",val:": bool | None = None"},{name:"hub_model_id",val:": str | None = None"},{name:"hub_strategy",val:": transformers.trainer_utils.HubStrategy | str = 'every_save'"},{name:"hub_always_push",val:": bool = False"},{name:"hub_revision",val:": str | None = None"},{name:"load_best_model_at_end",val:": bool = False"},{name:"metric_for_best_model",val:": str | None = None"},{name:"greater_is_better",val:": bool | None = None"},{name:"ignore_data_skip",val:": bool = False"},{name:"restore_callback_states_from_checkpoint",val:": bool = False"},{name:"full_determinism",val:": bool = False"},{name:"seed",val:": int = 42"},{name:"data_seed",val:": int | None = None"},{name:"use_cpu",val:": bool = False"},{name:"accelerator_config",val:": dict | str | None = None"},{name:"parallelism_config",val:": accelerate.parallelism_config.ParallelismConfig | None = None"},{name:"dataloader_drop_last",val:": bool = False"},{name:"dataloader_num_workers",val:": int = 0"},{name:"dataloader_pin_memory",val:": bool = True"},{name:"dataloader_persistent_workers",val:": bool = False"},{name:"dataloader_prefetch_factor",val:": int | None = None"},{name:"remove_unused_columns",val:": bool = False"},{name:"label_names",val:": list[str] | None = None"},{name:"train_sampling_strategy",val:": str = 'random'"},{name:"length_column_name",val:": str = 'length'"},{name:"ddp_find_unused_parameters",val:": bool | None = None"},{name:"ddp_bucket_cap_mb",val:": int | None = None"},{name:"ddp_broadcast_buffers",val:": bool | None = None"},{name:"ddp_backend",val:": str | None = None"},{name:"ddp_timeout",val:": int = 1800"},{name:"fsdp",val:": list[transformers.trainer_utils.FSDPOption] | str | None = None"},{name:"fsdp_config",val:": dict[str, typing.Any] | str | None = None"},{name:"deepspeed",val:": dict | str | None = None"},{name:"debug",val:": str | list[transformers.debug_utils.DebugOption] = ''"},{name:"skip_memory_metrics",val:": bool = True"},{name:"do_train",val:": bool = False"},{name:"do_eval",val:": bool = False"},{name:"do_predict",val:": bool = False"},{name:"resume_from_checkpoint",val:": str | None = None"},{name:"warmup_ratio",val:": float | None = None"},{name:"logging_dir",val:": str | None = None"},{name:"local_rank",val:": int = -1"},{name:"model_init_kwargs",val:": dict[str, typing.Any] | None = None"},{name:"disable_dropout",val:": bool = False"},{name:"max_prompt_length",val:": int | None = 512"},{name:"num_generations",val:": int = 8"},{name:"num_generations_eval",val:": int | None = None"},{name:"max_completion_length",val:": int | None = 256"},{name:"ds3_gather_for_generation",val:": bool = True"},{name:"shuffle_dataset",val:": bool = True"},{name:"generation_batch_size",val:": int | None = None"},{name:"steps_per_generation",val:": int | None = None"},{name:"temperature",val:": float = 1.0"},{name:"top_p",val:": float = 1.0"},{name:"top_k",val:": int = 0"},{name:"min_p",val:": float | None = None"},{name:"generation_kwargs",val:": dict[str, typing.Any] | None = None"},{name:"chat_template_kwargs",val:": dict[str, typing.Any] | None = None"},{name:"repetition_penalty",val:": float = 1.0"},{name:"cache_implementation",val:": str | None = None"},{name:"use_vllm",val:": bool = False"},{name:"vllm_mode",val:": str = 'colocate'"},{name:"vllm_model_impl",val:": str = 'vllm'"},{name:"vllm_enable_sleep_mode",val:": bool = False"},{name:"vllm_server_base_url",val:": str | None = None"},{name:"vllm_server_host",val:": str = '0.0.0.0'"},{name:"vllm_server_port",val:": int = 8000"},{name:"vllm_group_port",val:": int = 51216"},{name:"vllm_server_timeout",val:": float = 240.0"},{name:"vllm_tensor_parallel_size",val:": int = 1"},{name:"vllm_gpu_memory_utilization",val:": float = 0.3"},{name:"vllm_max_model_length",val:": int | None = None"},{name:"beta",val:": float = 0.0"},{name:"num_iterations",val:": int = 1"},{name:"epsilon",val:": float = 0.2"},{name:"epsilon_high",val:": float | None = None"},{name:"importance_sampling_level",val:": str = 'token'"},{name:"reward_weights",val:": list[float] | None = None"},{name:"scale_rewards",val:": str | bool = 'group'"},{name:"loss_type",val:": str = 'dapo'"},{name:"mask_truncated_completions",val:": bool = False"},{name:"sync_ref_model",val:": bool = False"},{name:"ref_model_mixup_alpha",val:": float = 0.6"},{name:"ref_model_sync_steps",val:": int = 512"},{name:"top_entropy_quantile",val:": float = 1.0"},{name:"distillation_alpha",val:": float = 1.0"},{name:"distillation_topk",val:": int | None = None"},{name:"full_logit_distillation",val:": bool = False"},{name:"distillation_is_clip",val:": float | None = 2.0"},{name:"distillation_add_tail",val:": bool = False"},{name:"distillation_weight",val:": float = 1.0"},{name:"diagnostics_warning_interval",val:": int = 10"},{name:"diagnostics_flat_tolerance",val:": float = 1e-08"},{name:"dont_reprompt_on_self_success",val:": bool = True"},{name:"sdpo_policy_loss_mode",val:": str = 'distillation_only'"},{name:"teacher_regularization",val:": str = 'ema'"},{name:"teacher_update_rate",val:": float | None = None"},{name:"ema_update_rate",val:": float = 0.05"},{name:"max_reprompt_len",val:": int = 10240"},{name:"use_successful_as_teacher",val:": bool = True"},{name:"success_reward_threshold",val:": float = 1.0"},{name:"reprompt_template",val:": str = '{prompt}{solution}{feedback}\\n\\nCorrectly solve the original question.\\n'"},{name:"solution_template",val:": str = '\\nCorrect solution:\\n\\n{successful_previous_attempt}\\n\\n'"},{name:"feedback_template",val:": str = '\\nThe following is feedback from your unsuccessful earlier attempt:\\n\\n{feedback_raw}\\n\\n'"},{name:"include_environment_feedback",val:": bool = False"},{name:"environment_feedback_only_without_solution",val:": bool = False"},{name:"remove_thinking_from_demonstration",val:": bool = False"}],source:"https://github.com/huggingface/trl/blob/vr_5607/trl/experimental/sdpo/sdpo_config.py#L21",parameterGroups:[{title:"Parameters that control the SDPO loss",parametersDescription:[{anchor:"trl.experimental.sdpo.SDPOConfig.sdpo_policy_loss_mode",description:`<strong>sdpo_policy_loss_mode</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;distillation_only&quot;</code>) &#x2014;
How SDPO combines the online policy loss and self-distillation loss. Supported: <code>distillation_only</code>,
<code>hybrid</code>.`,name:"sdpo_policy_loss_mode"},{anchor:"trl.experimental.sdpo.SDPOConfig.distillation_alpha",description:`<strong>distillation_alpha</strong> (<code>float</code>, <em>optional</em>, defaults to <code>1.0</code>) &#x2014;
Divergence interpolation coefficient. Token-level SDPO requires the official reverse-KL setting
<code>distillation_alpha=1.0</code>.`,name:"distillation_alpha"},{anchor:"trl.experimental.sdpo.SDPOConfig.distillation_topk",description:`<strong>distillation_topk</strong> (<code>int</code> or <code>None</code>, <em>optional</em>) &#x2014;
Top-k approximation for logit-level SDPO. Requires <code>full_logit_distillation=True</code>.`,name:"distillation_topk"}]},{title:"Parameters that control the teacher",parametersDescription:[{anchor:"trl.experimental.sdpo.SDPOConfig.teacher_regularization",description:`<strong>teacher_regularization</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;ema&quot;</code>) &#x2014;
Teacher update strategy. Supported: <code>ema</code>, <code>none</code>.`,name:"teacher_regularization"},{anchor:"trl.experimental.sdpo.SDPOConfig.teacher_update_rate",description:`<strong>teacher_update_rate</strong> (<code>float</code> or <code>None</code>, <em>optional</em>) &#x2014;
EMA update rate used when <code>teacher_regularization=&quot;ema&quot;</code>.`,name:"teacher_update_rate"},{anchor:"trl.experimental.sdpo.SDPOConfig.ema_update_rate",description:`<strong>ema_update_rate</strong> (<code>float</code>, <em>optional</em>, defaults to <code>0.05</code>) &#x2014;
Deprecated alias for <code>teacher_update_rate</code>.`,name:"ema_update_rate"}]},{title:"Parameters that control reprompting",parametersDescription:[{anchor:"trl.experimental.sdpo.SDPOConfig.use_successful_as_teacher",description:`<strong>use_successful_as_teacher</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether successful rollouts are turned into teacher demonstrations.`,name:"use_successful_as_teacher"},{anchor:"trl.experimental.sdpo.SDPOConfig.success_reward_threshold",description:`<strong>success_reward_threshold</strong> (<code>float</code>, <em>optional</em>, defaults to <code>1.0</code>) &#x2014;
Minimum reward for a rollout to count as successful.`,name:"success_reward_threshold"},{anchor:"trl.experimental.sdpo.SDPOConfig.include_environment_feedback",description:`<strong>include_environment_feedback</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>False</code>) &#x2014;
Whether <code>privileged_context</code> is injected into teacher reprompts when available.`,name:"include_environment_feedback"}]}]}}),E=new te({props:{title:"SDPOTrainer",local:"trl.experimental.sdpo.SDPOTrainer",headingTag:"h2"}}),H=new ue({props:{name:"class trl.experimental.sdpo.SDPOTrainer",anchor:"trl.experimental.sdpo.SDPOTrainer",parameters:[{name:"model",val:": str | transformers.modeling_utils.PreTrainedModel | torch.nn.modules.module.Module"},{name:"reward_funcs",val:": typing.Union[typing.Any, list[typing.Any], NoneType] = None"},{name:"args",val:": trl.experimental.sdpo.sdpo_config.SDPOConfig | None = None"},{name:"train_dataset",val:": datasets.arrow_dataset.Dataset | datasets.iterable_dataset.IterableDataset | None = None"},{name:"eval_dataset",val:": datasets.arrow_dataset.Dataset | datasets.iterable_dataset.IterableDataset | dict[str, datasets.arrow_dataset.Dataset | datasets.iterable_dataset.IterableDataset] | None = None"},{name:"processing_class",val:": transformers.tokenization_utils_base.PreTrainedTokenizerBase | transformers.processing_utils.ProcessorMixin | None = None"},{name:"reward_processing_classes",val:": transformers.tokenization_utils_base.PreTrainedTokenizerBase | list[transformers.tokenization_utils_base.PreTrainedTokenizerBase] | None = None"},{name:"callbacks",val:": list[transformers.trainer_callback.TrainerCallback] | None = None"},{name:"optimizers",val:": tuple = (None, None)"},{name:"peft_config",val:" = None"}],source:"https://github.com/huggingface/trl/blob/vr_5607/trl/experimental/sdpo/sdpo_trainer.py#L237"}}),Q=new ue({props:{name:"train",anchor:"trl.experimental.sdpo.SDPOTrainer.train",parameters:[{name:"resume_from_checkpoint",val:": str | bool | None = None"},{name:"trial",val:": optuna.Trial | dict[str, Any] | None = None"},{name:"ignore_keys_for_eval",val:": list[str] | None = None"}],parametersDescription:[{anchor:"trl.experimental.sdpo.SDPOTrainer.train.resume_from_checkpoint",description:`<strong>resume_from_checkpoint</strong> (<code>str</code> or <code>bool</code>, <em>optional</em>) &#x2014;
If a <code>str</code>, local path to a saved checkpoint as saved by a previous instance of <code>Trainer</code>. If a
<code>bool</code> and equals <code>True</code>, load the last checkpoint in <em>args.output_dir</em> as saved by a previous instance
of <code>Trainer</code>. If present, training will resume from the model/optimizer/scheduler states loaded here.`,name:"resume_from_checkpoint"},{anchor:"trl.experimental.sdpo.SDPOTrainer.train.trial",description:`<strong>trial</strong> (<code>optuna.Trial</code> or <code>dict[str, Any]</code>, <em>optional</em>) &#x2014;
The trial run or the hyperparameter dictionary for hyperparameter search.`,name:"trial"},{anchor:"trl.experimental.sdpo.SDPOTrainer.train.ignore_keys_for_eval",description:`<strong>ignore_keys_for_eval</strong> (<code>list[str]</code>, <em>optional</em>) &#x2014;
A list of keys in the output of your model (if it is a dictionary) that should be ignored when
gathering predictions for evaluation during the training.`,name:"ignore_keys_for_eval"}],source:"https://github.com/huggingface/trl/blob/vr_5607/transformers/trainer.py#L1323",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>Object containing the global step count, training loss, and metrics.</p>
`,returnType:`<script context="module">export const metadata = 'undefined';<\/script>


<p><code>~trainer_utils.TrainOutput</code></p>
`}}),q=new ue({props:{name:"save_model",anchor:"trl.experimental.sdpo.SDPOTrainer.save_model",parameters:[{name:"output_dir",val:": str | None = None"},{name:"_internal_call",val:": bool = False"}],source:"https://github.com/huggingface/trl/blob/vr_5607/transformers/trainer.py#L3746"}}),G=new ue({props:{name:"push_to_hub",anchor:"trl.experimental.sdpo.SDPOTrainer.push_to_hub",parameters:[{name:"commit_message",val:": str | None = 'End of training'"},{name:"blocking",val:": bool = True"},{name:"token",val:": str | None = None"},{name:"revision",val:": str | None = None"},{name:"**kwargs",val:""}],parametersDescription:[{anchor:"trl.experimental.sdpo.SDPOTrainer.push_to_hub.commit_message",description:`<strong>commit_message</strong> (<code>str</code>, <em>optional</em>, defaults to <code>&quot;End of training&quot;</code>) &#x2014;
Message to commit while pushing.`,name:"commit_message"},{anchor:"trl.experimental.sdpo.SDPOTrainer.push_to_hub.blocking",description:`<strong>blocking</strong> (<code>bool</code>, <em>optional</em>, defaults to <code>True</code>) &#x2014;
Whether the function should return only when the <code>git push</code> has finished.`,name:"blocking"},{anchor:"trl.experimental.sdpo.SDPOTrainer.push_to_hub.token",description:`<strong>token</strong> (<code>str</code>, <em>optional</em>, defaults to <code>None</code>) &#x2014;
Token with write permission to overwrite Trainer&#x2019;s original args.`,name:"token"},{anchor:"trl.experimental.sdpo.SDPOTrainer.push_to_hub.revision",description:`<strong>revision</strong> (<code>str</code>, <em>optional</em>) &#x2014;
The git revision to commit from. Defaults to the head of the &#x201C;main&#x201D; branch.`,name:"revision"},{anchor:"trl.experimental.sdpo.SDPOTrainer.push_to_hub.kwargs",description:`<strong>kwargs</strong> (<code>dict[str, Any]</code>, <em>optional</em>) &#x2014;
Additional keyword arguments passed along to <code>~Trainer.create_model_card</code>.`,name:"kwargs"}],source:"https://github.com/huggingface/trl/blob/vr_5607/transformers/trainer.py#L3993",returnDescription:`<script context="module">export const metadata = 'undefined';<\/script>


<p>The URL of the repository where the model was pushed if <code>blocking=False</code>, or a <code>Future</code> object tracking the
progress of the commit if <code>blocking=True</code>.</p>
`}}),Y=new Pt({props:{source:"https://github.com/huggingface/trl/blob/main/docs/source/sdpo_trainer.md"}}),{c(){y=r("meta"),fe=l(),pe=r("p"),ve=l(),c(w.$$.fragment),he=l(),c(J.$$.fragment),ge=l(),N=r("p"),N.innerHTML=ot,be=l(),x=r("blockquote"),x.innerHTML=rt,ye=l(),k=r("p"),k.textContent=st,Me=l(),C=r("p"),C.textContent=it,Te=l(),U=r("ul"),U.innerHTML=dt,we=l(),c(j.$$.fragment),Je=l(),I=r("p"),I.textContent=mt,Ne=l(),$=r("ul"),$.innerHTML=ct,xe=l(),c(S.$$.fragment),ke=l(),c(D.$$.fragment),Ce=l(),P=r("p"),P.innerHTML=pt,Ue=l(),c(F.$$.fragment),je=l(),A=r("p"),A.textContent=_t,Ie=l(),O=r("p"),O.textContent=ut,$e=l(),B=r("ul"),B.innerHTML=ft,Se=l(),L=r("p"),L.textContent=vt,De=l(),R=r("ul"),R.innerHTML=ht,Pe=l(),c(Z.$$.fragment),Fe=l(),V=r("p"),V.innerHTML=gt,Ae=l(),c(X.$$.fragment),Oe=l(),c(z.$$.fragment),Be=l(),g=r("div"),c(W.$$.fragment),Ee=l(),ae=r("p"),ae.innerHTML=bt,He=l(),ne=r("p"),ne.innerHTML=yt,Le=l(),c(E.$$.fragment),Re=l(),m=r("div"),c(H.$$.fragment),Qe=l(),le=r("p"),le.textContent=Mt,qe=l(),oe=r("p"),oe.textContent=Tt,Ge=l(),M=r("div"),c(Q.$$.fragment),Ye=l(),re=r("p"),re.textContent=wt,Ke=l(),b=r("div"),c(q.$$.fragment),et=l(),se=r("p"),se.innerHTML=Jt,tt=l(),ie=r("p"),ie.textContent=Nt,at=l(),T=r("div"),c(G.$$.fragment),nt=l(),de=r("p"),de.innerHTML=xt,Ze=l(),c(Y.$$.fragment),Ve=l(),_e=r("p"),this.h()},l(e){const t=St("svelte-u9bgzb",document.head);y=s(t,"META",{name:!0,content:!0}),t.forEach(a),fe=o(e),pe=s(e,"P",{}),K(pe).forEach(a),ve=o(e),p(w.$$.fragment,e),he=o(e),p(J.$$.fragment,e),ge=o(e),N=s(e,"P",{"data-svelte-h":!0}),d(N)!=="svelte-j8kgiw"&&(N.innerHTML=ot),be=o(e),x=s(e,"BLOCKQUOTE",{"data-svelte-h":!0}),d(x)!=="svelte-s0n0bh"&&(x.innerHTML=rt),ye=o(e),k=s(e,"P",{"data-svelte-h":!0}),d(k)!=="svelte-rxp2xi"&&(k.textContent=st),Me=o(e),C=s(e,"P",{"data-svelte-h":!0}),d(C)!=="svelte-1k81tbl"&&(C.textContent=it),Te=o(e),U=s(e,"UL",{"data-svelte-h":!0}),d(U)!=="svelte-9r2xaf"&&(U.innerHTML=dt),we=o(e),p(j.$$.fragment,e),Je=o(e),I=s(e,"P",{"data-svelte-h":!0}),d(I)!=="svelte-wjd8dj"&&(I.textContent=mt),Ne=o(e),$=s(e,"UL",{"data-svelte-h":!0}),d($)!=="svelte-g3jtzk"&&($.innerHTML=ct),xe=o(e),p(S.$$.fragment,e),ke=o(e),p(D.$$.fragment,e),Ce=o(e),P=s(e,"P",{"data-svelte-h":!0}),d(P)!=="svelte-19yhene"&&(P.innerHTML=pt),Ue=o(e),p(F.$$.fragment,e),je=o(e),A=s(e,"P",{"data-svelte-h":!0}),d(A)!=="svelte-bwf7b5"&&(A.textContent=_t),Ie=o(e),O=s(e,"P",{"data-svelte-h":!0}),d(O)!=="svelte-1dzg5pm"&&(O.textContent=ut),$e=o(e),B=s(e,"UL",{"data-svelte-h":!0}),d(B)!=="svelte-qdul72"&&(B.innerHTML=ft),Se=o(e),L=s(e,"P",{"data-svelte-h":!0}),d(L)!=="svelte-f3f18i"&&(L.textContent=vt),De=o(e),R=s(e,"UL",{"data-svelte-h":!0}),d(R)!=="svelte-12xjvno"&&(R.innerHTML=ht),Pe=o(e),p(Z.$$.fragment,e),Fe=o(e),V=s(e,"P",{"data-svelte-h":!0}),d(V)!=="svelte-7pln8l"&&(V.innerHTML=gt),Ae=o(e),p(X.$$.fragment,e),Oe=o(e),p(z.$$.fragment,e),Be=o(e),g=s(e,"DIV",{class:!0});var me=K(g);p(W.$$.fragment,me),Ee=o(me),ae=s(me,"P",{"data-svelte-h":!0}),d(ae)!=="svelte-1grzie"&&(ae.innerHTML=bt),He=o(me),ne=s(me,"P",{"data-svelte-h":!0}),d(ne)!=="svelte-1ls7sng"&&(ne.innerHTML=yt),me.forEach(a),Le=o(e),p(E.$$.fragment,e),Re=o(e),m=s(e,"DIV",{class:!0});var h=K(m);p(H.$$.fragment,h),Qe=o(h),le=s(h,"P",{"data-svelte-h":!0}),d(le)!=="svelte-19t2129"&&(le.textContent=Mt),qe=o(h),oe=s(h,"P",{"data-svelte-h":!0}),d(oe)!=="svelte-1m9q4x4"&&(oe.textContent=Tt),Ge=o(h),M=s(h,"DIV",{class:!0});var ze=K(M);p(Q.$$.fragment,ze),Ye=o(ze),re=s(ze,"P",{"data-svelte-h":!0}),d(re)!=="svelte-1cilnet"&&(re.textContent=wt),ze.forEach(a),Ke=o(h),b=s(h,"DIV",{class:!0});var ce=K(b);p(q.$$.fragment,ce),et=o(ce),se=s(ce,"P",{"data-svelte-h":!0}),d(se)!=="svelte-r8h4ov"&&(se.innerHTML=Jt),tt=o(ce),ie=s(ce,"P",{"data-svelte-h":!0}),d(ie)!=="svelte-1e6bius"&&(ie.textContent=Nt),ce.forEach(a),at=o(h),T=s(h,"DIV",{class:!0});var We=K(T);p(G.$$.fragment,We),nt=o(We),de=s(We,"P",{"data-svelte-h":!0}),d(de)!=="svelte-8tudwd"&&(de.innerHTML=xt),We.forEach(a),h.forEach(a),Ze=o(e),p(Y.$$.fragment,e),Ve=o(e),_e=s(e,"P",{}),K(_e).forEach(a),this.h()},h(){ee(y,"name","hf:doc:metadata"),ee(y,"content",At),ee(g,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),ee(M,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),ee(b,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),ee(T,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"),ee(m,"class","docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8")},m(e,t){i(document.head,y),n(e,fe,t),n(e,pe,t),n(e,ve,t),_(w,e,t),n(e,he,t),_(J,e,t),n(e,ge,t),n(e,N,t),n(e,be,t),n(e,x,t),n(e,ye,t),n(e,k,t),n(e,Me,t),n(e,C,t),n(e,Te,t),n(e,U,t),n(e,we,t),_(j,e,t),n(e,Je,t),n(e,I,t),n(e,Ne,t),n(e,$,t),n(e,xe,t),_(S,e,t),n(e,ke,t),_(D,e,t),n(e,Ce,t),n(e,P,t),n(e,Ue,t),_(F,e,t),n(e,je,t),n(e,A,t),n(e,Ie,t),n(e,O,t),n(e,$e,t),n(e,B,t),n(e,Se,t),n(e,L,t),n(e,De,t),n(e,R,t),n(e,Pe,t),_(Z,e,t),n(e,Fe,t),n(e,V,t),n(e,Ae,t),_(X,e,t),n(e,Oe,t),_(z,e,t),n(e,Be,t),n(e,g,t),_(W,g,null),i(g,Ee),i(g,ae),i(g,He),i(g,ne),n(e,Le,t),_(E,e,t),n(e,Re,t),n(e,m,t),_(H,m,null),i(m,Qe),i(m,le),i(m,qe),i(m,oe),i(m,Ge),i(m,M),_(Q,M,null),i(M,Ye),i(M,re),i(m,Ke),i(m,b),_(q,b,null),i(b,et),i(b,se),i(b,tt),i(b,ie),i(m,at),i(m,T),_(G,T,null),i(T,nt),i(T,de),n(e,Ze,t),_(Y,e,t),n(e,Ve,t),n(e,_e,t),Xe=!0},p:Ut,i(e){Xe||(u(w.$$.fragment,e),u(J.$$.fragment,e),u(j.$$.fragment,e),u(S.$$.fragment,e),u(D.$$.fragment,e),u(F.$$.fragment,e),u(Z.$$.fragment,e),u(X.$$.fragment,e),u(z.$$.fragment,e),u(W.$$.fragment,e),u(E.$$.fragment,e),u(H.$$.fragment,e),u(Q.$$.fragment,e),u(q.$$.fragment,e),u(G.$$.fragment,e),u(Y.$$.fragment,e),Xe=!0)},o(e){f(w.$$.fragment,e),f(J.$$.fragment,e),f(j.$$.fragment,e),f(S.$$.fragment,e),f(D.$$.fragment,e),f(F.$$.fragment,e),f(Z.$$.fragment,e),f(X.$$.fragment,e),f(z.$$.fragment,e),f(W.$$.fragment,e),f(E.$$.fragment,e),f(H.$$.fragment,e),f(Q.$$.fragment,e),f(q.$$.fragment,e),f(G.$$.fragment,e),f(Y.$$.fragment,e),Xe=!1},d(e){e&&(a(fe),a(pe),a(ve),a(he),a(ge),a(N),a(be),a(x),a(ye),a(k),a(Me),a(C),a(Te),a(U),a(we),a(Je),a(I),a(Ne),a($),a(xe),a(ke),a(Ce),a(P),a(Ue),a(je),a(A),a(Ie),a(O),a($e),a(B),a(Se),a(L),a(De),a(R),a(Pe),a(Fe),a(V),a(Ae),a(Oe),a(Be),a(g),a(Le),a(Re),a(m),a(Ze),a(Ve),a(_e)),a(y),v(w,e),v(J,e),v(j,e),v(S,e),v(D,e),v(F,e),v(Z,e),v(X,e),v(z,e),v(W),v(E,e),v(H),v(Q),v(q),v(G),v(Y,e)}}}const At='{"title":"SDPO","local":"sdpo","sections":[{"title":"Expected dataset columns","local":"expected-dataset-columns","sections":[],"depth":2},{"title":"Usage","local":"usage","sections":[],"depth":2},{"title":"Callbacks","local":"callbacks","sections":[],"depth":2},{"title":"Example script","local":"example-script","sections":[],"depth":2},{"title":"SDPOConfig","local":"trl.experimental.sdpo.SDPOConfig","sections":[],"depth":2},{"title":"SDPOTrainer","local":"trl.experimental.sdpo.SDPOTrainer","sections":[],"depth":2}],"depth":1}';function Ot(lt){return jt(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Xt extends It{constructor(y){super(),$t(this,y,Ot,Ft,Ct,{})}}export{Xt as component};