Skip to content

GitLab

"docs/source/en/sagemaker.md" did not exist on "b189226e8cb31361916d638c5a1e2540e3208a61"
Unverified Commit 588faad1 authored by Joao Gante's avatar Joao Gante Committed by GitHub
Browse files

Generate: TF XLA beam sample (#20927) 

* beam sample in beam search

* rag now works with the updated beam search

* delete legacy (non-XLA) generation code related to beam sample
parent 375801d5
Show whitespace changes
Inline Side-by-side
Showing with 163 additions and 1160 deletions
src/transformers/generation/tf_utils.py
View file @ 588faad1
......@@ -435,956 +435,19 @@ class TFGenerationMixin:
`top_k>1`
- *multinomial sampling* by calling [`~generation.TFGenerationMixin.sample`] if `num_beams=1` and
`do_sample=True`.
- *beam-search decoding* by calling [`~generation.TFGenerationMixin.beam_search`] if `num_beams>1` and
`do_sample=False`.
"""
_seed_generator = None
@property
def seed_generator(self):
warnings.warn("`seed_generator` is deprecated and will be removed in a future version.", UserWarning)
if self._seed_generator is None:
self._seed_generator = tf.random.Generator.from_non_deterministic_state()
return self._seed_generator
supports_xla_generation = True
def _use_cache(self, outputs, use_cache):
"""During generation, decide whether to pass the `past` variable to the next forward pass."""
use_cache = getattr(self.config, "use_cache", False)
if len(outputs) <= 1 or use_cache is False:
return False
if hasattr(self.config, "mem_len") and self.config.mem_len == 0:
return False
return True
def generate(
self,
input_ids=None,
max_length=None,
max_new_tokens=None,
min_length=None,
do_sample=None,
early_stopping=None,
num_beams=None,
temperature=None,
penalty_alpha=None,
top_k=None,
top_p=None,
repetition_penalty=None,
bad_words_ids=None,
bos_token_id=None,
pad_token_id=None,
eos_token_id=None,
length_penalty=None,
no_repeat_ngram_size=None,
num_return_sequences=None,
attention_mask=None,
decoder_start_token_id=None,
use_cache=None,
output_scores=None,
output_attentions=None,
output_hidden_states=None,
return_dict_in_generate=None,
forced_bos_token_id=None,
forced_eos_token_id=None,
suppress_tokens: Optional[List[int]] = None,
begin_suppress_tokens: Optional[List[int]] = None,
forced_decoder_ids: Optional[List[List[int]]] = None,
**model_kwargs,
) -> Union[TFGenerateOutput, tf.Tensor]:
r"""
Generates sequences of token ids for models with a language modeling head. The method supports the following
generation methods for text-decoder, text-to-text, speech-to-text, and vision-to-text models:
- *greedy decoding* by calling [`~generation.TFGenerationMixin.greedy_search`] if `num_beams=1` and
`do_sample=False`.
- *contrastive search* by calling [`~generation.TFGenerationMixin.contrastive_search`] if `penalty_alpha>0`
and `top_k>1`
- *multinomial sampling* by calling [`~generation.TFGenerationMixin.sample`] if `num_beams=1` and
`do_sample=True`.
- *beam-search decoding* by calling [`~generation.TFGenerationMixin.beam_search`] if `num_beams>1` and
`do_sample=False`.
Adapted in part from [Facebook's XLM beam search
code](https://github.com/facebookresearch/XLM/blob/9e6f6814d17be4fe5b15f2e6c43eb2b2d76daeb4/src/model/transformer.py#L529).
Apart from `input_ids` and `attention_mask`, all the arguments below will default to the value of the attribute
of the same name inside the [`PretrainedConfig`] of the model. The default values indicated are the default
values of those config.
Most of these parameters are explained in more detail in [this blog
post](https://huggingface.co/blog/how-to-generate).
Parameters:
input_ids (`tf.Tensor` of shape `(batch_size, sequence_length)`, `(batch_size, sequence_length,
feature_dim)` or `(batch_size, num_channels, height, width)`, *optional*):
The sequence used as a prompt for the generation or as model inputs to the encoder. If `None` the
method initializes it with `bos_token_id` and a batch size of 1. For decoder-only models `inputs`
should of in the format of `input_ids`. For encoder-decoder models *inputs* can represent any of
`input_ids`, `input_values`, `input_features`, or `pixel_values`.
max_length (`int`, *optional*, defaults to `model.config.max_length`):
The maximum length the generated tokens can have. Corresponds to the length of the input prompt +
`max_new_tokens`. In general, prefer the use of `max_new_tokens`, which ignores the number of tokens in
the prompt.
max_new_tokens (`int`, *optional*):
The maximum numbers of tokens to generate, ignoring the number of tokens in the prompt.
min_length (`int`, *optional*, defaults to 10):
The minimum length of the sequence to be generated.
do_sample (`bool`, *optional*, defaults to `False`):
Whether or not to use sampling ; use greedy decoding otherwise.
early_stopping (`bool`, *optional*, defaults to `False`):
Whether to stop the beam search when at least `num_beams` sentences are finished per batch or not.
num_beams (`int`, *optional*, defaults to 1):
Number of beams for beam search. 1 means no beam search.
temperature (`float`, *optional*, defaults to 1.0):
The value used to module the next token probabilities.
penalty_alpha (`float`, *optional*):
The values balance the model confidence and the degeneration penalty in contrastive search decoding.
top_k (`int`, *optional*, defaults to 50):
The number of highest probability vocabulary tokens to keep for top-k-filtering.
top_p (`float`, *optional*, defaults to 1.0):
If set to float < 1, only the most probable tokens with probabilities that add up to `top_p` or higher
are kept for generation.
repetition_penalty (`float`, *optional*, defaults to 1.0):
The parameter for repetition penalty. 1.0 means no penalty. See [this
paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
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 (`int`, *optional*):
The id of the *end-of-sequence* token.
length_penalty (`float`, *optional*, defaults to 1.0):
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*, defaults to 0):
If set to int > 0, all ngrams of that size can only occur once.
bad_words_ids(`List[int]`, *optional*):
List of token ids that are not allowed to be generated. In order to get the tokens of the words that
should not appear in the generated text, use `tokenizer.encode(bad_word, add_prefix_space=True)`.
num_return_sequences(`int`, *optional*, defaults to 1):
The number of independently computed returned sequences for each element in the batch.
attention_mask (`tf.Tensor` of `dtype=tf.int32` and shape `(batch_size, sequence_length)`, *optional*):
Mask to avoid performing attention on padding token indices. Mask values are in `[0, 1]`, 1 for tokens
that are not masked, and 0 for masked tokens.
If not provided, will default to a tensor the same shape as `input_ids` that masks the pad token.
[What are attention masks?](../glossary#attention-mask)
decoder_start_token_id (`int`, *optional*):
If an encoder-decoder model starts decoding with a different token than *bos*, the id of that token.
use_cache (`bool`, *optional*, defaults to `True`):
Whether or not the model should use the past last key/values attentions (if applicable to the model) to
speed up decoding.
output_attentions (`bool`, *optional*, defaults to `False`):
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
returned tensors for more details.
output_hidden_states (`bool`, *optional*, defaults to `False`):
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
for more details.
output_scores (`bool`, *optional*, defaults to `False`):
Whether or not to return the prediction scores. See `scores` under returned tensors for more details.
return_dict_in_generate (`bool`, *optional*, defaults to `False`):
Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
forced_bos_token_id (`int`, *optional*):
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`, *optional*):
The id of the token to force as the last generated token when `max_length` is reached.
suppress_tokens (`List[int]`, *optional*, defaults to `model.config.suppress_tokens`):
A list of tokens that will be supressed at generation. The `SupressTokens` logit processor will set
their log probs to `-inf` so that they are not sampled.
begin_suppress_tokens (`List[int]`, *optional*, defaults to `model.config.begin_suppress_tokens`):
A list of tokens that will be supressed at the begining of the generation. The `SupressBeginTokens`
logit processor will set their log probs to `-inf` so that they are not sampled.
forced_decoder_ids (`List[List[int]]`, *optional*, defaults to `model.config.forced_decoder_ids`):
A list of pairs of integers which indicates a mapping from generation indices to token indices that
will be forced before sampling. For example, `[[1, 123]]` means the second generated token will always
be a token of index 123.
model_specific_kwargs:
Additional model specific kwargs will be forwarded to the `forward` function of the model.
Return:
[`~utils.ModelOutput`] or `tf.Tensor`: A [`~utils.ModelOutput`] (if `return_dict_in_generate=True` or when
`config.return_dict_in_generate=True`) or a `tf.Tensor`.
If the model is *not* an encoder-decoder model (`model.config.is_encoder_decoder=False`), the possible
[`~utils.ModelOutput`] types are:
- [`~generation.TFGreedySearchDecoderOnlyOutput`],
- [`~generation.TFSampleDecoderOnlyOutput`],
- [`~generation.TFBeamSearchDecoderOnlyOutput`],
- [`~generation.TFBeamSampleDecoderOnlyOutput`]
If the model is an encoder-decoder model (`model.config.is_encoder_decoder=True`), the possible
[`~utils.ModelOutput`] types are:
- [`~generation.TFGreedySearchEncoderDecoderOutput`],
- [`~generation.TFSampleEncoderDecoderOutput`],
- [`~generation.TFBeamSearchEncoderDecoderOutput`],
- [`~generation.TFBeamSampleEncoderDecoderOutput`]
Examples:
```python
tokenizer = AutoTokenizer.from_pretrained("distilgpt2") # Initialize tokenizer
model = TFAutoModelWithLMHead.from_pretrained(
"distilgpt2"
) # Download model and configuration from huggingface.co and cache.
outputs = model.generate(max_length=40) # do greedy decoding
print(f"Generated: {tokenizer.decode(outputs[0], skip_special_tokens=True)}")
tokenizer = AutoTokenizer.from_pretrained("openai-gpt") # Initialize tokenizer
model = TFAutoModelWithLMHead.from_pretrained(
"openai-gpt"
) # Download model and configuration from huggingface.co and cache.
input_context = "The dog"
input_ids = tokenizer.encode(input_context, return_tensors="tf") # encode input context
outputs = model.generate(
input_ids=input_ids, num_beams=5, num_return_sequences=3, temperature=1.5
) # generate 3 independent sequences using beam search decoding (5 beams) with sampling from initial context 'The dog'
for i in range(3): # 3 output sequences were generated
print(f"Generated {i}: {tokenizer.decode(outputs[i], skip_special_tokens=True)}")
tokenizer = AutoTokenizer.from_pretrained("distilgpt2") # Initialize tokenizer
model = TFAutoModelWithLMHead.from_pretrained(
"distilgpt2"
) # Download model and configuration from huggingface.co and cache.
input_context = "The dog"
input_ids = tokenizer.encode(input_context, return_tensors="tf") # encode input context
outputs = model.generate(
input_ids=input_ids, max_length=40, temperature=0.7, num_return_sequences=3, do_sample=True
) # generate 3 candidates using sampling
for i in range(3): # 3 output sequences were generated
print(f"Generated {i}: {tokenizer.decode(outputs[i], skip_special_tokens=True)}")
tokenizer = AutoTokenizer.from_pretrained("ctrl") # Initialize tokenizer
model = TFAutoModelWithLMHead.from_pretrained(
"ctrl"
) # Download model and configuration from huggingface.co and cache.
input_context = "Legal My neighbor is" # "Legal" is one of the control codes for ctrl
input_ids = tokenizer.encode(input_context, return_tensors="tf") # encode input context
outputs = model.generate(
input_ids=input_ids, max_length=50, temperature=0.7, repetition_penalty=1.2
) # generate sequences
print(f"Generated: {tokenizer.decode(outputs[0], skip_special_tokens=True)}")
tokenizer = AutoTokenizer.from_pretrained("gpt2") # Initialize tokenizer
model = TFAutoModelWithLMHead.from_pretrained(
"gpt2"
) # Download model and configuration from huggingface.co and cache.
input_context = "My cute dog"
bad_words_ids = [
tokenizer.encode(bad_word, add_prefix_space=True) for bad_word in ["idiot", "stupid", "shut up"]
]
input_ids = tokenizer.encode(input_context, return_tensors="tf") # encode input context
outputs = model.generate(
input_ids=input_ids, max_length=100, do_sample=True, bad_words_ids=bad_words_ids
) # generate sequences without allowing bad_words to be generated
```"""
num_beams = num_beams if num_beams is not None else self.config.num_beams
do_sample = do_sample if do_sample is not None else self.config.do_sample
if do_sample is False or num_beams == 1:
seed = model_kwargs.pop("seed", None)
return self._generate(
input_ids=input_ids,
max_length=max_length,
max_new_tokens=max_new_tokens,
min_length=min_length,
do_sample=do_sample,
early_stopping=early_stopping,
num_beams=num_beams,
temperature=temperature,
penalty_alpha=penalty_alpha,
top_k=top_k,
top_p=top_p,
repetition_penalty=repetition_penalty,
bad_words_ids=bad_words_ids,
bos_token_id=bos_token_id,
pad_token_id=pad_token_id,
eos_token_id=eos_token_id,
length_penalty=length_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
num_return_sequences=num_return_sequences,
attention_mask=attention_mask,
decoder_start_token_id=decoder_start_token_id,
use_cache=use_cache,
seed=seed,
output_scores=output_scores,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict_in_generate=return_dict_in_generate,
forced_bos_token_id=forced_bos_token_id,
forced_eos_token_id=forced_eos_token_id,
suppress_tokens=suppress_tokens,
begin_suppress_tokens=begin_suppress_tokens,
forced_decoder_ids=forced_decoder_ids,
**model_kwargs,
)
# We cannot generate if the model does not have a LM head
if self.get_output_embeddings() is None:
raise AttributeError(
"You tried to generate sequences with a model that does not have a LM Head. Please use another model"
" class (e.g. `TFOpenAIGPTLMHeadModel`, `TFXLNetLMHeadModel`, `TFGPT2LMHeadModel`,"
" `TFCTRLLMHeadModel`, `TFT5ForConditionalGeneration`, `TFTransfoXLLMHeadModel`)"
)
max_length = max_length if max_length is not None else self.config.max_length
min_length = min_length if min_length is not None else self.config.min_length
early_stopping = early_stopping if early_stopping is not None else self.config.early_stopping
temperature = temperature if temperature is not None else self.config.temperature
top_k = top_k if top_k is not None else self.config.top_k
top_p = top_p if top_p is not None else self.config.top_p
repetition_penalty = repetition_penalty if repetition_penalty is not None else self.config.repetition_penalty
bos_token_id = bos_token_id if bos_token_id is not None else self.config.bos_token_id
pad_token_id = pad_token_id if pad_token_id is not None else self.config.pad_token_id
eos_token_id = eos_token_id if eos_token_id is not None else self.config.eos_token_id
length_penalty = length_penalty if length_penalty is not None else self.config.length_penalty
no_repeat_ngram_size = (
no_repeat_ngram_size if no_repeat_ngram_size is not None else self.config.no_repeat_ngram_size
)
bad_words_ids = bad_words_ids if bad_words_ids is not None else self.config.bad_words_ids
num_return_sequences = (
num_return_sequences if num_return_sequences is not None else self.config.num_return_sequences
)
decoder_start_token_id = (
decoder_start_token_id if decoder_start_token_id is not None else self.config.decoder_start_token_id
)
forced_bos_token_id = (
forced_bos_token_id if forced_bos_token_id is not None else self.config.forced_bos_token_id
)
forced_eos_token_id = (
forced_eos_token_id if forced_eos_token_id is not None else self.config.forced_eos_token_id
)
suppress_tokens = suppress_tokens if suppress_tokens is not None else self.config.suppress_tokens
begin_suppress_tokens = (
begin_suppress_tokens if begin_suppress_tokens is not None else self.config.begin_suppress_tokens
)
if forced_decoder_ids is None and hasattr(self.config, "forced_decoder_ids"):
forced_decoder_ids = self.config.forced_decoder_ids
output_scores = output_scores if output_scores is not None else self.config.output_scores
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
return_dict_in_generate = (
return_dict_in_generate if return_dict_in_generate is not None else self.config.return_dict_in_generate
)
model_kwargs["output_scores"] = output_scores
model_kwargs["output_attentions"] = output_attentions
model_kwargs["output_hidden_states"] = output_hidden_states
if self.config.is_encoder_decoder:
model_kwargs["encoder_attentions"] = None
model_kwargs["encoder_hidden_states"] = None
if input_ids is not None:
batch_size = shape_list(input_ids)[0] # overridden by the input batch_size
else:
batch_size = 1
assert isinstance(max_length, int) and max_length > 0, "`max_length` should be a strictly positive integer."
assert isinstance(min_length, int) and min_length >= 0, "`min_length` should be a positive integer."
assert isinstance(do_sample, bool), "`do_sample` should be a boolean."
assert isinstance(early_stopping, bool), "`early_stopping` should be a boolean."
assert isinstance(num_beams, int) and num_beams > 0, "`num_beams` should be a strictly positive integer."
assert temperature > 0, "`temperature` should be strictly positive."
assert isinstance(top_k, int) and top_k >= 0, "`top_k` should be a positive integer."
assert 0 <= top_p <= 1, "`top_p` should be between 0 and 1."
assert repetition_penalty >= 1.0, "`repetition_penalty` should be >= 1."
assert input_ids is not None or (
isinstance(bos_token_id, int) and bos_token_id >= 0
), "If input_ids is not defined, `bos_token_id` should be a positive integer."
assert pad_token_id is None or (
isinstance(pad_token_id, int) and (pad_token_id >= 0)
), "`pad_token_id` should be a positive integer."
assert (eos_token_id is None) or (
isinstance(eos_token_id, int) and (eos_token_id >= 0)
), "`eos_token_id` should be a positive integer."
assert length_penalty > 0, "`length_penalty` should be strictly positive."
assert (
isinstance(num_return_sequences, int) and num_return_sequences > 0
), "`num_return_sequences` should be a strictly positive integer."
assert (
bad_words_ids is None or isinstance(bad_words_ids, list) and isinstance(bad_words_ids[0], list)
), "`bad_words_ids` is either `None` or a list of lists of tokens that should not be generated"
# This block corresponds to the following line in `generation`:
# "input_ids = self._prepare_input_ids_for_generation(bos_token_id, model_kwargs.get("encoder_outputs"))"
# with the following differences:
# 1. In PT, `generate()`'s `model_kwargs` can accept `encoder_outputs`, but not the case in TF.
# 2. There is no shape checking in PT.
# In both PT/TF, if `input_ids` is `None`, we try to create it as it is for a text model.
if input_ids is None:
assert isinstance(bos_token_id, int) and bos_token_id >= 0, (
"you should either supply a context to complete as `input_ids` input "
"or a `bos_token_id` (integer >= 0) as a first token to start the generation."
)
input_ids = tf.fill((batch_size, 1), bos_token_id)
# not allow to duplicate outputs when greedy decoding
if do_sample is False:
if num_beams == 1:
# no_beam_search greedy generation conditions
assert num_return_sequences == 1, (
"Greedy decoding will always produce the same output for num_beams == 1 and num_return_sequences >"
" 1. Please set num_return_sequences = 1"
)
else:
# beam_search greedy generation conditions
assert num_beams >= num_return_sequences, (
"Greedy beam search decoding cannot return more sequences than it has beams. Please set num_beams"
" >= num_return_sequences"
)
# create attention mask if necessary
accepts_attention_mask = "attention_mask" in set(inspect.signature(self.call).parameters.keys())
if accepts_attention_mask:
if (attention_mask is None) and (pad_token_id is not None) and (pad_token_id in input_ids.numpy()):
attention_mask = tf.cast(tf.math.not_equal(input_ids, pad_token_id), dtype=tf.int32)
elif attention_mask is None:
attention_mask = tf.ones(shape_list(input_ids)[:2], dtype=tf.int32)
if pad_token_id is None and eos_token_id is not None:
logger.warning(f"Setting `pad_token_id` to {eos_token_id} (first `eos_token_id`) to generate sequence")
pad_token_id = eos_token_id
# current position and vocab size
cur_len = shape_list(input_ids)[1] # unused
vocab_size = getattr(self.config, "vocab_size", None)
if vocab_size is None and self.config.is_encoder_decoder:
decoder_config = getattr(self.config, "decoder", None)
if decoder_config is not None:
vocab_size = getattr(self.config.decoder, "vocab_size", None)
# set effective batch size and effective batch multiplier according to do_sample
if do_sample:
effective_batch_size = batch_size * num_return_sequences
effective_batch_mult = num_return_sequences
else:
effective_batch_size = batch_size
effective_batch_mult = 1
if self.config.is_encoder_decoder:
if decoder_start_token_id is None:
decoder_start_token_id = bos_token_id
assert (
decoder_start_token_id is not None
), "decoder_start_token_id or bos_token_id has to be defined for encoder-decoder generation"
assert hasattr(self, "get_encoder"), f"{self} should have a 'get_encoder' function defined"
assert callable(self.get_encoder), f"{self.get_encoder} should be a method"
# get encoder and store encoder outputs
encoder = self.get_encoder()
encoder_kwargs = {
"output_attentions": output_attentions,
"output_hidden_states": output_hidden_states,
"return_dict": return_dict_in_generate,
}
if accepts_attention_mask:
encoder_kwargs["attention_mask"] = attention_mask
encoder_outputs = encoder(input_ids, **encoder_kwargs)
if return_dict_in_generate:
if output_attentions:
model_kwargs["encoder_attentions"] = encoder_outputs.attentions
if output_hidden_states:
model_kwargs["encoder_hidden_states"] = encoder_outputs.hidden_states
expanded_batch_idxs = tf.reshape(
tf.repeat(tf.expand_dims(tf.range(batch_size), -1), repeats=num_beams * effective_batch_mult, axis=1),
shape=(-1,),
)
# prepares text-based inputs
if len(shape_list(input_ids)) == 2:
input_ids = tf.gather(input_ids, expanded_batch_idxs, axis=0)
if accepts_attention_mask:
attention_mask = tf.gather(attention_mask, expanded_batch_idxs, axis=0)
if self.config.is_encoder_decoder:
# create empty decoder_input_ids
input_ids = (
tf.ones(
(effective_batch_size * num_beams, 1),
dtype=tf.int32,
)
* decoder_start_token_id
)
cur_len = 1
assert (
batch_size == encoder_outputs[0].shape[0]
), f"expected encoder_outputs[0] to have 1st dimension bs={batch_size}, got {encoder_outputs[0].shape[0]} "
# expand encoder_outputs
encoder_outputs = (tf.gather(encoder_outputs[0], expanded_batch_idxs, axis=0),)
else:
encoder_outputs = None
cur_len = shape_list(input_ids)[-1]
assert cur_len < max_length, (
f"The context has {cur_len} number of tokens, but `max_length` is only {max_length}. Please make sure that"
" `max_length` is bigger than the number of tokens, by setting either `generate(max_length=...,...)` or"
" `config.max_length = ...`"
)
return self._generate_beam_search(
input_ids,
cur_len=cur_len,
max_length=max_length,
min_length=min_length,
do_sample=do_sample,
early_stopping=early_stopping,
temperature=temperature,
top_k=top_k,
top_p=top_p,
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
bad_words_ids=bad_words_ids,
pad_token_id=pad_token_id,
eos_token_id=eos_token_id,
batch_size=effective_batch_size,
num_return_sequences=num_return_sequences,
length_penalty=length_penalty,
num_beams=num_beams,
vocab_size=vocab_size,
encoder_outputs=encoder_outputs,
attention_mask=attention_mask,
use_cache=use_cache,
forced_bos_token_id=forced_bos_token_id,
forced_eos_token_id=forced_eos_token_id,
return_dict_in_generate=return_dict_in_generate,
**model_kwargs,
)
def _generate_beam_search(
self,
input_ids,
cur_len,
max_length,
min_length,
do_sample,
early_stopping,
temperature,
top_k,
top_p,
repetition_penalty,
no_repeat_ngram_size,
bad_words_ids,
pad_token_id,
eos_token_id,
batch_size,
num_return_sequences,
length_penalty,
num_beams,
vocab_size,
encoder_outputs,
attention_mask,
use_cache,
forced_bos_token_id,
forced_eos_token_id,
return_dict_in_generate,
**kwargs,
) -> Union[TFBeamSearchOutput, TFBeamSampleOutput, tf.Tensor]:
"""Generate sequences for each example with beam search."""
# generated hypotheses
generated_hyps = [
BeamHypotheses(num_beams, max_length, length_penalty, early_stopping=early_stopping)
for _ in range(batch_size)
]
# for greedy decoding it is made sure that only tokens of the first beam are considered to avoid sampling the exact same tokens three times
if do_sample is False:
beam_scores_begin = tf.zeros((batch_size, 1), dtype=tf.float32)
beam_scores_end = tf.ones((batch_size, num_beams - 1), dtype=tf.float32) * (-1e9)
beam_scores = tf.concat([beam_scores_begin, beam_scores_end], -1)
else:
beam_scores = tf.zeros((batch_size, num_beams), dtype=tf.float32)
beam_scores = tf.reshape(beam_scores, (batch_size * num_beams,))
# variable to cache compute states
past = None
# init attention / hidden states / scores tuples
scores = () if (return_dict_in_generate and kwargs["output_scores"]) else None
decoder_attentions = () if (return_dict_in_generate and kwargs["output_attentions"]) else None
cross_attentions = () if (return_dict_in_generate and kwargs["output_attentions"]) else None
decoder_hidden_states = () if (return_dict_in_generate and kwargs["output_hidden_states"]) else None
# if model is an encoder-decoder, retrieve encoder attention weights and hidden states
if self.config.is_encoder_decoder:
encoder_attentions = (
kwargs["encoder_attentions"] if (return_dict_in_generate and kwargs["encoder_attentions"]) else None
)
encoder_hidden_states = (
kwargs["encoder_hidden_states"]
if (return_dict_in_generate and kwargs["encoder_hidden_states"])
else None
)
# the refactored generate, without the encoder outputs in `past`, expects the `encoder_outputs`
# variable to contain all (encoder_outputs, encoder_hidden_states, encoder_attentions) in
# `prepare_inputs_for_generation`
if encoder_hidden_states is not None:
encoder_outputs = (*encoder_outputs, encoder_hidden_states)
if encoder_attentions is not None:
encoder_outputs = (*encoder_outputs, encoder_attentions)
# done sentences
done = [False for _ in range(batch_size)]
while cur_len < max_length:
model_inputs = self.prepare_inputs_for_generation(
input_ids,
past=past,
attention_mask=attention_mask,
use_cache=use_cache,
encoder_outputs=encoder_outputs,
**kwargs,
)
outputs = self(
**model_inputs,
return_dict=True,
output_attentions=kwargs["output_attentions"],
output_hidden_states=kwargs["output_hidden_states"],
)
next_token_logits = outputs.logits[:, -1, :] # (batch_size * num_beams, vocab_size)
# if model has past, then set the past variable to speed up decoding
if self._use_cache(outputs, use_cache):
past = outputs[1]
# repetition penalty (from CTRL paper https://arxiv.org/abs/1909.05858)
if repetition_penalty != 1.0:
next_token_logits_penalties = _create_next_token_logits_penalties(
input_ids, next_token_logits, repetition_penalty
)
next_token_logits = tf.math.multiply(next_token_logits, next_token_logits_penalties)
# Temperature (higher temperature => more likely to sample low probability tokens)
if temperature != 1.0:
next_token_logits = next_token_logits / temperature
if self.config.is_encoder_decoder and do_sample is False:
next_token_logits = self.adjust_logits_during_generation(
next_token_logits,
cur_len=cur_len,
max_length=max_length,
forced_bos_token_id=forced_bos_token_id,
forced_eos_token_id=forced_eos_token_id,
)
# calculate log softmax score
scores = tf.nn.log_softmax(next_token_logits, axis=-1) # (batch_size * num_beams, vocab_size)
# set eos token prob to zero if min_length is not reached
if eos_token_id is not None and cur_len < min_length:
# create eos_token_id boolean mask
num_batch_hypotheses = batch_size * num_beams
is_token_logit_eos_token = tf.convert_to_tensor(
[True if token == eos_token_id else False for token in range(vocab_size)], dtype=tf.bool
)
eos_token_indices_mask = tf.broadcast_to(is_token_logit_eos_token, [num_batch_hypotheses, vocab_size])
scores = tf.where(eos_token_indices_mask, -float("inf"), scores)
if no_repeat_ngram_size > 0:
# calculate a list of banned tokens to prevent repetitively generating the same ngrams
# from fairseq: https://github.com/pytorch/fairseq/blob/a07cb6f40480928c9e0548b737aadd36ee66ac76/fairseq/sequence_generator.py#L345
num_batch_hypotheses = batch_size * num_beams
banned_tokens = calc_banned_ngram_tokens(
input_ids, num_batch_hypotheses, no_repeat_ngram_size, cur_len
)
# create banned_tokens boolean mask
banned_tokens_indices_mask = []
for banned_tokens_slice in banned_tokens:
banned_tokens_indices_mask.append(
[True if token in banned_tokens_slice else False for token in range(vocab_size)]
)
scores = tf.where(
tf.convert_to_tensor(banned_tokens_indices_mask, dtype=tf.bool), -float("inf"), scores
)
if bad_words_ids is not None:
# calculate a list of banned tokens according to bad words
banned_tokens = calc_banned_bad_words_ids(input_ids, bad_words_ids)
banned_tokens_indices_mask = []
for banned_tokens_slice in banned_tokens:
banned_tokens_indices_mask.append(
[True if token in banned_tokens_slice else False for token in range(vocab_size)]
)
scores = tf.where(
tf.convert_to_tensor(banned_tokens_indices_mask, dtype=tf.bool), -float("inf"), scores
)
assert shape_list(scores) == [batch_size * num_beams, vocab_size]
if do_sample:
_scores = scores + tf.broadcast_to(
beam_scores[:, None], (batch_size * num_beams, vocab_size)
) # (batch_size * num_beams, vocab_size)
# Top-p/top-k filtering
_scores = tf_top_k_top_p_filtering(
_scores, top_k=top_k, top_p=top_p, min_tokens_to_keep=2
) # (batch_size * num_beams, vocab_size)
# Sample 2 next tokens for each beam (so we have some spare tokens and match output of greedy beam search)
_scores = tf.reshape(_scores, (batch_size, num_beams * vocab_size))
next_tokens = sample_without_replacement(
_scores, num_samples=2 * num_beams
) # (batch_size, 2 * num_beams)
# Compute next scores
next_scores = tf.gather(_scores, next_tokens, batch_dims=1) # (batch_size, 2 * num_beams)
# sort the sampled vector to make sure that the first num_beams samples are the best
next_scores_indices = tf.argsort(next_scores, direction="DESCENDING", axis=1)
next_scores = tf.gather(next_scores, next_scores_indices, batch_dims=1) # (batch_size, num_beams * 2)
next_tokens = tf.gather(next_tokens, next_scores_indices, batch_dims=1) # (batch_size, num_beams * 2)
else:
# Add the log prob of the new beams to the log prob of the beginning of the sequence (sum of logs == log of the product)
next_scores = scores + tf.broadcast_to(
beam_scores[:, None], (batch_size * num_beams, vocab_size)
) # (batch_size * num_beams, vocab_size)
# re-organize to group the beam together (we are keeping top hypothesis across beams)
next_scores = tf.reshape(
next_scores, (batch_size, num_beams * vocab_size)
) # (batch_size, num_beams * vocab_size)
next_scores, next_tokens = tf.math.top_k(next_scores, k=2 * num_beams, sorted=True)
assert shape_list(next_scores) == shape_list(next_tokens) == [batch_size, 2 * num_beams]
# Store scores, attentions and hidden_states when required
if return_dict_in_generate:
if kwargs["output_scores"]:
scores += (next_token_logits,)
if kwargs["output_attentions"]:
decoder_attentions += (
(outputs.decoder_attentions,) if self.config.is_encoder_decoder else (outputs.attentions,)
)
if self.config.is_encoder_decoder:
cross_attentions += (outputs.cross_attentions,)
if kwargs["output_hidden_states"]:
decoder_hidden_states += (
(outputs.decoder_hidden_states,)
if self.config.is_encoder_decoder
else (outputs.hidden_states,)
)
# next batch beam content
next_batch_beam = []
# for each sentence
for batch_idx in range(batch_size):
# if we are done with this sentence
if done[batch_idx]:
assert (
len(generated_hyps[batch_idx]) >= num_beams
), f"Batch can only be done if at least {num_beams} beams have been generated."
assert (
eos_token_id is not None and pad_token_id is not None
), "generated beams >= num_beams -> eos_token_id and pad_token have to be defined"
next_batch_beam.extend([(0, pad_token_id, 0)] * num_beams) # pad the batch
continue
# next sentence beam content
next_sent_beam = []
# next tokens for this sentence
for beam_token_rank, (beam_token_id, beam_token_score) in enumerate(
zip(next_tokens[batch_idx], next_scores[batch_idx])
):
# get beam and token IDs
beam_id = beam_token_id // vocab_size
token_id = beam_token_id % vocab_size
effective_beam_id = batch_idx * num_beams + beam_id
# add to generated hypotheses if end of sentence or last iteration
if (eos_token_id is not None) and (token_id.numpy() == eos_token_id):
# if beam_token does not belong to top num_beams tokens, it should not be added
is_beam_token_worse_than_top_num_beams = beam_token_rank >= num_beams
if is_beam_token_worse_than_top_num_beams:
continue
generated_hyps[batch_idx].add(
tf.identity(input_ids[effective_beam_id]), beam_token_score.numpy()
)
else:
# add next predicted token if it is not eos_token
next_sent_beam.append((beam_token_score, token_id, effective_beam_id))
# the beam for next step is full
if len(next_sent_beam) == num_beams:
break
# Check if we are done so that we can save a pad step if all(done)
done[batch_idx] = done[batch_idx] or generated_hyps[batch_idx].is_done(
tf.reduce_max(next_scores[batch_idx]).numpy(), cur_len
)
# update next beam content
assert len(next_sent_beam) == num_beams, "Beam should always be full"
next_batch_beam.extend(next_sent_beam)
assert len(next_batch_beam) == num_beams * (batch_idx + 1)
# stop when we are done with each sentence
if all(done):
break
# sanity check / prepare next batch
assert len(next_batch_beam) == batch_size * num_beams
beam_scores = tf.convert_to_tensor([x[0] for x in next_batch_beam], dtype=tf.float32)
beam_tokens = tf.convert_to_tensor([x[1] for x in next_batch_beam], dtype=tf.int32)
beam_idx = tf.convert_to_tensor([x[2] for x in next_batch_beam], dtype=tf.int32)
# re-order batch and update current length
input_ids = tf.stack([tf.identity(input_ids[x, :]) for x in beam_idx])
input_ids = tf.concat([input_ids, tf.expand_dims(beam_tokens, 1)], axis=-1)
cur_len = cur_len + 1
# re-order internal states
if past is not None:
past = self._reorder_cache(past, beam_idx)
- *beam-search decoding* by calling [`~generation.TFGenerationMixin.beam_search`] if `num_beams>1`.
"""
# extend attention_mask for new generated input if only decoder
if self.config.is_encoder_decoder is False:
attention_mask = tf.concat(
[attention_mask, tf.ones((shape_list(attention_mask)[0], 1), dtype=tf.int32)], axis=-1
)
_seed_generator = None
# finalize all open beam hypotheses and end to generated hypotheses
for batch_idx in range(batch_size):
# Add all open beam hypothesis to generated_hyps
if done[batch_idx]:
continue
# test that beam scores match previously calculated scores if not eos and batch_idx not done
if eos_token_id is not None and all(
(token_id % vocab_size).numpy().item() != eos_token_id for token_id in next_tokens[batch_idx]
):
if not tf.reduce_all(
next_scores[batch_idx, :num_beams] == tf.reshape(beam_scores, (batch_size, num_beams))[batch_idx]
):
raise ValueError(
f"If batch_idx is not done, final next scores: {next_scores[:, :num_beams][batch_idx]} have "
"to equal to accumulated beam_scores: "
f"{tf.reshape(beam_scores, (batch_size, num_beams))[batch_idx]}"
)
# need to add best num_beams hypotheses to generated hyps
for beam_id in range(num_beams):
effective_beam_id = batch_idx * num_beams + beam_id
final_score = beam_scores[effective_beam_id].numpy().item()
final_tokens = input_ids[effective_beam_id]
generated_hyps[batch_idx].add(final_tokens, final_score)
# depending on whether greedy generation is wanted or not define different output_batch_size and output_num_return_sequences_per_batch
output_batch_size = batch_size if do_sample else batch_size * num_return_sequences
output_num_return_sequences_per_batch = 1 if do_sample else num_return_sequences
# select the best hypotheses
sent_lengths_list = []
best = []
# retrieve best hypotheses
for i, hypotheses in enumerate(generated_hyps):
sorted_hyps = sorted(hypotheses.beams, key=lambda x: x[0])
for j in range(output_num_return_sequences_per_batch):
best_hyp = sorted_hyps.pop()[1]
sent_lengths_list.append(len(best_hyp))
best.append(best_hyp)
assert output_batch_size == len(
best
), f"Output batch size {output_batch_size} must match output beam hypotheses {len(best)}"
sent_lengths = tf.convert_to_tensor(sent_lengths_list, dtype=tf.int32)
# shorter batches are filled with pad_token
if tf.reduce_min(sent_lengths).numpy() != tf.reduce_max(sent_lengths).numpy():
assert pad_token_id is not None, "`Pad_token_id` has to be defined"
sent_max_len = min(tf.reduce_max(sent_lengths).numpy() + 1, max_length)
decoded_list = []
# fill with hypothesis and eos_token_id if necessary
for i, hypo in enumerate(best):
assert sent_lengths[i] == shape_list(hypo)[0]
# if sent_length is max_len do not pad
if sent_lengths[i] == sent_max_len:
decoded_slice = hypo
else:
# else pad to sent_max_len
num_pad_tokens = sent_max_len - sent_lengths[i]
padding = pad_token_id * tf.ones((num_pad_tokens,), dtype=tf.int32)
decoded_slice = tf.concat([hypo, padding], axis=-1)
# finish sentence with EOS token
if sent_lengths[i] < max_length:
decoded_slice = tf.where(
tf.range(sent_max_len, dtype=tf.int32) == sent_lengths[i],
eos_token_id * tf.ones((sent_max_len,), dtype=tf.int32),
decoded_slice,
)
# add to list
decoded_list.append(decoded_slice)
decoded = tf.stack(decoded_list)
else:
# none of the hypotheses have an eos_token
assert (len(hypo) == max_length for hypo in best)
decoded = tf.stack(best)
@property
def seed_generator(self):
warnings.warn("`seed_generator` is deprecated and will be removed in a future version.", UserWarning)
if self._seed_generator is None:
self._seed_generator = tf.random.Generator.from_non_deterministic_state()
return self._seed_generator
if return_dict_in_generate:
if do_sample and self.config.is_encoder_decoder:
return TFBeamSampleEncoderDecoderOutput(
sequences=decoded,
scores=scores,
encoder_attentions=encoder_attentions,
encoder_hidden_states=encoder_hidden_states,
decoder_attentions=decoder_attentions,
cross_attentions=cross_attentions,
decoder_hidden_states=decoder_hidden_states,
)
elif do_sample and not self.config.is_encoder_decoder:
return TFBeamSampleDecoderOnlyOutput(
sequences=decoded,
scores=scores,
attentions=decoder_attentions,
hidden_states=decoder_hidden_states,
)
elif self.config.is_encoder_decoder:
return TFBeamSearchEncoderDecoderOutput(
sequences=decoded,
scores=scores,
encoder_attentions=encoder_attentions,
encoder_hidden_states=encoder_hidden_states,
decoder_attentions=decoder_attentions,
cross_attentions=cross_attentions,
decoder_hidden_states=decoder_hidden_states,
)
else:
return TFBeamSearchDecoderOnlyOutput(
sequences=decoded,
scores=scores,
attentions=decoder_attentions,
hidden_states=decoder_hidden_states,
)
else:
return decoded
supports_xla_generation = True
@staticmethod
def _reorder_cache(past, beam_idx):
......@@ -1459,7 +522,7 @@ class TFGenerationMixin:
" generate arguments will also show up in this list)"
)
def _generate(
def generate(
self,
input_ids=None,
max_length=None,
......@@ -1831,13 +894,13 @@ class TFGenerationMixin:
)
# 7. determine generation mode
# TODO(Matt, Joao, Patrick) - add more use cases here
is_contrastive_search_gen_mode = (
top_k is not None and top_k > 1 and do_sample is False and penalty_alpha is not None and penalty_alpha > 0
)
is_greedy_gen_mode = not is_contrastive_search_gen_mode and (num_beams == 1) and do_sample is False
is_beam_gen_mode = not is_contrastive_search_gen_mode and (num_beams > 1) and do_sample is False
is_sample_gen_mode = (num_beams == 1) and do_sample is True
is_beam_sample_gen_mode = (num_beams > 1) and do_sample is True
# 8. prepare distribution pre_processing samplers
logits_processor = self._get_logits_processor(
......@@ -1919,7 +982,7 @@ class TFGenerationMixin:
elif is_beam_gen_mode:
if num_beams < num_return_sequences:
raise ValueError(
"Greedy beam search decoding cannot return more sequences than it has beams. Please set "
"Beam search decoding cannot return more sequences than it has beams. Please set "
f"num_beams >= num_return_sequences, got {num_beams} and {num_return_sequences} (respectivelly)"
)
......@@ -1950,9 +1013,44 @@ class TFGenerationMixin:
**model_kwargs,
)
else:
# TODO(Matt, Joao, Patrick) - add more sub-generation methods here
raise NotImplementedError("Beam sampling is currently not implemented.")
elif is_beam_sample_gen_mode:
if num_beams < num_return_sequences:
raise ValueError(
"Beam search decoding cannot return more sequences than it has beams. Please set "
f"num_beams >= num_return_sequences, got {num_beams} and {num_return_sequences} (respectivelly)"
)
# 10. prepare logits warper
logits_warper = self._get_logits_warper(top_k=top_k, top_p=top_p, temperature=temperature)
# 11. broadcast inputs to the desired number of beams
input_ids = self._expand_to_num_beams(input_ids, num_beams=num_beams)
if "encoder_outputs" in model_kwargs:
model_kwargs["encoder_outputs"]["last_hidden_state"] = self._expand_to_num_beams(
model_kwargs["encoder_outputs"]["last_hidden_state"], num_beams=num_beams
)
if "attention_mask" in model_kwargs:
model_kwargs["attention_mask"] = self._expand_to_num_beams(
model_kwargs["attention_mask"], num_beams=num_beams
)
# 12. run beam sample (beam search with sampling)
return self.beam_search(
input_ids,
do_sample=True,
max_length=max_length,
pad_token_id=pad_token_id,
eos_token_id=eos_token_id,
length_penalty=length_penalty,
early_stopping=early_stopping,
logits_processor=logits_processor,
logits_warper=logits_warper,
return_dict_in_generate=return_dict_in_generate,
num_return_sequences=num_return_sequences,
**model_kwargs,
)
@staticmethod
def _expand_to_num_beams(tensor: tf.Tensor, num_beams: int) -> tf.Tensor:
......@@ -2851,28 +1949,54 @@ class TFGenerationMixin:
else:
return generated
@staticmethod
def _gather_beams(nested, beam_indices, batch_axis=0):
"""Gathers the beam slices indexed by beam_indices into new beam array."""
def gather_fn(tensor):
if batch_axis > 0:
# pushes all dimentions before the batch to the end, so we get (batch, beam_id, ...)
perm = tf.concat((tf.range(tf.rank(tensor))[batch_axis:], tf.range(batch_axis)), axis=0)
tensor = tf.transpose(tensor, perm=perm)
gathered_tensor = tf.gather(params=tensor, indices=beam_indices, axis=1, batch_dims=1)
if batch_axis > 0:
# transposes back to the original dimensions
perm = tf.concat((tf.range(tf.rank(tensor))[batch_axis:], tf.range(batch_axis)), axis=0)
perm = tf.math.invert_permutation(perm)
gathered_tensor = tf.transpose(gathered_tensor, perm=perm)
return gathered_tensor
return tf.nest.map_structure(gather_fn, nested)
def beam_search(
self,
input_ids: tf.Tensor,
do_sample: bool = False,
max_length: Optional[int] = None,
pad_token_id: Optional[int] = None,
eos_token_id: Optional[int] = None,
length_penalty: Optional[float] = None,
early_stopping: Optional[bool] = None,
logits_processor: Optional[TFLogitsProcessorList] = None,
logits_warper: Optional[TFLogitsProcessorList] = None,
num_return_sequences: Optional[int] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
output_scores: Optional[bool] = None,
return_dict_in_generate: Optional[bool] = None,
**model_kwargs,
) -> Union[TFBeamSearchOutput, tf.Tensor]:
) -> Union[TFBeamSearchOutput, TFBeamSampleOutput, tf.Tensor]:
r"""
Generates sequences for models with a language modeling head using beam search with multinomial sampling.
Generates sequences for models with a language modeling head using beam search. If `do_sample` is `False`, uses
a greedy approach, otherwise does multinomial sampling without replacement.
Parameters:
input_ids (`tf.Tensor` of shape `(batch_size, sequence_length)`):
The sequence used as a prompt for the generation.
do_sample (`bool`, *optional*, defaults to `False`):
Whether or not to use sampling ; use greedy decoding otherwise.
max_length (`int`, *optional*, defaults to 20):
The maximum length of the sequence to be generated.
pad_token_id (`int`, *optional*):
......@@ -2889,6 +2013,10 @@ class TFGenerationMixin:
logits_processor (`[TFLogitsProcessorList]`, *optional*):
An instance of [`TFLogitsProcessorList`]. List of instances of class derived from [`TFLogitsProcessor`]
used to modify the prediction scores of the language modeling head applied at each generation step.
logits_warper (`TFLogitsProcessorList`, *optional*):
An instance of [`TFLogitsProcessorList`]. List of instances of class derived from [`TFLogitsWarper`]
used to warp the prediction score distribution of the language modeling head applied before multinomial
sampling at each generation step.
num_return_sequences(`int`, *optional*, defaults to 1):
The number of independently computed returned sequences for each element in the batch.
output_attentions (`bool`, *optional*, defaults to `False`):
......@@ -2946,7 +2074,6 @@ class TFGenerationMixin:
... )
>>> outputs = model.beam_search(input_ids, logits_processor=logits_processor, **model_kwargs)
>>> tokenizer.batch_decode(outputs, skip_special_tokens=True)
['Wie alt bist du?']
```"""
......@@ -2959,33 +2086,14 @@ class TFGenerationMixin:
shape[:batch_axis] + [shape[batch_axis] * shape[batch_axis + 1]] + shape[batch_axis + 2 :],
)
def unflatten_beam_dim(tensor, batch_size, num_beams, batch_axis=0):
def unflatten_beam_dim(tensor, num_beams, batch_axis=0):
"""Unflattens the first, flat batch*beam dimension of a non-scalar array."""
shape = shape_list(tensor)
return tf.reshape(tensor, shape[:batch_axis] + [batch_size, num_beams] + shape[batch_axis + 1 :])
def gather_beams(nested, beam_indices, batch_axis=0):
"""Gathers the beam slices indexed by beam_indices into new beam array."""
def gather_fn(tensor):
if batch_axis > 0:
# pushes all dimentions before the batch to the end, so we get (batch, beam_id, ...)
perm = tf.concat((tf.range(tf.rank(tensor))[batch_axis:], tf.range(batch_axis)), axis=0)
tensor = tf.transpose(tensor, perm=perm)
gathered_tensor = tf.gather(params=tensor, indices=beam_indices, axis=1, batch_dims=1)
if batch_axis > 0:
# transposes back to the original dimensions
perm = tf.concat((tf.range(tf.rank(tensor))[batch_axis:], tf.range(batch_axis)), axis=0)
perm = tf.math.invert_permutation(perm)
gathered_tensor = tf.transpose(gathered_tensor, perm=perm)
return gathered_tensor
return tf.nest.map_structure(gather_fn, nested)
return tf.reshape(tensor, shape[:batch_axis] + [-1, num_beams] + shape[batch_axis + 1 :])
# 1. init beam_search values
logits_processor = logits_processor if logits_processor is not None else TFLogitsProcessorList()
logits_warper = logits_warper if logits_warper is not None else TFLogitsProcessorList()
max_length = max_length if max_length is not None else self.config.max_length
pad_token_id = pad_token_id if pad_token_id is not None else self.config.pad_token_id
......@@ -3102,7 +2210,7 @@ class TFGenerationMixin:
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
)
logits = unflatten_beam_dim(model_outputs.logits[:, -1], batch_size, num_beams)
logits = unflatten_beam_dim(model_outputs.logits[:, -1], num_beams)
# Store scores, attentions and hidden_states when required
if not use_xla and return_dict_in_generate:
......@@ -3125,8 +2233,11 @@ class TFGenerationMixin:
# add new logprobs to existing running logprobs scores.
log_probs = tf.nn.log_softmax(logits)
log_probs = logits_processor(flatten_beam_dim(running_sequences), flatten_beam_dim(log_probs), cur_len)
log_probs = unflatten_beam_dim(log_probs, batch_size, num_beams)
log_probs = unflatten_beam_dim(log_probs, num_beams)
log_probs = log_probs + tf.expand_dims(running_scores, axis=2)
if do_sample:
log_probs = logits_warper(flatten_beam_dim(running_sequences), flatten_beam_dim(log_probs), cur_len)
log_probs = unflatten_beam_dim(log_probs, num_beams)
vocab_size = log_probs.shape[2]
log_probs = tf.reshape(log_probs, (batch_size, num_beams * vocab_size))
......@@ -3141,9 +2252,13 @@ class TFGenerationMixin:
# Recover token id by modulo division and expand Id array for broadcasting.
# Update sequences for the 2*K top-k new sequences.
beams_to_keep = 2 * num_beams
if do_sample:
topk_indices = sample_without_replacement(log_probs, beams_to_keep)
topk_log_probs = tf.gather(log_probs, topk_indices, axis=1, batch_dims=1)
else:
topk_log_probs, topk_indices = tf.math.top_k(log_probs, k=beams_to_keep)
topk_beam_indices = topk_indices // vocab_size
topk_running_sequences = gather_beams(running_sequences, topk_beam_indices)
topk_running_sequences = self._gather_beams(running_sequences, topk_beam_indices)
topk_ids = topk_indices % vocab_size
# writes the new token
......@@ -3178,7 +2293,7 @@ class TFGenerationMixin:
# Determine the top k beam indices (from top 2*k beams) from log probs and gather top k beams
# (from top 2*k beams).
next_topk_indices = tf.math.top_k(running_topk_log_probs, k=num_beams)[1]
next_running_sequences, next_running_scores = gather_beams(
next_running_sequences, next_running_scores = self._gather_beams(
[topk_sequences, running_topk_log_probs], next_topk_indices
)
......@@ -3204,7 +2319,7 @@ class TFGenerationMixin:
merged_scores = tf.concat([scores, topk_log_probs], axis=1)
merged_is_sent_finished = tf.concat([is_sent_finished, did_topk_just_finished], axis=1)
topk_merged_indices = tf.math.top_k(merged_scores, k=num_beams)[1]
next_sequences, next_scores, next_is_sent_finished = gather_beams(
next_sequences, next_scores, next_is_sent_finished = self._gather_beams(
[merged_sequences, merged_scores, merged_is_sent_finished], topk_merged_indices
)
......@@ -3214,11 +2329,11 @@ class TFGenerationMixin:
cur_len = cur_len + 1
if "past_key_values" in model_outputs:
cache = tf.nest.map_structure(
lambda tensor: unflatten_beam_dim(tensor, batch_size, num_beams, batch_axis=cache_batch_axis),
lambda tensor: unflatten_beam_dim(tensor, num_beams, batch_axis=cache_batch_axis),
model_outputs.past_key_values,
)
next_running_indices = gather_beams(topk_beam_indices, next_topk_indices)
next_cache = gather_beams(cache, next_running_indices, batch_axis=cache_batch_axis)
next_running_indices = self._gather_beams(topk_beam_indices, next_topk_indices)
next_cache = self._gather_beams(cache, next_running_indices, batch_axis=cache_batch_axis)
model_outputs["past_key_values"] = tf.nest.map_structure(
lambda tensor: flatten_beam_dim(tensor, batch_axis=cache_batch_axis), next_cache
)
......@@ -3303,7 +2418,8 @@ class TFGenerationMixin:
model_kwargs["encoder_outputs"].get("hidden_states") if output_hidden_states else None
)
return TFBeamSearchEncoderDecoderOutput(
output_cls = TFBeamSampleEncoderDecoderOutput if do_sample else TFBeamSearchEncoderDecoderOutput
return output_cls(
sequences=sequences,
scores=scores,
encoder_attentions=encoder_attentions,
......@@ -3313,7 +2429,8 @@ class TFGenerationMixin:
decoder_hidden_states=decoder_hidden_states,
)
else:
return TFBeamSearchDecoderOnlyOutput(
output_cls = TFBeamSampleDecoderOnlyOutput if do_sample else TFBeamSearchDecoderOnlyOutput
return output_cls(
sequences=sequences,
scores=scores,
attentions=decoder_attentions,
......@@ -3714,79 +2831,6 @@ class TFGenerationMixin:
return generated
def _create_next_token_logits_penalties(input_ids, logits, repetition_penalty):
# create logit penalties for already seen input_ids
token_penalties = np.ones(shape_list(logits))
prev_input_ids = [np.unique(input_id) for input_id in input_ids.numpy()]
for i, prev_input_id in enumerate(prev_input_ids):
logit_penalized = logits[i].numpy()[prev_input_id]
logit_penalties = np.zeros(logit_penalized.shape)
# if previous logit score is < 0 then multiply repetition penalty else divide
logit_penalties[logit_penalized < 0] = repetition_penalty
logit_penalties[logit_penalized > 0] = 1 / repetition_penalty
np.put(token_penalties[i], prev_input_id, logit_penalties)
return tf.convert_to_tensor(token_penalties, dtype=tf.float32)
def calc_banned_ngram_tokens(prev_input_ids, num_hypos, no_repeat_ngram_size, cur_len):
# Copied from fairseq for no_repeat_ngram in beam_search
if cur_len + 1 < no_repeat_ngram_size:
# return no banned tokens if we haven't generated no_repeat_ngram_size tokens yet
return [[] for _ in range(num_hypos)]
generated_ngrams = [{} for _ in range(num_hypos)]
for idx in range(num_hypos):
gen_tokens = prev_input_ids[idx].numpy().tolist()
generated_ngram = generated_ngrams[idx]
for ngram in zip(*[gen_tokens[i:] for i in range(no_repeat_ngram_size)]):
prev_ngram_tuple = tuple(ngram[:-1])
generated_ngram[prev_ngram_tuple] = generated_ngram.get(prev_ngram_tuple, []) + [ngram[-1]]
def _get_generated_ngrams(hypo_idx):
# Before decoding the next token, prevent decoding of ngrams that have already appeared
start_idx = cur_len + 1 - no_repeat_ngram_size
ngram_idx = tuple(prev_input_ids[hypo_idx, start_idx:cur_len].numpy().tolist())
return generated_ngrams[hypo_idx].get(ngram_idx, [])
banned_tokens = [_get_generated_ngrams(hypo_idx) for hypo_idx in range(num_hypos)]
return banned_tokens
def calc_banned_bad_words_ids(prev_input_ids, bad_words_ids):
banned_tokens = []
def _tokens_match(prev_tokens, tokens):
if len(tokens) == 0:
# if bad word tokens is just one token always ban it
return True
if len(tokens) > len(prev_tokens):
# if bad word tokens are longer than prev tokens they can't be equal
return False
if prev_tokens[-len(tokens) :] == tokens:
# if tokens match
return True
else:
return False
for prev_input_ids_slice in prev_input_ids:
banned_tokens_slice = []
for banned_token_seq in bad_words_ids:
assert (
len(banned_token_seq) > 0
), f"Banned words token sequences { bad_words_ids} cannot have an empty list"
if _tokens_match(prev_input_ids_slice.numpy().tolist(), banned_token_seq[:-1]) is False:
# if tokens do not match continue
continue
banned_tokens_slice.append(banned_token_seq[-1])
banned_tokens.append(banned_tokens_slice)
return banned_tokens
def tf_top_k_top_p_filtering(logits, top_k=0, top_p=1.0, filter_value=-float("Inf"), min_tokens_to_keep=1):
"""
Filter a distribution of logits using top-k and/or nucleus (top-p) filtering
......@@ -3857,58 +2901,11 @@ def sample_without_replacement(logits, num_samples):
categorical sampling without replacement is currently not implemented the gumbel-max trick will do for now see
https://github.com/tensorflow/tensorflow/issues/9260 for more info
"""
z = -tf.math.log(tf.random.uniform(shape_list(logits), 0, 1))
z = -tf.math.log(-tf.math.log(tf.random.uniform(shape_list(logits), 0, 1)))
_, indices = tf.nn.top_k(logits + z, num_samples)
return indices
class BeamHypotheses(object):
def __init__(self, num_beams, max_length, length_penalty, early_stopping):
"""
Initialize n-best list of hypotheses.
"""
self.max_length = max_length - 1 # ignoring bos_token
self.length_penalty = length_penalty
self.early_stopping = early_stopping
self.num_beams = num_beams
self.beams = []
self.worst_score = 1e9
def __len__(self):
"""
Number of hypotheses in the list.
"""
return len(self.beams)
def add(self, hyp, sum_logprobs):
"""
Add a new hypothesis to the list.
"""
score = sum_logprobs / len(hyp) ** self.length_penalty
if len(self) < self.num_beams or score > self.worst_score:
self.beams.append((score, hyp))
if len(self) > self.num_beams:
sorted_scores = sorted([(s, idx) for idx, (s, _) in enumerate(self.beams)])
del self.beams[sorted_scores[0][1]]
self.worst_score = sorted_scores[1][0]
else:
self.worst_score = min(score, self.worst_score)
def is_done(self, best_sum_logprobs, cur_len):
"""
If there are enough hypotheses and that none of the hypotheses being generated can become better than the worst
one in the heap, then we are done with this sentence.
"""
if len(self) < self.num_beams:
return False
elif self.early_stopping:
return True
else:
cur_score = best_sum_logprobs / cur_len**self.length_penalty
ret = self.worst_score >= cur_score
return ret
def _ranking_fast(
context_hidden: tf.Tensor,
next_hidden: tf.Tensor,
......
src/transformers/models/rag/modeling_tf_rag.py
View file @ 588faad1
......@@ -817,6 +817,32 @@ class TFRagTokenForGeneration(TFRagPreTrainedModel, TFCausalLanguageModelingLoss
return reordered_past
@staticmethod
def _gather_beams(nested, beam_indices, batch_axis=0):
"""
RAG-specific `_gather_beams`: gathers the beam slices indexed by beam_indices into new beam array. If the
nested tensor has a shape mismatch with the beam indices, then it means it is the cache. In that case, isolates
and takes care of the extra dimension for ndocs.
"""
def gather_fn(tensor):
is_rag_cache = tensor.shape[0] != beam_indices.shape[0]
if is_rag_cache:
n_docs = tensor.shape[0] // beam_indices.shape[0]
batch_size = beam_indices.shape[0]
# reshapes into (batch size, num beams, n_docs, ...), the cache format expected by RAG
tensor = tf.reshape(tensor, (batch_size, -1, n_docs, *tensor.shape[2:]))
gathered_tensor = tf.gather(params=tensor, indices=beam_indices, axis=1, batch_dims=1)
if is_rag_cache:
# reshapes back into the shape expected by beam search
gathered_tensor = tf.reshape(gathered_tensor, (batch_size * n_docs, -1, *gathered_tensor.shape[3:]))
return gathered_tensor
return tf.nest.map_structure(gather_fn, nested)
def marginalize(self, seq_logits, doc_scores, n_docs=None):
n_docs = n_docs if n_docs is not None else self.config.n_docs
......@@ -1129,12 +1155,6 @@ class TFRagTokenForGeneration(TFRagPreTrainedModel, TFCausalLanguageModelingLoss
return_dict_in_generate if return_dict_in_generate is not None else self.config.return_dict_in_generate
)
model_kwargs["output_scores"] = output_scores
model_kwargs["output_attentions"] = output_attentions
model_kwargs["output_hidden_states"] = output_hidden_states
model_kwargs["encoder_attentions"] = None
model_kwargs["encoder_hidden_states"] = None
# retrieve docs
if self.retriever is not None and context_input_ids is None:
question_hidden_states = self.question_encoder(input_ids, attention_mask=attention_mask)[0]
......@@ -1211,71 +1231,55 @@ class TFRagTokenForGeneration(TFRagPreTrainedModel, TFCausalLanguageModelingLoss
doc_scores = tf.repeat(doc_scores, num_beams, axis=0)
# define start_len & additional parameters
cur_len = 1
vocab_size = self.config.generator.vocab_size
model_kwargs["doc_scores"] = doc_scores
model_kwargs["encoder_outputs"] = encoder_outputs
model_kwargs["attention_mask"] = context_attention_mask
model_kwargs["n_docs"] = n_docs
# not needed. TODO(PVP): change after generate refactor
do_sample = False
temperature = self.config.temperature
top_k = self.config.top_k
top_p = self.config.top_p
repetition_penalty = self.config.repetition_penalty
if num_beams > 1:
return self._generate_beam_search(
decoder_input_ids,
cur_len=cur_len,
max_length=max_length,
min_length=min_length,
do_sample=do_sample,
early_stopping=early_stopping,
temperature=temperature,
top_k=top_k,
top_p=top_p,
repetition_penalty=repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
bad_words_ids=bad_words_ids,
pad_token_id=pad_token_id,
eos_token_id=eos_token_id,
batch_size=batch_size,
num_return_sequences=num_return_sequences,
length_penalty=length_penalty,
num_beams=num_beams,
vocab_size=vocab_size,
attention_mask=context_attention_mask,
use_cache=use_cache,
forced_bos_token_id=None,
forced_eos_token_id=None,
return_dict_in_generate=return_dict_in_generate,
**model_kwargs, # encoder_outputs is here as in Pytorch's version
)
else:
pre_processor = self._get_logits_processor(
repetition_penalty=repetition_penalty,
repetition_penalty=self.config.repetition_penalty,
no_repeat_ngram_size=no_repeat_ngram_size,
bad_words_ids=bad_words_ids,
min_length=min_length,
max_length=max_length,
eos_token_id=eos_token_id,
forced_bos_token_id=None,
forced_eos_token_id=None,
forced_bos_token_id=self.config.generator.forced_bos_token_id,
forced_eos_token_id=self.config.generator.forced_eos_token_id,
input_ids_seq_length=tf.shape(decoder_input_ids)[-1],
)
model_kwargs["attention_mask"] = context_attention_mask
if model_kwargs.get("encoder_attentions", None) is None:
model_kwargs.pop("encoder_attentions", None)
if model_kwargs.get("encoder_hidden_states", None) is None:
model_kwargs.pop("encoder_hidden_states", None)
if num_beams == 1:
return self.greedy_search(
input_ids=decoder_input_ids,
max_length=max_length,
pad_token_id=pad_token_id,
eos_token_id=eos_token_id,
logits_processor=pre_processor,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
output_scores=output_scores,
return_dict_in_generate=return_dict_in_generate,
**model_kwargs,
)
elif num_beams > 1:
if num_beams < num_return_sequences:
raise ValueError(
"Beam search decoding cannot return more sequences than it has beams. Please set "
f"num_beams >= num_return_sequences, got {num_beams} and {num_return_sequences} (respectivelly)"
)
model_kwargs.pop("output_hidden_states", None)
model_kwargs.pop("output_attentions", None)
model_kwargs.pop("output_scores", None)
def unflatten_beam_dim(tensor):
"""Unflattens the first, flat batch*beam dimension of a non-scalar array."""
shape = shape_list(tensor)
return tf.reshape(tensor, [-1, num_beams] + shape[1:])
return self.greedy_search(
decoder_input_ids = unflatten_beam_dim(decoder_input_ids)
model_kwargs["attention_mask"] = unflatten_beam_dim(model_kwargs["attention_mask"])
model_kwargs["encoder_outputs"]["last_hidden_state"] = unflatten_beam_dim(
model_kwargs["encoder_outputs"]["last_hidden_state"]
)
return self.beam_search(
input_ids=decoder_input_ids,
max_length=max_length,
pad_token_id=pad_token_id,
......@@ -1287,6 +1291,8 @@ class TFRagTokenForGeneration(TFRagPreTrainedModel, TFCausalLanguageModelingLoss
return_dict_in_generate=return_dict_in_generate,
**model_kwargs,
)
else:
raise ValueError(f"`num_beams` has to be an integer strictly superior to 0 (≥ 1), but is {num_beams}")
def get_input_embeddings(self):
return self.rag.generator.get_input_embeddings()
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment