rhQFddlmZddlZddlmZmZddlmZmZddl m Z ddl m Z ddlmZddlmZmZej&eZGd d eZGd d eZeGd de Zy)) annotationsN) dataclassfield)CallableUnion)TrainingArguments) ParallelMode) ExplicitEnum)DefaultBatchSamplerMultiDatasetDefaultBatchSamplerceZdZdZdZdZdZy) BatchSamplersac Stores the acceptable string identifiers for batch samplers. The batch sampler is responsible for determining how samples are grouped into batches during training. Valid options are: - ``BatchSamplers.BATCH_SAMPLER``: **[default]** Uses :class:`~sentence_transformers.sampler.DefaultBatchSampler`, the default PyTorch batch sampler. - ``BatchSamplers.NO_DUPLICATES``: Uses :class:`~sentence_transformers.sampler.NoDuplicatesBatchSampler`, ensuring no duplicate samples in a batch. Recommended for losses that use in-batch negatives, such as: - :class:`~sentence_transformers.losses.MultipleNegativesRankingLoss` - :class:`~sentence_transformers.losses.CachedMultipleNegativesRankingLoss` - :class:`~sentence_transformers.losses.MultipleNegativesSymmetricRankingLoss` - :class:`~sentence_transformers.losses.CachedMultipleNegativesSymmetricRankingLoss` - :class:`~sentence_transformers.losses.MegaBatchMarginLoss` - :class:`~sentence_transformers.losses.GISTEmbedLoss` - :class:`~sentence_transformers.losses.CachedGISTEmbedLoss` - ``BatchSamplers.GROUP_BY_LABEL``: Uses :class:`~sentence_transformers.sampler.GroupByLabelBatchSampler`, ensuring that each batch has 2+ samples from the same label. Recommended for losses that require multiple samples from the same label, such as: - :class:`~sentence_transformers.losses.BatchAllTripletLoss` - :class:`~sentence_transformers.losses.BatchHardSoftMarginTripletLoss` - :class:`~sentence_transformers.losses.BatchHardTripletLoss` - :class:`~sentence_transformers.losses.BatchSemiHardTripletLoss` If you want to use a custom batch sampler, then you can subclass :class:`~sentence_transformers.sampler.DefaultBatchSampler` and pass the class (not an instance) to the ``batch_sampler`` argument in :class:`~sentence_transformers.training_args.SentenceTransformerTrainingArguments` (or :class:`~sentence_transformers.cross_encoder.training_args.CrossEncoderTrainingArguments`, etc.). Alternatively, you can pass a function that accepts ``dataset``, ``batch_size``, ``drop_last``, ``valid_label_columns``, ``generator``, and ``seed`` and returns a :class:`~sentence_transformers.sampler.DefaultBatchSampler` instance. Usage: :: from sentence_transformers import SentenceTransformer, SentenceTransformerTrainer, SentenceTransformerTrainingArguments from sentence_transformers.training_args import BatchSamplers from sentence_transformers.losses import MultipleNegativesRankingLoss from datasets import Dataset model = SentenceTransformer("microsoft/mpnet-base") train_dataset = Dataset.from_dict({ "anchor": ["It's nice weather outside today.", "He drove to work."], "positive": ["It's so sunny.", "He took the car to the office."], }) loss = MultipleNegativesRankingLoss(model) args = SentenceTransformerTrainingArguments( output_dir="checkpoints", batch_sampler=BatchSamplers.NO_DUPLICATES, ) trainer = SentenceTransformerTrainer( model=model, args=args, train_dataset=train_dataset, loss=loss, ) trainer.train() batch_sampler no_duplicatesgroup_by_labelN)__name__ __module__ __qualname____doc__ BATCH_SAMPLER NO_DUPLICATESGROUP_BY_LABELv/var/www/html/ai-insurance-compliance-backend/venv/lib/python3.12/site-packages/sentence_transformers/training_args.pyrrs<|$M#M%NrrceZdZdZdZdZy)MultiDatasetBatchSamplersa Stores the acceptable string identifiers for multi-dataset batch samplers. The multi-dataset batch sampler is responsible for determining in what order batches are sampled from multiple datasets during training. Valid options are: - ``MultiDatasetBatchSamplers.ROUND_ROBIN``: Uses :class:`~sentence_transformers.sampler.RoundRobinBatchSampler`, which uses round-robin sampling from each dataset until one is exhausted. With this strategy, it's likely that not all samples from each dataset are used, but each dataset is sampled from equally. - ``MultiDatasetBatchSamplers.PROPORTIONAL``: **[default]** Uses :class:`~sentence_transformers.sampler.ProportionalBatchSampler`, which samples from each dataset in proportion to its size. With this strategy, all samples from each dataset are used and larger datasets are sampled from more frequently. If you want to use a custom multi-dataset batch sampler, then you can subclass :class:`~sentence_transformers.sampler.MultiDatasetDefaultBatchSampler` and pass the class (not an instance) to the ``multi_dataset_batch_sampler`` argument in :class:`~sentence_transformers.training_args.SentenceTransformerTrainingArguments`. (or :class:`~sentence_transformers.cross_encoder.training_args.CrossEncoderTrainingArguments`, etc.). Alternatively, you can pass a function that accepts ``dataset`` (a :class:`~torch.utils.data.ConcatDataset`), ``batch_samplers`` (i.e. a list of batch sampler for each of the datasets in the :class:`~torch.utils.data.ConcatDataset`), ``generator``, and ``seed`` and returns a :class:`~sentence_transformers.sampler.MultiDatasetDefaultBatchSampler` instance. Usage: :: from sentence_transformers import SentenceTransformer, SentenceTransformerTrainer, SentenceTransformerTrainingArguments from sentence_transformers.training_args import MultiDatasetBatchSamplers from sentence_transformers.losses import CoSENTLoss from datasets import Dataset, DatasetDict model = SentenceTransformer("microsoft/mpnet-base") train_general = Dataset.from_dict({ "sentence_A": ["It's nice weather outside today.", "He drove to work."], "sentence_B": ["It's so sunny.", "He took the car to the bank."], "score": [0.9, 0.4], }) train_medical = Dataset.from_dict({ "sentence_A": ["The patient has a fever.", "The doctor prescribed medication.", "The patient is sweating."], "sentence_B": ["The patient feels hot.", "The medication was given to the patient.", "The patient is perspiring."], "score": [0.8, 0.6, 0.7], }) train_legal = Dataset.from_dict({ "sentence_A": ["This contract is legally binding.", "The parties agree to the terms and conditions."], "sentence_B": ["Both parties acknowledge their obligations.", "By signing this agreement, the parties enter into a legal relationship."], "score": [0.7, 0.8], }) train_dataset = DatasetDict({ "general": train_general, "medical": train_medical, "legal": train_legal, }) loss = CoSENTLoss(model) args = SentenceTransformerTrainingArguments( output_dir="checkpoints", multi_dataset_batch_sampler=MultiDatasetBatchSamplers.PROPORTIONAL, ) trainer = SentenceTransformerTrainer( model=model, args=args, train_dataset=train_dataset, loss=loss, ) trainer.train() round_robin proportionalN)rrrr ROUND_ROBIN PROPORTIONALrrrrrTs@D K!LrrceZdZUdZgdZedddiZded<eejdd iZ d ed <ee jdd iZ d ed<eeddiZded<eeddiZded<fdZfdZxZS)$SentenceTransformerTrainingArgumentsa SentenceTransformerTrainingArguments extends :class:`~transformers.TrainingArguments` with additional arguments specific to Sentence Transformers. See :class:`~transformers.TrainingArguments` for the complete list of available arguments. Args: output_dir (`str`): The output directory where the model checkpoints will be written. prompts (`Union[Dict[str, Dict[str, str]], Dict[str, str], str]`, *optional*): The prompts to use for each column in the training, evaluation and test datasets. Four formats are accepted: 1. `str`: A single prompt to use for all columns in the datasets, regardless of whether the training/evaluation/test datasets are :class:`datasets.Dataset` or a :class:`datasets.DatasetDict`. 2. `Dict[str, str]`: A dictionary mapping column names to prompts, regardless of whether the training/evaluation/test datasets are :class:`datasets.Dataset` or a :class:`datasets.DatasetDict`. 3. `Dict[str, str]`: A dictionary mapping dataset names to prompts. This should only be used if your training/evaluation/test datasets are a :class:`datasets.DatasetDict` or a dictionary of :class:`datasets.Dataset`. 4. `Dict[str, Dict[str, str]]`: A dictionary mapping dataset names to dictionaries mapping column names to prompts. This should only be used if your training/evaluation/test datasets are a :class:`datasets.DatasetDict` or a dictionary of :class:`datasets.Dataset`. batch_sampler (Union[:class:`~sentence_transformers.training_args.BatchSamplers`, `str`, :class:`~sentence_transformers.sampler.DefaultBatchSampler`, Callable[[...], :class:`~sentence_transformers.sampler.DefaultBatchSampler`]], *optional*): The batch sampler to use. See :class:`~sentence_transformers.training_args.BatchSamplers` for valid options. Defaults to ``BatchSamplers.BATCH_SAMPLER``. multi_dataset_batch_sampler (Union[:class:`~sentence_transformers.training_args.MultiDatasetBatchSamplers`, `str`, :class:`~sentence_transformers.sampler.MultiDatasetDefaultBatchSampler`, Callable[[...], :class:`~sentence_transformers.sampler.MultiDatasetDefaultBatchSampler`]], *optional*): The multi-dataset batch sampler to use. See :class:`~sentence_transformers.training_args.MultiDatasetBatchSamplers` for valid options. Defaults to ``MultiDatasetBatchSamplers.PROPORTIONAL``. router_mapping (`Dict[str, str] | Dict[str, Dict[str, str]]`, *optional*): A mapping of dataset column names to Router routes, like "query" or "document". This is used to specify which Router submodule to use for each dataset. Two formats are accepted: 1. `Dict[str, str]`: A mapping of column names to routes. 2. `Dict[str, Dict[str, str]]`: A mapping of dataset names to a mapping of column names to routes for multi-dataset training/evaluation. learning_rate_mapping (`Dict[str, float] | None`, *optional*): A mapping of parameter name regular expressions to learning rates. This allows you to set different learning rates for different parts of the model, e.g., `{'SparseStaticEmbedding\.*': 1e-3}` for the SparseStaticEmbedding module. This is useful when you want to fine-tune specific parts of the model with different learning rates. )accelerator_config fsdp_config deepspeedgradient_checkpointing_kwargslr_scheduler_kwargspromptsrouter_mappinglearning_rate_mappingNhelpzThe prompts to use for each column in the datasets. Either 1) a single string prompt, 2) a mapping of column names to prompts, 3) a mapping of dataset names to prompts, or 4) a mapping of dataset names to a mapping of column names to prompts.)defaultmetadataz;Union[str, None, dict[str, str], dict[str, dict[str, str]]]r)zThe batch sampler to use.zRUnion[BatchSamplers, str, DefaultBatchSampler, Callable[..., DefaultBatchSampler]]rz'The multi-dataset batch sampler to use.zvUnion[MultiDatasetBatchSamplers, str, MultiDatasetDefaultBatchSampler, Callable[..., MultiDatasetDefaultBatchSampler]]multi_dataset_batch_samplerzA mapping of dataset column names to Router routes, like "query" or "document". Either 1) a mapping of column names to routes or 2) a mapping of dataset names to a mapping of column names to routes for multi-dataset training/evaluation. )default_factoryr.r*zA mapping of parameter name regular expressions to learning rates. This allows you to set different learning rates for different parts of the model, e.g., {'SparseStaticEmbedding\.*': 1e-3} for the SparseStaticEmbedding module.z"Union[str, None, dict[str, float]]r+ct|t|jtrt |jn |j|_t|j trt|j n |j |_|j |jni|_t|jtr td|j |jni|_ t|jtr tdd|_ d|_ |jtjk(r&|j dk7rt"j%dyy|jtj&k(r9|j(s,|j dk7rt"j%dd|_yyy)NzThe `router_mapping` argument must be a dictionary mapping dataset column names to Router routes, like 'query' or 'document'. A stringified dictionary also works.zThe `learning_rate_mapping` argument must be a dictionary mapping parameter name regular expressions to learning rates. A stringified dictionary also works.TFunusedzCurrently using DataParallel (DP) for multi-gpu training, while DistributedDataParallel (DDP) is recommended for faster training. See https://sbert.net/docs/sentence_transformer/training/distributed.html for more information.zWhen using DistributedDataParallel (DDP), it is recommended to set `dataloader_drop_last=True` to avoid hanging issues with an uneven last batch. Setting `dataloader_drop_last=True`.)super __post_init__ isinstancerstrrr/rr* ValueErrorr+prediction_loss_onlyddp_broadcast_buffers parallel_moder NOT_DISTRIBUTED output_dirloggerwarning DISTRIBUTEDdataloader_drop_last)self __class__s rr4z2SentenceTransformerTrainingArguments.__post_init__s 2rUsl"("K3+^   8 $A&LA&HD" D"N Y"+HY" Y"r