# Copyright 2022 The HuggingFace Inc. team. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Generation configuration class and utilities.""" import copy import json import os from abc import ABC, abstractmethod from collections.abc import Callable from dataclasses import dataclass, is_dataclass from typing import TYPE_CHECKING, Any, Optional, Union from huggingface_hub import create_repo from .. import __version__ from ..utils import ( GENERATION_CONFIG_NAME, ExplicitEnum, PushToHubMixin, cached_file, extract_commit_hash, is_torch_available, logging, ) if TYPE_CHECKING: import torch from ..configuration_utils import PreTrainedConfig from ..modeling_utils import PreTrainedModel logger = logging.get_logger(__name__) METADATA_FIELDS = ("_from_model_config", "_commit_hash", "_original_object_hash", "transformers_version") STATIC_CACHE_IMPLEMENTATIONS = ("static", "offloaded_static") DYNAMIC_CACHE_IMPLEMENTATIONS = ("dynamic", "dynamic_full", "offloaded", "quantized") # All the following are redundant and deprecated, but kept for BC DEPRECATED_STATIC_CACHE_IMPLEMENTATIONS = ( "sliding_window", "hybrid", "hybrid_chunked", "offloaded_hybrid", "offloaded_hybrid_chunked", ) ALL_STATIC_CACHE_IMPLEMENTATIONS = STATIC_CACHE_IMPLEMENTATIONS + DEPRECATED_STATIC_CACHE_IMPLEMENTATIONS ALL_CACHE_IMPLEMENTATIONS = ALL_STATIC_CACHE_IMPLEMENTATIONS + DYNAMIC_CACHE_IMPLEMENTATIONS if is_torch_available(): from .logits_process import SynthIDTextWatermarkLogitsProcessor, WatermarkLogitsProcessor class GenerationMode(ExplicitEnum): """ Possible generation modes, downstream of the [`~generation.GenerationMixin.generate`] method. """ # Non-beam methods CONTRASTIVE_SEARCH = "contrastive_search" GREEDY_SEARCH = "greedy_search" SAMPLE = "sample" ASSISTED_GENERATION = "assisted_generation" DOLA_GENERATION = "dola_generation" # Beam methods BEAM_SEARCH = "beam_search" BEAM_SAMPLE = "beam_sample" CONSTRAINED_BEAM_SEARCH = "constrained_beam_search" GROUP_BEAM_SEARCH = "group_beam_search" class GenerationConfig(PushToHubMixin): # no-format """ Class that holds a configuration for a generation task. A `generate` call supports the following generation methods for text-decoder, text-to-text, speech-to-text, and vision-to-text models: - *greedy decoding* if `num_beams=1` and `do_sample=False` - *multinomial sampling* if `num_beams=1` and `do_sample=True` - *beam-search decoding* if `num_beams>1` and `do_sample=False` - *beam-search multinomial sampling* if `num_beams>1` and `do_sample=True` - *assisted decoding* if `assistant_model` or `prompt_lookup_num_tokens` is passed to `.generate()` To learn more about decoding strategies refer to the [text generation strategies guide](../generation_strategies). A large number of these flags control the logits or the stopping criteria of the generation. Make sure you check the [generate-related classes](https://huggingface.co/docs/transformers/internal/generation_utils) for a full description of the possible manipulations, as well as examples of their usage. Note: the configuration fields that are still `None` will be overridden by `GenerationConfig._get_default_generation_params()` during the generation loop. If you want to use different values for these fields, make sure to explicitly set them in the generation config. Args: > Parameters that control the length of the output max_length (`int`, *optional*): `max_new_tokens` is recommended for controlling how many tokens the model generates. `max_length` remains for backward compatibility. max_new_tokens (`int`, *optional*): The maximum numbers of tokens to generate, ignoring the number of tokens in the prompt. min_length (`int`, *optional*): The minimum length of the sequence to be generated. Corresponds to the length of the input prompt + `min_new_tokens`. Its effect is overridden by `min_new_tokens`, if also set. min_new_tokens (`int`, *optional*): The minimum numbers of tokens to generate, ignoring the number of tokens in the prompt. early_stopping (`bool` or `str`, *optional*): Controls the stopping condition for beam-based methods, like beam-search. It accepts the following values: `True`, where the generation stops as soon as there are `num_beams` complete candidates; `False`, where an heuristic is applied and the generation stops when is it very unlikely to find better candidates; `"never"`, where the beam search procedure only stops when there cannot be better candidates (canonical beam search algorithm). max_time (`float`, *optional*): The maximum amount of time you allow the computation to run for in seconds. generation will still finish the current pass after allocated time has been passed. stop_strings (`str or list[str]`, *optional*): A string or a list of strings that should terminate generation if the model outputs them. > Parameters that control the generation strategy used do_sample (`bool`): Whether or not to use sampling ; use greedy decoding otherwise. num_beams (`int`, *optional*): Number of beams for beam search. 1 means no beam search. > Parameters that control the cache use_cache (`bool`): Whether or not the model should use the past last key/values attentions (if applicable to the model) to speed up decoding. cache_implementation (`str`, *optional*): Name of the cache class that will be instantiated in `generate`, for faster decoding. Possible values are: - `"dynamic"`: [`DynamicCache`] - `"static"`: [`StaticCache`] - `"offloaded"`: [`DynamicCache(offloaded=True)`] - `"offloaded_static"`: [`StaticCache(offloaded=True)`] - `"quantized"`: [`QuantizedCache`] If none is specified, we will use the default cache for the model (which is often [`DynamicCache`]). See our [cache documentation](https://huggingface.co/docs/transformers/en/kv_cache) for further information. cache_config (`dict`, *optional*, default to `None`): Arguments used in the key-value cache class can be passed in `cache_config`. > Parameters for manipulation of the model output logits temperature (`float`, *optional*): The value used to module the next token probabilities. This value is set in a model's `generation_config.json` file. If it isn't set, the default value is 1.0 top_k (`int`, *optional*): The number of highest probability vocabulary tokens to keep for top-k-filtering. This value is set in a model's `generation_config.json` file. If it isn't set, the default value is 50. top_p (`float`, *optional*): If set to float < 1, only the smallest set of most probable tokens with probabilities that add up to `top_p` or higher are kept for generation. This value is set in a model's `generation_config.json` file. If it isn't set, the default value is 1.0 min_p (`float`, *optional*): Minimum token probability, which will be scaled by the probability of the most likely token. It must be a value between 0 and 1. Typical values are in the 0.01-0.2 range, comparably selective as setting `top_p` in the 0.99-0.8 range (use the opposite of normal `top_p` values). top_h (`float`, *optional*): Entropy budget scaling factor, which controls how much of the distribution’s entropy is preserved when sampling. Must be a value between 0 and 1. At each step, tokens are sorted by probability, and the smallest prefix of tokens is kept whose *renormalized* entropy is less than or equal to `top_h` times the entropy of the full distribution. Smaller values (e.g., 0.2–0.5) lead to more focused, deterministic outputs, while values closer to 1.0 allow more randomness and diversity. Typical values are in the 0.3–0.6 range. typical_p (`float`, *optional*): Local typicality measures how similar the conditional probability of predicting a target token next is to the expected conditional probability of predicting a random token next, given the partial text already generated. If set to float < 1, the smallest set of the most locally typical tokens with probabilities that add up to `typical_p` or higher are kept for generation. See [this paper](https://huggingface.co/papers/2202.00666) for more details. epsilon_cutoff (`float`, *optional*): If set to float strictly between 0 and 1, only tokens with a conditional probability greater than `epsilon_cutoff` will be sampled. In the paper, suggested values range from 3e-4 to 9e-4, depending on the size of the model. See [Truncation Sampling as Language Model Desmoothing](https://huggingface.co/papers/2210.15191) for more details. eta_cutoff (`float`, *optional*): Eta sampling is a hybrid of locally typical sampling and epsilon sampling. If set to float strictly between 0 and 1, a token is only considered if it is greater than either `eta_cutoff` or `sqrt(eta_cutoff) * exp(-entropy(softmax(next_token_logits)))`. The latter term is intuitively the expected next token probability, scaled by `sqrt(eta_cutoff)`. In the paper, suggested values range from 3e-4 to 2e-3, depending on the size of the model. See [Truncation Sampling as Language Model Desmoothing](https://huggingface.co/papers/2210.15191) for more details. repetition_penalty (`float`, *optional*): The parameter for repetition penalty. 1.0 means no penalty. See [this paper](https://huggingface.co/papers/1909.05858) for more details. encoder_repetition_penalty (`float`, *optional*): The parameter for encoder_repetition_penalty. An exponential penalty on sequences that are not in the original input. 1.0 means no penalty. length_penalty (`float`, *optional*): Exponential penalty to the length that is used with beam-based generation. It is applied as an exponent to the sequence length, which in turn is used to divide the score of the sequence. Since the score is the log likelihood of the sequence (i.e. negative), `length_penalty` > 0.0 promotes longer sequences, while `length_penalty` < 0.0 encourages shorter sequences. no_repeat_ngram_size (`int`, *optional*): If set to int > 0, all ngrams of that size can only occur once. bad_words_ids (`list[list[int]]`, *optional*): List of list of token ids that are not allowed to be generated. Check [`~generation.NoBadWordsLogitsProcessor`] for further documentation and examples. renormalize_logits (`bool`): Whether to renormalize the logits after applying all the logits processors (including the custom ones). It's highly recommended to set this flag to `True` as the search algorithms suppose the score logits are normalized but some logit processors break the normalization. forced_bos_token_id (`int`, *optional*, defaults to `model.config.forced_bos_token_id`): The id of the token to force as the first generated token after the `decoder_start_token_id`. Useful for multilingual models like [mBART](../model_doc/mbart) where the first generated token needs to be the target language token. forced_eos_token_id (`int` or list[int]`, *optional*, defaults to `model.config.forced_eos_token_id`): The id of the token to force as the last generated token when `max_length` is reached. Optionally, use a list to set multiple *end-of-sequence* tokens. remove_invalid_values (`bool`): Whether to remove possible *nan* and *inf* outputs of the model to prevent the generation method to crash. Note that using `remove_invalid_values` can slow down generation. exponential_decay_length_penalty (`tuple(int, float)`, *optional*): This Tuple adds an exponentially increasing length penalty, after a certain amount of tokens have been generated. The tuple shall consist of: `(start_index, decay_factor)` where `start_index` indicates where penalty starts and `decay_factor` represents the factor of exponential decay suppress_tokens (`list[int]`, *optional*): A list of tokens that will be suppressed at generation. The `SuppressTokens` logit processor will set their log probs to `-inf` so that they are not sampled. begin_suppress_tokens (`list[int]`, *optional*): A list of tokens that will be suppressed at the beginning of the generation. The `SuppressBeginTokens` logit processor will set their log probs to `-inf` so that they are not sampled. sequence_bias (`dict[tuple[int], float]`, *optional*)): Dictionary that maps a sequence of tokens to its bias term. Positive biases increase the odds of the sequence being selected, while negative biases do the opposite. Check [`~generation.SequenceBiasLogitsProcessor`] for further documentation and examples. token_healing (`bool`): Heal tail tokens of prompts by replacing them with their appropriate extensions. This enhances the quality of completions for prompts affected by greedy tokenization bias. guidance_scale (`float`, *optional*): The guidance scale for classifier free guidance (CFG). CFG is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages the model to generate samples that are more closely linked to the input prompt, usually at the expense of poorer quality. watermarking_config (`BaseWatermarkingConfig` or `dict`, *optional*): Arguments used to watermark the model outputs by adding a small bias to randomly selected set of "green" tokens. See the docs of [`SynthIDTextWatermarkingConfig`] and [`WatermarkingConfig`] for more details. If passed as `Dict`, it will be converted to a `WatermarkingConfig` internally. > Parameters that define the output variables of generate num_return_sequences (`int`, *optional*): The number of independently computed returned sequences for each element in the batch. output_attentions (`bool`): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more details. output_hidden_states (`bool`): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more details. output_scores (`bool`): Whether or not to return the prediction scores. See `scores` under returned tensors for more details. output_logits (`bool`): Whether or not to return the unprocessed prediction logit scores. See `logits` under returned tensors for more details. return_dict_in_generate (`bool`): Whether or not to return a [`~utils.ModelOutput`], as opposed to returning exclusively the generated sequence. This flag must be set to `True` to return the generation cache (when `use_cache` is `True`) or optional outputs (see flags starting with `output_`) > Special tokens that can be used at generation time pad_token_id (`int`, *optional*): The id of the *padding* token. bos_token_id (`int`, *optional*): The id of the *beginning-of-sequence* token. eos_token_id (`Union[int, list[int]]`, *optional*): The id of the *end-of-sequence* token. Optionally, use a list to set multiple *end-of-sequence* tokens. > Generation parameters exclusive to encoder-decoder models encoder_no_repeat_ngram_size (`int`, *optional*): If set to int > 0, all ngrams of that size that occur in the `encoder_input_ids` cannot occur in the `decoder_input_ids`. decoder_start_token_id (`int` or `list[int]`, *optional*): If an encoder-decoder model starts decoding with a different token than *bos*, the id of that token or a list of length `batch_size`. Indicating a list enables different start ids for each element in the batch (e.g. multilingual models with different target languages in one batch) > Generation parameters exclusive to assistant generation is_assistant (`bool`): Whether the model is an assistant (draft) model. num_assistant_tokens (`int`, *optional*): Defines the number of _speculative tokens_ that shall be generated by the assistant model before being checked by the target model at each iteration. Higher values for `num_assistant_tokens` make the generation more _speculative_ : If the assistant model is performant larger speed-ups can be reached, if the assistant model requires lots of corrections, lower speed-ups are reached. num_assistant_tokens_schedule (`str`, *optional*): Defines the schedule at which max assistant tokens shall be changed during inference. - `"heuristic"`: When all speculative tokens are correct, increase `num_assistant_tokens` by 2 else reduce by 1. `num_assistant_tokens` value is persistent over multiple generation calls with the same assistant model. - `"heuristic_transient"`: Same as `"heuristic"` but `num_assistant_tokens` is reset to its initial value after each generation call. - `"constant"`: `num_assistant_tokens` stays unchanged during generation assistant_confidence_threshold (`float`, *optional*): The confidence threshold for the assistant model. If the assistant model's confidence in its prediction for the current token is lower than this threshold, the assistant model stops the current token generation iteration, even if the number of _speculative tokens_ (defined by `num_assistant_tokens`) is not yet reached. The assistant's confidence threshold is adjusted throughout the speculative iterations to reduce the number of unnecessary draft and target forward passes, biased towards avoiding false negatives. `assistant_confidence_threshold` value is persistent over multiple generation calls with the same assistant model. It is an unsupervised version of the dynamic speculation lookahead from Dynamic Speculation Lookahead Accelerates Speculative Decoding of Large Language Models . prompt_lookup_num_tokens (`int`, *optional*): The number of tokens to be output as candidate tokens. max_matching_ngram_size (`int`, *optional*): The maximum ngram size to be considered for matching in the prompt. Default to 2 if not provided. assistant_early_exit(`int`, *optional*): If set to a positive integer, early exit of the model will be used as an assistant. Can only be used with models that support early exit (i.e. models where logits from intermediate layers can be interpreted by the LM head). assistant_lookbehind(`int`, *optional*): If set to a positive integer, the re-encodeing process will additionally consider the last `assistant_lookbehind` assistant tokens to correctly align tokens. Can only be used with different tokenizers in speculative decoding. See this [blog](https://huggingface.co/blog/universal_assisted_generation) for more details. target_lookbehind(`int`, *optional*): If set to a positive integer, the re-encodeing process will additionally consider the last `target_lookbehind` target tokens to correctly align tokens. Can only be used with different tokenizers in speculative decoding. See this [blog](https://huggingface.co/blog/universal_assisted_generation) for more details. > Parameters related to performances and compilation compile_config (CompileConfig, *optional*): If using a compilable cache, this controls how `generate` will `compile` the forward pass for faster inference. disable_compile (`bool`): Whether to disable the automatic compilation of the forward pass. Automatic compilation happens when specific criteria are met, including using a compilable cache. Please open an issue if you find the need to use this flag. """ extra_output_flags = ("output_attentions", "output_hidden_states", "output_scores", "output_logits") # Tensor versions of token IDs, set by _prepare_special_tokens() at generation time _bos_token_tensor: "torch.Tensor | None" _eos_token_tensor: "torch.Tensor | None" _pad_token_tensor: "torch.Tensor | None" _decoder_start_token_tensor: "torch.Tensor | None" # Hash to detect whether the instance was modified after loading _original_object_hash: int | None def __init__(self, **kwargs): # Parameters that control the length of the output self.max_length = kwargs.pop("max_length", None) self.max_new_tokens = kwargs.pop("max_new_tokens", None) self.min_length = kwargs.pop("min_length", None) self.min_new_tokens = kwargs.pop("min_new_tokens", None) self.early_stopping = kwargs.pop("early_stopping", None) self.max_time = kwargs.pop("max_time", None) self.stop_strings = kwargs.pop("stop_strings", None) # Parameters that control the generation strategy used self.do_sample = kwargs.pop("do_sample", None) self.num_beams = kwargs.pop("num_beams", None) # Parameters that control the cache self.use_cache = kwargs.pop("use_cache", None) self.cache_implementation = kwargs.pop("cache_implementation", None) self.cache_config = kwargs.pop("cache_config", None) # Parameters for manipulation of the model output logits self.temperature = kwargs.pop("temperature", None) self.top_k = kwargs.pop("top_k", None) self.top_p = kwargs.pop("top_p", None) self.min_p = kwargs.pop("min_p", None) self.top_h = kwargs.pop("top_h", None) self.typical_p = kwargs.pop("typical_p", None) self.epsilon_cutoff = kwargs.pop("epsilon_cutoff", None) self.eta_cutoff = kwargs.pop("eta_cutoff", None) self.repetition_penalty = kwargs.pop("repetition_penalty", None) self.encoder_repetition_penalty = kwargs.pop("encoder_repetition_penalty", None) self.length_penalty = kwargs.pop("length_penalty", None) self.no_repeat_ngram_size = kwargs.pop("no_repeat_ngram_size", None) self.bad_words_ids = kwargs.pop("bad_words_ids", None) self.renormalize_logits = kwargs.pop("renormalize_logits", None) self.forced_bos_token_id = kwargs.pop("forced_bos_token_id", None) self.forced_eos_token_id = kwargs.pop("forced_eos_token_id", None) self.remove_invalid_values = kwargs.pop("remove_invalid_values", None) self.exponential_decay_length_penalty = kwargs.pop("exponential_decay_length_penalty", None) self.suppress_tokens = kwargs.pop("suppress_tokens", None) self.begin_suppress_tokens = kwargs.pop("begin_suppress_tokens", None) self.sequence_bias = kwargs.pop("sequence_bias", None) self.token_healing = kwargs.pop("token_healing", None) self.guidance_scale = kwargs.pop("guidance_scale", None) self.watermarking_config = kwargs.pop("watermarking_config", None) if isinstance(self.watermarking_config, dict): self.watermarking_config = WatermarkingConfig.from_dict(self.watermarking_config) # Parameters that define the output variables of `generate` self.num_return_sequences = kwargs.pop("num_return_sequences", None) self.output_attentions = kwargs.pop("output_attentions", None) self.output_hidden_states = kwargs.pop("output_hidden_states", None) self.output_scores = kwargs.pop("output_scores", None) self.output_logits = kwargs.pop("output_logits", None) self.return_dict_in_generate = kwargs.pop("return_dict_in_generate", None) # Special tokens that can be used at generation time self.pad_token_id = kwargs.pop("pad_token_id", None) self.bos_token_id = kwargs.pop("bos_token_id", None) self.eos_token_id = kwargs.pop("eos_token_id", None) # Generation parameters exclusive to encoder-decoder models self.encoder_no_repeat_ngram_size = kwargs.pop("encoder_no_repeat_ngram_size", None) self.decoder_start_token_id = kwargs.pop("decoder_start_token_id", None) # Assistant generation self.is_assistant = kwargs.pop("is_assistant", None) self.num_assistant_tokens = kwargs.pop("num_assistant_tokens", None) self.num_assistant_tokens_schedule = kwargs.pop("num_assistant_tokens_schedule", None) self.assistant_confidence_threshold = kwargs.pop("assistant_confidence_threshold", None) self.prompt_lookup_num_tokens = kwargs.pop("prompt_lookup_num_tokens", None) self.max_matching_ngram_size = kwargs.pop("max_matching_ngram_size", None) self.assistant_early_exit = kwargs.pop("assistant_early_exit", None) self.assistant_lookbehind = kwargs.pop("assistant_lookbehind", None) self.target_lookbehind = kwargs.pop("target_lookbehind", None) # Performance self.compile_config = kwargs.pop("compile_config", None) self.disable_compile = kwargs.pop("disable_compile", None) self.continuous_batching_config = kwargs.pop("continuous_batching_config", None) # Deprecated (moved to the Hub). TODO remove for v5 self.low_memory = kwargs.pop("low_memory", None) self.penalty_alpha = kwargs.pop("penalty_alpha", None) self.dola_layers = kwargs.pop("dola_layers", None) self.diversity_penalty = kwargs.pop("diversity_penalty", None) self.num_beam_groups = kwargs.pop("num_beam_groups", None) self.constraints = kwargs.pop("constraints", None) self.force_words_ids = kwargs.pop("force_words_ids", None) self.prefill_chunk_size = kwargs.pop("prefill_chunk_size", None) # Common attributes self._commit_hash = kwargs.pop("_commit_hash", None) self._from_model_config = kwargs.pop("_from_model_config", None) self.transformers_version = kwargs.pop("transformers_version", None) # Additional attributes without default values if not self._from_model_config: # we don't want to copy values from the model config if we're initializing # a `GenerationConfig` from a model's default configuration file for key, value in kwargs.items(): try: setattr(self, key, value) except AttributeError as err: logger.error(f"Can't set {key} with value {value} for {self}") raise err else: # Ensure backward compatibility for models that use `forced_bos_token_id` within their config if kwargs.get("force_bos_token_to_be_generated", False): self.forced_bos_token_id = self.bos_token_id logger.warning_once( f"Please make sure the generation config includes `forced_bos_token_id={self.bos_token_id}`. " ) # Validate the values of the attributes self.validate() def __hash__(self): return hash(self.to_json_string(ignore_metadata=True)) def __eq__(self, other): if not isinstance(other, GenerationConfig): return False self_without_metadata = self.to_json_string(use_diff=False, ignore_metadata=True) other_without_metadata = other.to_json_string(use_diff=False, ignore_metadata=True) return self_without_metadata == other_without_metadata def __repr__(self): return f"{self.__class__.__name__} {self.to_json_string(ignore_metadata=True)}" def get_generation_mode(self, assistant_model: Optional["PreTrainedModel"] = None) -> GenerationMode: """ Returns the generation mode triggered by the [`GenerationConfig`] instance. Arg: assistant_model (`PreTrainedModel`, *optional*): The assistant model to be used for assisted generation. If set, the generation mode will be assisted generation. Returns: `GenerationMode`: The generation mode triggered by the instance. """ # TODO joao: find out a way of not depending on external fields (e.g. `assistant_model`), then make this a # property and part of the `__repr__` if self.constraints is not None or self.force_words_ids is not None: generation_mode = GenerationMode.CONSTRAINED_BEAM_SEARCH elif self.num_beams is None or self.num_beams == 1: if self.do_sample is not True: if ( self.top_k is not None and self.top_k > 1 and self.penalty_alpha is not None and self.penalty_alpha > 0 ): generation_mode = GenerationMode.CONTRASTIVE_SEARCH else: generation_mode = GenerationMode.GREEDY_SEARCH else: generation_mode = GenerationMode.SAMPLE else: if self.num_beam_groups is not None and self.num_beam_groups > 1: generation_mode = GenerationMode.GROUP_BEAM_SEARCH elif self.do_sample is True: generation_mode = GenerationMode.BEAM_SAMPLE else: generation_mode = GenerationMode.BEAM_SEARCH # Assisted generation may extend some generation modes if ( assistant_model is not None or self.prompt_lookup_num_tokens is not None or self.assistant_early_exit is not None ): if generation_mode in ("greedy_search", "sample"): generation_mode = GenerationMode.ASSISTED_GENERATION else: logger.warning( "You've set `assistant_model`, which triggers assisted generate. Currently, assisted generate " "is only supported with Greedy Search and Sample. However, the base decoding mode (based on " f"current flags) is {generation_mode} -- some of the set flags will be ignored." ) # DoLa generation may extend some generation modes # TODO joao, manuel: remove this in v4.62.0 if self.dola_layers is not None: if generation_mode in ("greedy_search", "sample"): generation_mode = GenerationMode.DOLA_GENERATION else: logger.warning( "You've set `dola_layers`, which triggers DoLa generate. Currently, DoLa generate " "is only supported with Greedy Search and Sample. However, the base decoding mode (based on " f"current flags) is {generation_mode} -- some of the set flags will be ignored." ) return generation_mode @staticmethod def _get_default_generation_params() -> dict[str, Any]: return { "max_length": 20, "min_length": 0, "do_sample": False, "use_cache": True, "early_stopping": False, "num_beams": 1, "temperature": 1.0, "top_k": 50, "top_p": 1.0, "typical_p": 1.0, "repetition_penalty": 1.0, "length_penalty": 1.0, "no_repeat_ngram_size": 0, "encoder_no_repeat_ngram_size": 0, "bad_words_ids": None, "num_return_sequences": 1, "output_scores": False, "return_dict_in_generate": False, "forced_bos_token_id": None, "forced_eos_token_id": None, "remove_invalid_values": False, "exponential_decay_length_penalty": None, "suppress_tokens": None, "begin_suppress_tokens": None, "epsilon_cutoff": 0.0, "eta_cutoff": 0.0, "encoder_repetition_penalty": 1.0, "num_assistant_tokens": 20, "num_assistant_tokens_schedule": "constant", "assistant_confidence_threshold": 0.4, "assistant_lookbehind": 10, "target_lookbehind": 10, # Deprecated arguments (moved to the Hub). TODO joao, manuel: remove in v4.62.0 "num_beam_groups": 1, "diversity_penalty": 0.0, } def validate(self, strict=False): """ Validates the values of the attributes of the [`GenerationConfig`] instance. Raises exceptions in the presence of parameterization that can be detected as incorrect from the configuration instance alone. Note that some parameters not validated here are best validated at generate runtime, as they may depend on other inputs and/or the model, such as parameters related to the generation length. Args: strict (bool): If True, raise an exception for any issues found. If False, only log issues. """ minor_issues = {} # format: {attribute_name: issue_description} # 1. Validation of individual attributes # 1.1. Decoding attributes if self.early_stopping not in {None, True, False, "never"}: raise ValueError(f"`early_stopping` must be a boolean or 'never', but is {self.early_stopping}.") if self.max_new_tokens is not None and self.max_new_tokens <= 0: raise ValueError(f"`max_new_tokens` must be greater than 0, but is {self.max_new_tokens}.") if self.pad_token_id is not None and self.pad_token_id < 0: minor_issues["pad_token_id"] = ( f"`pad_token_id` should be positive but got {self.pad_token_id}. This will cause errors when batch " "generating, if there is padding. Please set `pad_token_id` explicitly as " "`model.generation_config.pad_token_id=PAD_TOKEN_ID` to avoid errors in generation" ) # 1.2. Cache attributes # "paged" re-routes to continuous batching and so it is a valid cache implementation. But we do not want to test # it with the `generate` as the other would be, so we we cannot add it to ALL_CACHE_IMPLEMENTATIONS valid_cache_implementations = ALL_CACHE_IMPLEMENTATIONS + ("paged",) if self.cache_implementation is not None and self.cache_implementation not in valid_cache_implementations: raise ValueError( f"Invalid `cache_implementation` ({self.cache_implementation}). Choose one of: " f"{valid_cache_implementations}" ) # 1.3. Performance attributes if self.compile_config is not None and not isinstance(self.compile_config, CompileConfig): raise ValueError( f"You provided `compile_config` as an instance of {type(self.compile_config)}, but it must be an " "instance of `CompileConfig`." ) # 1.4. Watermarking attributes if self.watermarking_config is not None: self.watermarking_config.validate() # 2. Validation of attribute combinations # 2.1. detect sampling-only parameterization when not in sampling mode # Note that we check `is not True` in purpose. Boolean fields can also be `None` so we # have to be explicit. Value of `None` is same as having `False`, i.e. the default value if self.do_sample is not True: greedy_wrong_parameter_msg = ( "`do_sample` is set not to set `True`. However, `{flag_name}` is set to `{flag_value}` -- this flag is only " "used in sample-based generation modes. You should set `do_sample=True` or unset `{flag_name}`." ) if self.temperature is not None and self.temperature != 1.0: minor_issues["temperature"] = greedy_wrong_parameter_msg.format( flag_name="temperature", flag_value=self.temperature ) if self.top_p is not None and self.top_p != 1.0: minor_issues["top_p"] = greedy_wrong_parameter_msg.format(flag_name="top_p", flag_value=self.top_p) if self.min_p is not None: minor_issues["min_p"] = greedy_wrong_parameter_msg.format(flag_name="min_p", flag_value=self.min_p) if self.top_h is not None: minor_issues["top_h"] = greedy_wrong_parameter_msg.format(flag_name="top_h", flag_value=self.top_h) if self.typical_p is not None and self.typical_p != 1.0: minor_issues["typical_p"] = greedy_wrong_parameter_msg.format( flag_name="typical_p", flag_value=self.typical_p ) if self.top_k is not None and self.top_k != 50: minor_issues["top_k"] = greedy_wrong_parameter_msg.format(flag_name="top_k", flag_value=self.top_k) if self.epsilon_cutoff is not None and self.epsilon_cutoff != 0.0: minor_issues["epsilon_cutoff"] = greedy_wrong_parameter_msg.format( flag_name="epsilon_cutoff", flag_value=self.epsilon_cutoff ) if self.eta_cutoff is not None and self.eta_cutoff != 0.0: minor_issues["eta_cutoff"] = greedy_wrong_parameter_msg.format( flag_name="eta_cutoff", flag_value=self.eta_cutoff ) # 2.2. detect beam-only parameterization when not in beam mode if self.num_beams is None or self.num_beams == 1: single_beam_wrong_parameter_msg = ( "`num_beams` is set to {num_beams}. However, `{flag_name}` is set to `{flag_value}` -- this flag is only used " "in beam-based generation modes. You should set `num_beams>1` or unset `{flag_name}`." ) if self.early_stopping is not None and self.early_stopping is not False: minor_issues["early_stopping"] = single_beam_wrong_parameter_msg.format( num_beams=self.num_beams, flag_name="early_stopping", flag_value=self.early_stopping ) if self.length_penalty is not None and self.length_penalty != 1.0: minor_issues["length_penalty"] = single_beam_wrong_parameter_msg.format( num_beams=self.num_beams, flag_name="length_penalty", flag_value=self.length_penalty ) # 2.4. check `num_return_sequences` if self.num_return_sequences is not None and self.num_return_sequences > 1: if self.num_beams is None or self.num_beams == 1: if not self.do_sample: raise ValueError( "Greedy methods (do_sample != True) without beam search do not support " f"`num_return_sequences` different than 1 (got {self.num_return_sequences})." ) elif ( self.num_beams is not None and self.num_return_sequences is not None and self.num_return_sequences > self.num_beams ): raise ValueError( f"`num_return_sequences` ({self.num_return_sequences}) has to be smaller or equal to `num_beams` " f"({self.num_beams})." ) # 2.5. check cache-related arguments if self.use_cache is False: # In this case, all cache-related arguments should be unset. However, since `use_cache=False` is often used # passed to `generate` directly to hot-fix cache issues, let's raise a warning instead of an error # (otherwise a user might need to overwrite several parameters). no_cache_warning = ( "You have not set `use_cache` to `True`, but {cache_arg} is set to {cache_arg_value}." "{cache_arg} will have no effect." ) for arg_name in ("cache_implementation", "cache_config"): if getattr(self, arg_name) is not None: minor_issues[arg_name] = no_cache_warning.format( cache_arg=arg_name, cache_arg_value=getattr(self, arg_name) ) # 2.6. other incorrect combinations if self.return_dict_in_generate is not True: for extra_output_flag in self.extra_output_flags: if getattr(self, extra_output_flag) is True: minor_issues[extra_output_flag] = ( f"`return_dict_in_generate` is NOT set to `True`, but `{extra_output_flag}` is. When " f"`return_dict_in_generate` is not `True`, `{extra_output_flag}` is ignored." ) # 3. Check common issue: passing `generate` arguments inside the generation config generate_arguments = ( "logits_processor", "stopping_criteria", "prefix_allowed_tokens_fn", "synced_gpus", "assistant_model", "streamer", "negative_prompt_ids", "negative_prompt_attention_mask", ) for arg in generate_arguments: if hasattr(self, arg): raise ValueError( f"Argument `{arg}` is not a valid argument of `GenerationConfig`. It should be passed to " "`generate()` (or a pipeline) directly." ) # Finally, handle caught minor issues. With default parameterization, we will throw a minimal warning. if len(minor_issues) > 0: # Full list of issues with potential fixes info_message = [] for attribute_name, issue_description in minor_issues.items(): info_message.append(f"- `{attribute_name}`: {issue_description}") info_message = "\n".join(info_message) info_message += ( "\nIf you're using a pretrained model, note that some of these attributes may be set through the " "model's `generation_config.json` file." ) if strict: raise ValueError("GenerationConfig is invalid: \n" + info_message) else: attributes_with_issues = list(minor_issues.keys()) warning_message = ( f"The following generation flags are not valid and may be ignored: {attributes_with_issues}." ) if logging.get_verbosity() >= logging.WARNING: warning_message += " Set `TRANSFORMERS_VERBOSITY=info` for more details." logger.warning_once(warning_message) logger.info_once(info_message) def save_pretrained( self, save_directory: str | os.PathLike, config_file_name: str | os.PathLike | None = None, push_to_hub: bool = False, **kwargs, ): r""" Save a generation configuration object to the directory `save_directory`, so that it can be re-loaded using the [`~GenerationConfig.from_pretrained`] class method. Args: save_directory (`str` or `os.PathLike`): Directory where the configuration JSON file will be saved (will be created if it does not exist). config_file_name (`str` or `os.PathLike`, *optional*, defaults to `"generation_config.json"`): Name of the generation configuration JSON file to be saved in `save_directory`. push_to_hub (`bool`, *optional*, defaults to `False`): Whether or not to push your model to the Hugging Face model hub after saving it. You can specify the repository you want to push to with `repo_id` (will default to the name of `save_directory` in your namespace). kwargs (`dict[str, Any]`, *optional*): Additional key word arguments passed along to the [`~utils.PushToHubMixin.push_to_hub`] method. """ # At save time, validate the instance enforcing strictness -- if any warning/exception would be thrown, we # refuse to save the instance. # This strictness is enforced to prevent bad configurations from being saved and re-used. try: self.validate(strict=True) except ValueError as exc: raise ValueError(str(exc) + "\n\nFix these issues to save the configuration.") config_file_name = config_file_name if config_file_name is not None else GENERATION_CONFIG_NAME if os.path.isfile(save_directory): raise AssertionError(f"Provided path ({save_directory}) should be a directory, not a file") os.makedirs(save_directory, exist_ok=True) if push_to_hub: commit_message = kwargs.pop("commit_message", None) repo_id = kwargs.pop("repo_id", str(save_directory).split(os.path.sep)[-1]) repo_id = create_repo(repo_id, exist_ok=True, **kwargs).repo_id files_timestamps = self._get_files_timestamps(save_directory) output_config_file = os.path.join(save_directory, config_file_name) self.to_json_file(output_config_file, use_diff=True, keys_to_pop=["compile_config"]) logger.info(f"Configuration saved in {output_config_file}") if push_to_hub: self._upload_modified_files( save_directory, repo_id, files_timestamps, commit_message=commit_message, token=kwargs.get("token"), ) @classmethod def from_pretrained( cls, pretrained_model_name: str | os.PathLike, config_file_name: str | os.PathLike | None = None, cache_dir: str | os.PathLike | None = None, force_download: bool = False, local_files_only: bool = False, token: str | bool | None = None, revision: str = "main", **kwargs, ) -> "GenerationConfig": r""" Instantiate a [`GenerationConfig`] from a generation configuration file. Args: pretrained_model_name (`str` or `os.PathLike`): This can be either: - a string, the *model id* of a pretrained model configuration hosted inside a model repo on huggingface.co. - a path to a *directory* containing a configuration file saved using the [`~GenerationConfig.save_pretrained`] method, e.g., `./my_model_directory/`. config_file_name (`str` or `os.PathLike`, *optional*, defaults to `"generation_config.json"`): Name of the generation configuration JSON file to be loaded from `pretrained_model_name`. cache_dir (`str` or `os.PathLike`, *optional*): Path to a directory in which a downloaded pretrained model configuration should be cached if the standard cache should not be used. force_download (`bool`, *optional*, defaults to `False`): Whether or not to force to (re-)download the configuration files and override the cached versions if they exist. proxies (`dict[str, str]`, *optional*): A dictionary of proxy servers to use by protocol or endpoint, e.g., `{'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}.` The proxies are used on each request. token (`str` or `bool`, *optional*): The token to use as HTTP bearer authorization for remote files. If `True`, or not specified, will use the token generated when running `hf auth login` (stored in `~/.huggingface`). revision (`str`, *optional*, defaults to `"main"`): The specific model version to use. It can be a branch name, a tag name, or a commit id, since we use a git-based system for storing models and other artifacts on huggingface.co, so `revision` can be any identifier allowed by git. To test a pull request you made on the Hub, you can pass `revision="refs/pr/"`. return_unused_kwargs (`bool`, *optional*, defaults to `False`): If `False`, then this function returns just the final configuration object. If `True`, then this functions returns a `Tuple(config, unused_kwargs)` where *unused_kwargs* is a dictionary consisting of the key/value pairs whose keys are not configuration attributes: i.e., the part of `kwargs` which has not been used to update `config` and is otherwise ignored. subfolder (`str`, *optional*, defaults to `""`): In case the relevant files are located inside a subfolder of the model repo on huggingface.co, you can specify the folder name here. kwargs (`dict[str, Any]`, *optional*): The values in kwargs of any keys which are configuration attributes will be used to override the loaded values. Behavior concerning key/value pairs whose keys are *not* configuration attributes is controlled by the `return_unused_kwargs` keyword parameter. Returns: [`GenerationConfig`]: The configuration object instantiated from this pretrained model. Examples: ```python >>> from transformers import GenerationConfig >>> # Download configuration from huggingface.co and cache. >>> generation_config = GenerationConfig.from_pretrained("openai-community/gpt2") >>> # E.g. config was saved using *save_pretrained('./test/saved_model/')* >>> generation_config.save_pretrained("./test/saved_model/") >>> generation_config = GenerationConfig.from_pretrained("./test/saved_model/") >>> # You can also specify configuration names to your generation configuration file >>> generation_config.save_pretrained("./test/saved_model/", config_file_name="my_configuration.json") >>> generation_config = GenerationConfig.from_pretrained("./test/saved_model/", "my_configuration.json") >>> # If you'd like to try a minor variation to an existing configuration, you can also pass generation >>> # arguments to `.from_pretrained()`. Be mindful that typos and unused arguments will be ignored >>> generation_config, unused_kwargs = GenerationConfig.from_pretrained( ... "openai-community/gpt2", top_k=1, foo=False, do_sample=True, return_unused_kwargs=True ... ) >>> generation_config.top_k 1 >>> unused_kwargs {'foo': False} ```""" config_file_name = config_file_name if config_file_name is not None else GENERATION_CONFIG_NAME proxies = kwargs.pop("proxies", None) subfolder = kwargs.pop("subfolder", "") from_pipeline = kwargs.pop("_from_pipeline", None) from_auto_class = kwargs.pop("_from_auto", False) commit_hash = kwargs.pop("_commit_hash", None) user_agent = {"file_type": "config", "from_auto_class": from_auto_class} if from_pipeline is not None: user_agent["using_pipeline"] = from_pipeline config_path = os.path.join(pretrained_model_name, config_file_name) config_path = str(config_path) is_local = os.path.exists(config_path) if os.path.isfile(os.path.join(subfolder, config_path)): # Special case when config_path is a local file resolved_config_file = config_path is_local = True else: configuration_file = config_file_name try: # Load from local folder or from cache or download from model Hub and cache resolved_config_file = cached_file( pretrained_model_name, configuration_file, cache_dir=cache_dir, force_download=force_download, proxies=proxies, local_files_only=local_files_only, token=token, user_agent=user_agent, revision=revision, subfolder=subfolder, _commit_hash=commit_hash, ) commit_hash = extract_commit_hash(resolved_config_file, commit_hash) except OSError: # Raise any environment error raise by `cached_file`. It will have a helpful error message adapted to # the original exception. raise except Exception: # For any other exception, we throw a generic error. raise OSError( f"Can't load the configuration of '{pretrained_model_name}'. If you were trying to load it" " from 'https://huggingface.co/models', make sure you don't have a local directory with the same" f" name. Otherwise, make sure '{pretrained_model_name}' is the correct path to a directory" f" containing a {configuration_file} file" ) try: # Load config dict config_dict = cls._dict_from_json_file(resolved_config_file) config_dict["_commit_hash"] = commit_hash except (json.JSONDecodeError, UnicodeDecodeError): raise OSError(f"It looks like the config file at '{resolved_config_file}' is not a valid JSON file.") if is_local: logger.info(f"loading configuration file {resolved_config_file}") else: logger.info(f"loading configuration file {configuration_file} from cache at {resolved_config_file}") if kwargs.get("_from_model_config", False): return cls.from_model_config(config_dict) elif kwargs.get("return_unused_kwargs") is True: config, unused_kwargs = cls.from_dict(config_dict, **kwargs) config._original_object_hash = hash(config) # Hash to detect whether the instance was modified return config, unused_kwargs else: config = cls.from_dict(config_dict, **kwargs) config._original_object_hash = hash(config) # Hash to detect whether the instance was modified return config @classmethod def _dict_from_json_file(cls, json_file: str | os.PathLike): with open(json_file, "r", encoding="utf-8") as reader: text = reader.read() return json.loads(text) @classmethod def from_dict(cls, config_dict: dict[str, Any], **kwargs) -> "GenerationConfig": """ Instantiates a [`GenerationConfig`] from a Python dictionary of parameters. Args: config_dict (`dict[str, Any]`): Dictionary that will be used to instantiate the configuration object. kwargs (`dict[str, Any]`): Additional parameters from which to initialize the configuration object. Returns: [`GenerationConfig`]: The configuration object instantiated from those parameters. """ return_unused_kwargs = kwargs.pop("return_unused_kwargs", False) # Those arguments may be passed along for our internal telemetry. # We remove them so they don't appear in `return_unused_kwargs`. kwargs.pop("_from_auto", None) kwargs.pop("_from_pipeline", None) # The commit hash might have been updated in the `config_dict`, we don't want the kwargs to erase that update. if "_commit_hash" in kwargs and "_commit_hash" in config_dict: kwargs["_commit_hash"] = config_dict["_commit_hash"] # The line below allows model-specific config to be loaded as well through kwargs, with safety checks. # See https://github.com/huggingface/transformers/pull/21269 config = cls(**{**config_dict, **kwargs}) unused_kwargs = config.update(**kwargs) logger.info(f"Generate config {config}") if return_unused_kwargs: return config, unused_kwargs else: return config def dict_dtype_to_str(self, d: dict[str, Any]) -> None: """ Checks whether the passed dictionary and its nested dicts have a *dtype* key and if it's not None, converts torch.dtype to a string of just the type. For example, `torch.float32` get converted into *"float32"* string, which can then be stored in the json format. """ if d.get("dtype") is not None and not isinstance(d["dtype"], str): d["dtype"] = str(d["dtype"]).split(".")[1] for value in d.values(): if isinstance(value, dict): self.dict_dtype_to_str(value) def to_diff_dict(self) -> dict[str, Any]: """ Removes all attributes from config which correspond to the default config attributes for better readability and serializes to a Python dictionary. Returns: `dict[str, Any]`: Dictionary of all the attributes that make up this configuration instance, """ config_dict = self.to_dict() # get the default config dict default_config_dict = GenerationConfig().to_dict() serializable_config_dict = {} # only serialize values that differ from the default config for key, value in config_dict.items(): if key not in default_config_dict or key == "transformers_version" or value != default_config_dict[key]: serializable_config_dict[key] = value self.dict_dtype_to_str(serializable_config_dict) return serializable_config_dict def to_dict(self) -> dict[str, Any]: """ Serializes this instance to a Python dictionary. Returns: `dict[str, Any]`: Dictionary of all the attributes that make up this configuration instance. """ output = copy.deepcopy(self.__dict__) # Fields to ignore at serialization time if "_commit_hash" in output: del output["_commit_hash"] if "_original_object_hash" in output: del output["_original_object_hash"] # Transformers version when serializing this file output["transformers_version"] = __version__ self.dict_dtype_to_str(output) return output def to_json_string( self, use_diff: bool = True, ignore_metadata: bool = False, keys_to_pop: list[str] | None = None ) -> str: """ Serializes this instance to a JSON string. Args: use_diff (`bool`, *optional*, defaults to `True`): If set to `True`, only the difference between the config instance and the default `GenerationConfig()` is serialized to JSON string. ignore_metadata (`bool`, *optional*, defaults to `False`): Whether to ignore the metadata fields present in the instance keys_to_pop (`list[str]`, *optional*): Keys to pop from the config dictionary before serializing Returns: `str`: String containing all the attributes that make up this configuration instance in JSON format. """ if use_diff is True: config_dict = self.to_diff_dict() else: config_dict = self.to_dict() if keys_to_pop is not None: for key in keys_to_pop: config_dict.pop(key, None) if ignore_metadata: for metadata_field in METADATA_FIELDS: config_dict.pop(metadata_field, None) def convert_keys_to_string(obj): if isinstance(obj, dict): return {str(key): convert_keys_to_string(value) for key, value in obj.items()} elif isinstance(obj, list): return [convert_keys_to_string(item) for item in obj] else: return obj def convert_dataclass_to_dict(obj): if isinstance(obj, dict): return {key: convert_dataclass_to_dict(value) for key, value in obj.items()} elif is_dataclass(obj): return obj.to_dict() else: return obj config_dict = convert_keys_to_string(config_dict) config_dict = convert_dataclass_to_dict(config_dict) return json.dumps(config_dict, indent=2, sort_keys=True) + "\n" def to_json_file( self, json_file_path: str | os.PathLike, use_diff: bool = True, keys_to_pop: list[str] | None = None ) -> None: """ Save this instance to a JSON file. Args: json_file_path (`str` or `os.PathLike`): Path to the JSON file in which this configuration instance's parameters will be saved. use_diff (`bool`, *optional*, defaults to `True`): If set to `True`, only the difference between the config instance and the default `GenerationConfig()` is serialized to JSON file. keys_to_pop (`list[str]`, *optional*): Keys to pop from the config dictionary before serializing """ with open(json_file_path, "w", encoding="utf-8") as writer: writer.write(self.to_json_string(use_diff=use_diff, keys_to_pop=keys_to_pop)) @classmethod def from_model_config(cls, model_config: Union["PreTrainedConfig", dict]) -> "GenerationConfig": """ Instantiates a [`GenerationConfig`] from a [`PreTrainedConfig`]. This function is useful to convert legacy [`PreTrainedConfig`] objects, which may contain generation parameters, into a stand-alone [`GenerationConfig`]. Args: model_config (`PreTrainedConfig | dict`): The model config that will be used to instantiate the generation config. Returns: [`GenerationConfig`]: The configuration object instantiated from those parameters. """ config_dict = model_config.to_dict() if not isinstance(model_config, dict) else model_config config_dict.pop("_from_model_config", None) # Removes all `None` from the model config dict -- this lets the generation config defaults to take hold config_dict = {key: value for key, value in config_dict.items() if value is not None} generation_config = cls.from_dict(config_dict, return_unused_kwargs=False, _from_model_config=True) # Special case: some models have generation attributes set in the decoder. Use them if still unset in the # generation config (which in turn is defined from the outer attributes of model config). if isinstance(model_config, dict): decoder_possible_text_config_names = ("decoder", "generator", "text_config") for text_config_name in decoder_possible_text_config_names: if text_config := model_config.get(text_config_name): model_config = text_config break else: model_config = model_config.get_text_config(decoder=True) model_config = model_config.to_dict() default_generation_config = GenerationConfig() for attr in generation_config.to_dict(): is_unset = getattr(generation_config, attr) == getattr(default_generation_config, attr) if attr in model_config and is_unset: setattr(generation_config, attr, model_config[attr]) # If any `output_...` flag is set to `True`, we ensure `return_dict_in_generate` is set to `True`. if not generation_config.return_dict_in_generate: if any( getattr(generation_config, extra_output_flag, False) for extra_output_flag in generation_config.extra_output_flags ): generation_config.return_dict_in_generate = True # Hash to detect whether the instance was modified generation_config._original_object_hash = hash(generation_config) return generation_config def update(self, defaults_only=False, allow_custom_entries=False, **kwargs): """ Updates attributes of this class instance with attributes from `kwargs` if they match existing attributes, returning all the unused kwargs. Args: defaults_only (`bool`, *optional*, defaults to `False`): Whether to update all keys in config with `kwargs` or only those that are set to `None` (i.e. default value). allow_custom_entries (`bool`, *optional*, defaults to `False`): Whether to allow updating custom entries into the config with `kwargs` if not present in the current config. kwargs (`dict[str, Any]`): Dictionary of attributes to tentatively update this class. Returns: `dict[str, Any]`: Dictionary containing all the key-value pairs that were not used to update the instance. """ to_remove = [] for key, value in kwargs.items(): if allow_custom_entries and not hasattr(self, key): setattr(self, key, value) to_remove.append(key) elif hasattr(self, key): if not defaults_only or getattr(self, key) is None: setattr(self, key, value) to_remove.append(key) # Confirm that the updated instance is still valid self.validate() # Remove all the attributes that were updated, without modifying the input dict unused_kwargs = {key: value for key, value in kwargs.items() if key not in to_remove} return unused_kwargs @dataclass class BaseWatermarkingConfig(ABC): """Generic watermarking config""" @classmethod def from_dict(cls, config_dict, **kwargs): """ Constructs a BaseWatermarkingConfig instance from a dictionary of parameters. Args: config_dict (dict[str, Any]): Dictionary containing configuration parameters. **kwargs: Additional keyword arguments to override dictionary values. Returns: BaseWatermarkingConfig: Instance of BaseWatermarkingConfig constructed from the dictionary. """ config = cls(**config_dict) to_remove = [] for key, value in kwargs.items(): if hasattr(config, key): setattr(config, key, value) to_remove.append(key) for key in to_remove: kwargs.pop(key, None) return config def to_json_file(self, json_file_path: str | os.PathLike): """ Save this instance to a JSON file. Args: json_file_path (Union[str, os.PathLike]): Path to the JSON file in which this configuration instance's parameters will be saved. """ with open(json_file_path, "w", encoding="utf-8") as writer: config_dict = self.to_dict() json_string = json.dumps(config_dict, indent=2, sort_keys=True) + "\n" writer.write(json_string) def to_dict(self) -> dict[str, Any]: """ Serializes this instance to a Python dictionary. Returns: dict[str, Any]: Dictionary of all the attributes that make up this configuration instance. """ output = copy.deepcopy(self.__dict__) return output def __iter__(self): yield from copy.deepcopy(self.__dict__).items() def __repr__(self): return f"{self.__class__.__name__} {self.to_json_string()}" def to_json_string(self): """ Serializes this instance to a JSON formatted string. Returns: str: JSON formatted string representing the configuration instance. """ return json.dumps(self.__dict__, indent=2) + "\n" def update(self, **kwargs): """ Update the configuration attributes with new values. Args: **kwargs: Keyword arguments representing configuration attributes and their new values. """ for key, value in kwargs.items(): if hasattr(self, key): setattr(self, key, value) @abstractmethod def validate(self): ... @abstractmethod def construct_processor(self, vocab_size): ... @dataclass class WatermarkingConfig(BaseWatermarkingConfig): """ Class that holds arguments for watermark generation and should be passed into `GenerationConfig` during `generate`. See [this paper](https://huggingface.co/papers/2306.04634) for more details on the arguments. Accepts the following keys: - greenlist_ratio (`float`): Used for watermarking. The ratio of "green" tokens used to the vocabulary size. Defaults to 0.25. - bias (`float`): Used with watermarking. The bias added to the selected "green" tokens' logits. Defaults to 2.0. - hashing_key (`int`): Hashing key used for watermarking. Defaults to 15485863 (the millionth prime). - seeding_scheme (`str`): Algorithm to use for watermarking. Accepts values: - "lefthash" (default): "green" tokens selection depend on the last token (Algorithm 2 from the paper) - "selfhash": "green" tokens selection depends on the current token itself (Algorithm 3 from the paper) The downside of this scheme is that it considers all possible next tokens and can be slower than "lefthash". - context_width(`int`): The context length of previous tokens to use in seeding. Higher context length makes watermarking more robust. """ def __init__( self, greenlist_ratio: float = 0.25, bias: float = 2.0, hashing_key: int = 15485863, seeding_scheme: str = "lefthash", context_width: int = 1, ): self.greenlist_ratio = greenlist_ratio self.bias = bias self.hashing_key = hashing_key self.seeding_scheme = seeding_scheme self.context_width = context_width def validate(self): watermark_missing_arg_msg = ( "Some of the keys in `watermarking_config` are defined incorrectly. `{key}` should be {correct_value}` " "but found {found_value}" ) if self.seeding_scheme not in ["selfhash", "lefthash"]: raise ValueError( watermark_missing_arg_msg.format( key="seeding_scheme", correct_value="[`selfhash`, `lefthash`]", found_value=self.seeding_scheme, ), ) if not 0.0 <= self.greenlist_ratio <= 1.0: raise ValueError( watermark_missing_arg_msg.format( key="greenlist_ratio", correct_value="in range between 0.0 and 1.0", found_value=self.seeding_scheme, ), ) if not self.context_width >= 1: raise ValueError( watermark_missing_arg_msg.format( key="context_width", correct_value="a positive integer", found_value=self.context_width, ), ) def construct_processor(self, vocab_size: int, device) -> "WatermarkLogitsProcessor": return WatermarkLogitsProcessor( vocab_size=vocab_size, device=device, greenlist_ratio=self.greenlist_ratio, bias=self.bias, hashing_key=self.hashing_key, seeding_scheme=self.seeding_scheme, context_width=self.context_width, ) @dataclass class SynthIDTextWatermarkingConfig(BaseWatermarkingConfig): """ Class that holds arguments for watermark generation and should be passed into `GenerationConfig` during `generate`. See [this paper](https://www.nature.com/articles/s41586-024-08025-4) for more details on the arguments. Args: ngram_len (`int`): Ngram length. keys (`list[int]`): A sequence of watermarking keys, one for each depth. context_history_size (`int`, *optional*, defaults to 1024): Size of the tensor to keep track of seen contexts. sampling_table_seed (`int`, *optional*, defaults to 0): Random seed to generate the sampling table. sampling_table_size (`int`, *optional*, defaults to 65536): Size of the sampling table. skip_first_ngram_calls (`bool`, *optional*, defaults to `False`): Whether to skip first ngram calls. debug_mode (`bool`, optional, *optional*, defaults to `False`): Logits are modified to uniform one got before watermarking modification is applied. This is to test the implementation. Examples: ```python >>> from transformers import AutoModelForCausalLM, AutoTokenizer, SynthIDTextWatermarkingConfig >>> tokenizer = AutoTokenizer.from_pretrained('google/gemma-2-2b', padding_side="left") >>> model = AutoModelForCausalLM.from_pretrained('google/gemma-2-2b') >>> # SynthID Text configuration >>> watermarking_config = SynthIDTextWatermarkingConfig( ... keys=[654, 400, 836, 123, 340, 443, 597, 160, 57], ... ngram_len=5, ... ) >>> # Generation with watermarking >>> tokenized_prompts = tokenizer(["Once upon a time, "], return_tensors="pt", padding=True) >>> output_sequences = model.generate( ... **tokenized_prompts, watermarking_config=watermarking_config, do_sample=True, max_new_tokens=10 ... ) >>> watermarked_text = tokenizer.batch_decode(output_sequences, skip_special_tokens=True) ``` """ def __init__( self, ngram_len: int, keys: list[int], context_history_size: int = 1024, sampling_table_seed: int = 0, sampling_table_size: int = 2**16, skip_first_ngram_calls: bool = False, debug_mode: bool = False, ): self.ngram_len = ngram_len self.keys = keys self.sampling_table_size = sampling_table_size self.sampling_table_seed = sampling_table_seed self.context_history_size = context_history_size self.skip_first_ngram_calls = skip_first_ngram_calls self.debug_mode = debug_mode def validate(self): watermark_missing_arg_msg = ( "Some of the keys in `watermarking_config` are defined incorrectly. `{key}` should be {correct_value}` " "but found {found_value}" ) if self.sampling_table_size > 2**24: raise ValueError( watermark_missing_arg_msg.format( key="sampling_table_size", correct_value="< 2**24", found_value=self.sampling_table_size, ), ) def construct_processor(self, vocab_size: int, device) -> "WatermarkLogitsProcessor": return SynthIDTextWatermarkLogitsProcessor( ngram_len=self.ngram_len, keys=self.keys, sampling_table_size=self.sampling_table_size, sampling_table_seed=self.sampling_table_seed, context_history_size=self.context_history_size, device=device, skip_first_ngram_calls=self.skip_first_ngram_calls, debug_mode=self.debug_mode, ) @dataclass class CompileConfig: """ Class that holds arguments relative to `torch.compile` behavior, when using automatic compilation in `generate`. See [`torch.compile`](https://pytorch.org/docs/stable/generated/torch.compile.html) for more details on the arguments. Args: fullgraph (`bool`, *optional*, defaults to `False`): If False (default), attempts to discover compilable regions that will be optimized. If True, then require that the entire function be capturable into a single graph. If this is not possible (that is, if there are graph breaks), then an error will be raised. dynamic (`bool` or `None`, *optional*): Whether to try to use dynamic shape graphs. backend (`str` or `Callable`, *optional*, defaults to `"inductor"`): Backend to be used. mode (`str`, *optional*, defaults to `"reduce-overhead"`): Controls balance between performance and overhead. options (`dict`, *optional*): A dictionary of options to pass to the backend. Examples: ```python >>> from transformers import AutoModelForCausalLM, AutoTokenizer, CompileConfig >>> tokenizer = AutoTokenizer.from_pretrained('google/gemma-2-2b') >>> model = AutoModelForCausalLM.from_pretrained('google/gemma-2-2b').cuda() >>> # Automatic compile configuration, used with static cache >>> compile_config = CompileConfig(dynamic=True) >>> # Generation with static cache and compile config >>> input = tokenizer.encode("Hello there, how", return_tensors="pt").cuda() >>> output = model.generate( ... input, do_sample=False, max_new_tokens=300, cache_implementation="static", compile_config=compile_config ... ) >>> output_text = tokenizer.batch_decode(output, skip_special_tokens=True)[0] ``` """ fullgraph: bool = False dynamic: bool | None = None backend: str | Callable = "inductor" mode: str = "reduce-overhead" options: dict | None = None # Used to flag our `generate` call to compile on e.g. CPU. Often not optimal, but useful for testing purposes. _compile_all_devices = None def to_dict(self) -> dict[str, Any]: """Serializes this instance to a Python dictionary.""" return copy.deepcopy({key: value for key, value in self.__dict__.items() if key != "_compile_all_devices"}) # TODO: add the @strict decorator to prevent attributes passed as args rather than kwargs @dataclass class ContinuousBatchingConfig: """ Class that holds arguments relative to continuous batching, when using continuous batching through the `generate_batch` method or the `continuous_batching_context_manager` context manager. Args: block_size (`int`, *optional*, defaults to 256): Size of each KV cache block in tokens. num_blocks (`int`, *optional*): Number of blocks in the KV cache. Auto-inferred from GPU memory when `None`. max_batch_tokens (`int`, *optional*): Maximum number of tokens in a batch. Auto-inferred from GPU memory when `None`. max_memory_percent (`float`, *optional*, defaults to 0.8): Maximum percentage of free GPU memory (after the model is loaded) to use for the KV cache. max_blocks_per_request (`int`, *optional*, defaults to 0): Maximum blocks per request, used in the `flash_attn_with_kvcache` fast decode path to dimension the block table. Setting this to 0 disables the fast decode path. allow_block_sharing (`bool`, *optional*, defaults to `True`): Whether to allow block sharing for prefix caching. Block sharing can only be allowed, never forced, as some models do not support it. Disable if you have few short prompts but long generation lengths. use_async_batching (`bool`, *optional*): Whether to enable async double-buffering, which removes CPU overhead from the continuous batching loop at the cost of doubled VRAM usage. Auto-detected when `None`. use_cuda_graph (`bool`, *optional*): Whether to enable CUDA graphs. Auto-inferred when `None`. q_padding_interval_size (`int`, *optional*, defaults to 0): Query padding granularity in tokens for CUDA graphs. Uses a preset from `continuous_api.py` when set to 0. kv_padding_interval_size (`int`, *optional*, defaults to 0): KV padding granularity in tokens for CUDA graphs. Uses a preset from `continuous_api.py` when set to 0. max_cached_graphs (`int`, *optional*, defaults to 0): Maximum number of cached CUDA graphs. Uses a preset from `continuous_api.py` when set to 0. varlen_compile_config (`CompileConfig`, *optional*): CompileConfig for varlen (prefill) path. Default is None (uses generation_config fallback) The varlen path handles batches with varying query and KV lengths, often benefiting from dynamic=True. decode_compile_config (`CompileConfig`, *optional*): CompileConfig for decode (fast) path. Default is None (uses generation_config fallback) The decode path handles batches has no dynamic KV length, so static shapes are a better fit. use_default_compile_configs (`bool`, *optional*, defaults to `False`): If True, a default compile config will be used for paths that are not explicitly set. scheduler_type (`str`, *optional*, defaults to `"fifo"`): Scheduler type to use. return_logprobs (`bool`, *optional*, defaults to `False`): Whether to return log probabilities along with the generated tokens. max_queue_size (`int`, *optional*, defaults to 0): Maximum request queue size for serving. 0 means unlimited. """ # Size of each KV cache block block_size: int = 256 # The number of blocks used in the KV cache and the maximum number of tokens in a batch. Once the block size is set, # these can be auto inferred using GPU size. num_blocks: int | None = None max_batch_tokens: int | None = None # The max percentage of free GPU memory (after the model is loaded) to use for the KV cache. max_memory_percent: float = 0.8 # This is only used in the flash_attn_with_kvcache fast decode path to dimension the block table. If it is set to 0, # the fast decode path will not be used. Currently turned off by default. max_blocks_per_request: int | None = 0 # Block sharing can only be allowed, but never forced: some model just do not support it. If you only have a few # short prompts, but long generation lengths, you might want to disable block sharing. allow_block_sharing: bool = True # Enables asynchronous batching. This removes the CPU overhead from the continuous batching loop, at the cost of # doubling the VRAM usage. If None, will be automatically detected. use_async_batching: bool | None = None # If any of these parameters are set to a non-default, CUDA graphs will be used. Otherwise we automatically infer # if they should be turned on. Padding interval sizes are in tokens and further explained in the docstring at the # top of the continuous_batching/continuous_api.py file. use_cuda_graph: bool | None = None q_padding_interval_size: int = 0 kv_padding_interval_size: int = 0 max_cached_graphs: int = 0 # Compile configs for the two execution paths. If None, uses the compile_config from generation_config as fallback. # The varlen path is used for prefill and when fast decode is unavailable. The decode path is used when # max_blocks_per_request > 0 (fast decode with block table). varlen_compile_config: CompileConfig | None = None decode_compile_config: CompileConfig | None = None # If this flag is set to True, a default compile config will be used for paths that are not explicitly set. use_default_compile_configs: bool = False # Scheduler type. FIFO by default. For all types available, checks SCHEDULER_MAPPING in scheduler.py scheduler_type: str = "fifo" # Whether to generate log probabilities, which is the log of the softmax of the processed logits. If True, the log # probabilities will be returned along with the generated tokens in the generation output. return_logprobs: bool = False # The parameters below are mostly useful in the context of serving max_queue_size: int = 0 def account_for_cb_deprecated_arguments( self, max_queue_size: int = 0, q_padding_interval_size: int = 0, kv_padding_interval_size: int = 0, allow_block_sharing: bool = True, use_async_batching: bool | None = None, max_cached_graphs: int = 0, ) -> None: """Some arguments given to `generate_batch`, `init_continuous_batching` or `continuous_batching_context_manager` are now deprecated and are expected inside the continuous batching config. This method checks if any were passed and accounts for them in the continuous batching config. It raises a deprecation warning if any were passed. """ kwargs_to_warn = [] if max_queue_size > 0: kwargs_to_warn.append("max_queue_size") self.max_queue_size = max_queue_size if q_padding_interval_size > 0: kwargs_to_warn.append("q_padding_interval_size") self.q_padding_interval_size = q_padding_interval_size if kv_padding_interval_size > 0: kwargs_to_warn.append("kv_padding_interval_size") self.kv_padding_interval_size = kv_padding_interval_size if not allow_block_sharing: # config default is True, so False means the user explicitly set it to False kwargs_to_warn.append("allow_block_sharing") self.allow_block_sharing = allow_block_sharing if use_async_batching is not None: kwargs_to_warn.append("use_async_batching") self.use_async_batching = use_async_batching if max_cached_graphs > 0: kwargs_to_warn.append("max_cached_graphs") self.max_cached_graphs = max_cached_graphs if kwargs_to_warn: logger.warning( "The following arguments were provided to a continuous batching entry point instead of being passed " "through the continuous_batching_config: " + ", ".join(kwargs_to_warn) ) def decide_use_cuda_graphs(self, compile_config: CompileConfig | None, is_attn_mask_needed: bool) -> bool: """Returns whether or not to use cuda graphs for continuous batching. If the user specified this in the config or if they specified a parameter related to cuda graphs, they are turned on. Otherwise, we use a heuristic based on the attention implementation: we turn on cuda graphs if and only if no attention mask is needed. This function modifies the `use_cuda_graph` attribute of the config in place. """ # If cuda is not available, we cannot use cuda graphs import torch if not torch.cuda.is_available(): if self.use_cuda_graph: # throw a warning only if the user intended to use cuda graphs logger.warning(f"use_cuda_graph is True but {torch.cuda.is_available() = }: turning off cuda graphs.") self.use_cuda_graph = False # Else if use_cuda_graph is specified, we follow the user's choice elif self.use_cuda_graph is not None: pass # nothing to do but catch this case and wait for the function to return later # Else if the user specified a parameter related to cuda graphs, we activate cuda graphs elif self.q_padding_interval_size or self.kv_padding_interval_size or self.max_cached_graphs: self.use_cuda_graph = True # Else if a compile config was found, turn off cuda graphs if the compile config already uses them elif compile_config is not None: options = torch._inductor.list_mode_options().get(compile_config.mode, compile_config.options) compile_uses_cudagraphs = options.get("triton.cudagraphs", False) if compile_uses_cudagraphs: logger.warning( f"Compile config {compile_config.mode = } uses cudagraphs, which usually does not work well with " "continuous batching. We recommend using mode 'default' or 'max-autotune-no-cudagraphs' instead." ) self.use_cuda_graph = not compile_uses_cudagraphs # TODO: should this also match the dynamic shapes? # Otherwise we have a default heuristic based on the attention implementation: # attention implementations where an attention mask is needed suffer a lot more from the padding associated # with cuda graphs, so default is to turn cuda graphs off for those implementations else: self.use_cuda_graph = not is_attn_mask_needed logger.warning( f"No behavior specified for use_cuda_graph, defaulting to {self.use_cuda_graph = } because " f"{is_attn_mask_needed = }. If you want to save memory, turn off cuda graphs, but they tend to improve " "performances by a lot." ) # Return the decision return self.use_cuda_graph def decide_use_async_batching(self, is_attn_mask_needed: bool) -> bool: """Returns whether or not to use asynchronous batching for continuous batching. If the user specified this in the config, we follow their choice. Otherwise, we turn on asynchronous batching if and only if CUDA graphs are turned on and no attention mask is needed. This function modifies the `use_async_batching` attribute of the config in place. """ # If the user specifies to use async or not, no need to decide ourselves if self.use_async_batching is None: self.use_async_batching = self.use_cuda_graph and not is_attn_mask_needed logger.info( f"No behavior specified for use_async_batching, choosing {self.use_async_batching = } because " f"{self.use_cuda_graph = } and {not is_attn_mask_needed = }. If you want to save memory, you can " "disable asynchronous batching but it will degrade performance." ) return self.use_async_batching def resolve_sentinel_values(self) -> None: """For some parameters (padding intervals and max cached graphs), the default is a sentinel value of 0: that way, if the user specifies a value for those parameters, we know they want it used, ie. we turn on cuda graphs. But in the case the user does not specify those values, we still need them to resolve to a non-zero value. This function takes care of that.""" # Interval sizes are in tokens for both Q and KV if self.q_padding_interval_size == 0: self.q_padding_interval_size = 64 if self.kv_padding_interval_size == 0: self.kv_padding_interval_size = 64 * 256 # 64 blocks of 256 tokens ie. 16384 tokens if self.max_cached_graphs == 0: self.max_cached_graphs = 32 def resolve_compile_configs( self, fallback_compile_config: CompileConfig | None, is_flash_attn: bool, decode_fast_path_available: bool ) -> None: """Resolve if the compile configs for varlen and decode paths, modifying these attributes in place if needed.""" logger_ = logging.get_logger("ContinuousBatchingLogger") # For each config, priority is: explicit config, default config, fallback config, None if self.varlen_compile_config is None: if self.use_default_compile_configs: # Flash does not support fullgraph but other (sdpa and eager) do fullgraph = not is_flash_attn varlen_config = CompileConfig(mode="max-autotune-no-cudagraphs", fullgraph=fullgraph, dynamic=True) elif fallback_compile_config is not None: varlen_config = fallback_compile_config else: varlen_config = None else: varlen_config = self.varlen_compile_config if self.decode_compile_config is None: if self.use_default_compile_configs: # Paged attention is wrapped in @torch.compiler.disable so we can't use fullgraph decode_config = CompileConfig(mode="max-autotune-no-cudagraphs", fullgraph=False, dynamic=False) elif fallback_compile_config is not None: decode_config = fallback_compile_config else: decode_config = None else: decode_config = self.decode_compile_config # For decode, we throw a warning if the fast decode path is not available and a compile config was found if not decode_fast_path_available and self.decode_compile_config is not None: decode_config = None logger_.warning("A decode_compile_config was set but fast decode path is not available. Ignoring it.") # Log what will be compiled if varlen_config is not None: logger_.info(f"Varlen path will be compiled with {varlen_config.to_dict()}") if decode_config is not None: logger_.info(f"Decode path will be compiled with {decode_config.to_dict()}") # Modify in place self.varlen_compile_config = varlen_config self.decode_compile_config = decode_config