1785 lines
80 KiB
Python
1785 lines
80 KiB
Python
# coding=utf-8
|
|
# Copyright 2022 The Fairseq Authors and The HuggingFace Inc. team. All rights reserved.
|
|
#
|
|
# 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.
|
|
"""PyTorch MVP model."""
|
|
|
|
import math
|
|
from typing import Optional, Union
|
|
|
|
import torch
|
|
import torch.utils.checkpoint
|
|
from torch import nn
|
|
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
|
|
|
|
from ...activations import ACT2FN
|
|
from ...cache_utils import Cache, EncoderDecoderCache
|
|
from ...generation import GenerationMixin
|
|
from ...modeling_attn_mask_utils import (
|
|
_prepare_4d_attention_mask,
|
|
_prepare_4d_causal_attention_mask,
|
|
)
|
|
from ...modeling_layers import GradientCheckpointingLayer
|
|
from ...modeling_outputs import (
|
|
BaseModelOutput,
|
|
BaseModelOutputWithPastAndCrossAttentions,
|
|
CausalLMOutputWithCrossAttentions,
|
|
Seq2SeqLMOutput,
|
|
Seq2SeqModelOutput,
|
|
Seq2SeqQuestionAnsweringModelOutput,
|
|
Seq2SeqSequenceClassifierOutput,
|
|
)
|
|
from ...modeling_utils import PreTrainedModel
|
|
from ...utils import auto_docstring, logging
|
|
from .configuration_mvp import MvpConfig
|
|
|
|
|
|
logger = logging.get_logger(__name__)
|
|
|
|
|
|
# Copied from transformers.models.bart.modeling_bart.shift_tokens_right
|
|
def shift_tokens_right(input_ids: torch.Tensor, pad_token_id: int, decoder_start_token_id: int):
|
|
"""
|
|
Shift input ids one token to the right.
|
|
"""
|
|
shifted_input_ids = input_ids.new_zeros(input_ids.shape)
|
|
shifted_input_ids[:, 1:] = input_ids[:, :-1].clone()
|
|
shifted_input_ids[:, 0] = decoder_start_token_id
|
|
|
|
if pad_token_id is None:
|
|
raise ValueError("self.model.config.pad_token_id has to be defined.")
|
|
# replace possible -100 values in labels by `pad_token_id`
|
|
shifted_input_ids.masked_fill_(shifted_input_ids == -100, pad_token_id)
|
|
|
|
return shifted_input_ids
|
|
|
|
|
|
# Copied from transformers.models.bart.modeling_bart.BartLearnedPositionalEmbedding with Bart->Mvp
|
|
class MvpLearnedPositionalEmbedding(nn.Embedding):
|
|
"""
|
|
This module learns positional embeddings up to a fixed maximum size.
|
|
"""
|
|
|
|
def __init__(self, num_embeddings: int, embedding_dim: int):
|
|
# Mvp is set up so that if padding_idx is specified then offset the embedding ids by 2
|
|
# and adjust num_embeddings appropriately. Other models don't have this hack
|
|
self.offset = 2
|
|
super().__init__(num_embeddings + self.offset, embedding_dim)
|
|
|
|
def forward(self, input_ids: torch.Tensor, past_key_values_length: int = 0, position_ids: torch.Tensor = None):
|
|
"""`input_ids' shape is expected to be [bsz x seqlen]."""
|
|
|
|
if position_ids is None:
|
|
bsz, seq_len = input_ids.shape[:2]
|
|
position_ids = torch.arange(
|
|
past_key_values_length, past_key_values_length + seq_len, dtype=torch.long, device=self.weight.device
|
|
).expand(bsz, -1)
|
|
else:
|
|
position_ids = position_ids.unsqueeze(0)
|
|
|
|
return super().forward(position_ids + self.offset)
|
|
|
|
|
|
class MvpAttention(nn.Module):
|
|
"""Multi-headed attention from 'Attention Is All You Need' paper"""
|
|
|
|
def __init__(
|
|
self,
|
|
embed_dim: int,
|
|
num_heads: int,
|
|
dropout: Optional[float] = 0.0,
|
|
is_decoder: Optional[bool] = False,
|
|
bias: Optional[bool] = True,
|
|
layer_idx: Optional[bool] = None,
|
|
):
|
|
super().__init__()
|
|
self.embed_dim = embed_dim
|
|
self.num_heads = num_heads
|
|
self.dropout = dropout
|
|
self.head_dim = embed_dim // num_heads
|
|
|
|
if (self.head_dim * num_heads) != self.embed_dim:
|
|
raise ValueError(
|
|
f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
|
|
f" and `num_heads`: {num_heads})."
|
|
)
|
|
self.scaling = self.head_dim**-0.5
|
|
self.is_decoder = is_decoder
|
|
self.layer_idx = layer_idx
|
|
|
|
self.k_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
|
|
self.v_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
|
|
self.q_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
|
|
self.out_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.Tensor,
|
|
key_value_states: Optional[torch.Tensor] = None,
|
|
past_key_value: Optional[Cache] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
layer_head_mask: Optional[torch.Tensor] = None,
|
|
attn_prompt: Optional[torch.Tensor] = None,
|
|
output_attentions: bool = False,
|
|
cache_position: Optional[torch.Tensor] = None,
|
|
) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
|
|
"""Input shape: Batch x Time x Channel"""
|
|
|
|
# if key_value_states are provided this layer is used as a cross-attention layer
|
|
# for the decoder
|
|
is_cross_attention = key_value_states is not None
|
|
|
|
bsz, tgt_len, _ = hidden_states.size()
|
|
|
|
# get query proj
|
|
query_states = self.q_proj(hidden_states) * self.scaling
|
|
|
|
if past_key_value is not None:
|
|
if isinstance(past_key_value, EncoderDecoderCache):
|
|
is_updated = past_key_value.is_updated.get(self.layer_idx)
|
|
if is_cross_attention:
|
|
# after the first generated id, we can subsequently re-use all key/value_states from cache
|
|
curr_past_key_value = past_key_value.cross_attention_cache
|
|
else:
|
|
curr_past_key_value = past_key_value.self_attention_cache
|
|
else:
|
|
curr_past_key_value = past_key_value
|
|
|
|
current_states = key_value_states if is_cross_attention else hidden_states
|
|
if is_cross_attention and past_key_value is not None and is_updated:
|
|
# reuse k,v, cross_attentions
|
|
key_states = curr_past_key_value.key_cache[self.layer_idx]
|
|
value_states = curr_past_key_value.value_cache[self.layer_idx]
|
|
else:
|
|
key_states = self.k_proj(current_states)
|
|
value_states = self.v_proj(current_states)
|
|
key_states = key_states.view(bsz, -1, self.num_heads, self.head_dim).transpose(1, 2)
|
|
value_states = value_states.view(bsz, -1, self.num_heads, self.head_dim).transpose(1, 2)
|
|
|
|
if past_key_value is not None:
|
|
# save all key/value_states to cache to be re-used for fast auto-regressive generation
|
|
cache_position = cache_position if not is_cross_attention else None
|
|
key_states, value_states = curr_past_key_value.update(
|
|
key_states, value_states, self.layer_idx, {"cache_position": cache_position}
|
|
)
|
|
# set flag that curr layer for cross-attn is already updated so we can re-use in subsequent calls
|
|
if is_cross_attention:
|
|
past_key_value.is_updated[self.layer_idx] = True
|
|
|
|
if attn_prompt is not None:
|
|
key_states = torch.cat([attn_prompt[0].expand(bsz, -1, -1, -1), key_states], dim=2)
|
|
value_states = torch.cat([attn_prompt[1].expand(bsz, -1, -1, -1), value_states], dim=2)
|
|
if attention_mask is not None:
|
|
prompt_mask = torch.zeros(bsz, 1, tgt_len, attn_prompt[0].size(1)).to(attention_mask.device)
|
|
attention_mask = torch.cat([prompt_mask, attention_mask], dim=(-1))
|
|
|
|
proj_shape = (bsz * self.num_heads, -1, self.head_dim)
|
|
query_states = query_states.view(bsz, tgt_len, self.num_heads, self.head_dim).transpose(1, 2)
|
|
query_states = query_states.reshape(*proj_shape)
|
|
key_states = key_states.reshape(*proj_shape)
|
|
value_states = value_states.reshape(*proj_shape)
|
|
|
|
src_len = key_states.size(1)
|
|
attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))
|
|
|
|
if attn_weights.size() != (bsz * self.num_heads, tgt_len, src_len):
|
|
raise ValueError(
|
|
f"Attention weights should be of size {(bsz * self.num_heads, tgt_len, src_len)}, but is"
|
|
f" {attn_weights.size()}"
|
|
)
|
|
|
|
if attention_mask is not None:
|
|
if attention_mask.size() != (bsz, 1, tgt_len, src_len):
|
|
raise ValueError(
|
|
f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}"
|
|
)
|
|
attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + attention_mask
|
|
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
|
|
|
|
attn_weights = nn.functional.softmax(attn_weights, dim=-1)
|
|
|
|
if layer_head_mask is not None:
|
|
if layer_head_mask.size() != (self.num_heads,):
|
|
raise ValueError(
|
|
f"Head mask for a single layer should be of size {(self.num_heads,)}, but is"
|
|
f" {layer_head_mask.size()}"
|
|
)
|
|
attn_weights = layer_head_mask.view(1, -1, 1, 1) * attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
|
|
attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)
|
|
|
|
if output_attentions:
|
|
# this operation is a bit awkward, but it's required to
|
|
# make sure that attn_weights keeps its gradient.
|
|
# In order to do so, attn_weights have to be reshaped
|
|
# twice and have to be reused in the following
|
|
attn_weights_reshaped = attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
|
|
attn_weights = attn_weights_reshaped.view(bsz * self.num_heads, tgt_len, src_len)
|
|
else:
|
|
attn_weights_reshaped = None
|
|
|
|
attn_probs = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)
|
|
|
|
attn_output = torch.bmm(attn_probs, value_states)
|
|
|
|
if attn_output.size() != (bsz * self.num_heads, tgt_len, self.head_dim):
|
|
raise ValueError(
|
|
f"`attn_output` should be of size {(bsz, self.num_heads, tgt_len, self.head_dim)}, but is"
|
|
f" {attn_output.size()}"
|
|
)
|
|
|
|
attn_output = attn_output.view(bsz, self.num_heads, tgt_len, self.head_dim)
|
|
attn_output = attn_output.transpose(1, 2)
|
|
|
|
# Use the `embed_dim` from the config (stored in the class) rather than `hidden_state` because `attn_output` can be
|
|
# partitioned aross GPUs when using tensor-parallelism.
|
|
attn_output = attn_output.reshape(bsz, tgt_len, self.embed_dim)
|
|
|
|
attn_output = self.out_proj(attn_output)
|
|
|
|
return attn_output, attn_weights_reshaped
|
|
|
|
|
|
class MvpEncoderLayer(GradientCheckpointingLayer):
|
|
def __init__(self, config: MvpConfig):
|
|
super().__init__()
|
|
self.embed_dim = config.d_model
|
|
self.self_attn = MvpAttention(
|
|
embed_dim=self.embed_dim,
|
|
num_heads=config.encoder_attention_heads,
|
|
dropout=config.attention_dropout,
|
|
)
|
|
self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
|
|
self.dropout = config.dropout
|
|
self.activation_fn = ACT2FN[config.activation_function]
|
|
self.activation_dropout = config.activation_dropout
|
|
self.fc1 = nn.Linear(self.embed_dim, config.encoder_ffn_dim)
|
|
self.fc2 = nn.Linear(config.encoder_ffn_dim, self.embed_dim)
|
|
self.final_layer_norm = nn.LayerNorm(self.embed_dim)
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
attention_mask: torch.FloatTensor,
|
|
layer_head_mask: torch.FloatTensor,
|
|
self_attn_prompt: torch.FloatTensor,
|
|
output_attentions: Optional[bool] = False,
|
|
) -> tuple[torch.FloatTensor, Optional[torch.FloatTensor]]:
|
|
"""
|
|
Args:
|
|
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
|
|
attention_mask (`torch.FloatTensor`): attention mask of size
|
|
`(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
|
|
layer_head_mask (`torch.FloatTensor`): mask for attention heads in a given layer of size
|
|
`(encoder_attention_heads,)`.
|
|
self_attn_prompt (`torch.FloatTensor`): prompt of self attention of shape
|
|
`(2, encoder_attention_heads, pro_len, head_dim)`.
|
|
output_attentions (`bool`, *optional*):
|
|
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
|
returned tensors for more detail.
|
|
"""
|
|
residual = hidden_states
|
|
hidden_states, attn_weights = self.self_attn(
|
|
hidden_states=hidden_states,
|
|
attention_mask=attention_mask,
|
|
layer_head_mask=layer_head_mask,
|
|
attn_prompt=self_attn_prompt,
|
|
output_attentions=output_attentions,
|
|
)
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
hidden_states = residual + hidden_states
|
|
hidden_states = self.self_attn_layer_norm(hidden_states)
|
|
|
|
residual = hidden_states
|
|
hidden_states = self.activation_fn(self.fc1(hidden_states))
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
|
|
hidden_states = self.fc2(hidden_states)
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
hidden_states = residual + hidden_states
|
|
hidden_states = self.final_layer_norm(hidden_states)
|
|
|
|
if hidden_states.dtype == torch.float16 and (
|
|
torch.isinf(hidden_states).any() or torch.isnan(hidden_states).any()
|
|
):
|
|
clamp_value = torch.finfo(hidden_states.dtype).max - 1000
|
|
hidden_states = torch.clamp(hidden_states, min=-clamp_value, max=clamp_value)
|
|
|
|
return hidden_states, attn_weights
|
|
|
|
|
|
class MvpDecoderLayer(GradientCheckpointingLayer):
|
|
def __init__(self, config: MvpConfig, layer_idx=None):
|
|
super().__init__()
|
|
self.embed_dim = config.d_model
|
|
|
|
self.self_attn = MvpAttention(
|
|
embed_dim=self.embed_dim,
|
|
num_heads=config.decoder_attention_heads,
|
|
dropout=config.attention_dropout,
|
|
is_decoder=True,
|
|
layer_idx=layer_idx,
|
|
)
|
|
self.dropout = config.dropout
|
|
self.activation_fn = ACT2FN[config.activation_function]
|
|
self.activation_dropout = config.activation_dropout
|
|
|
|
self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
|
|
self.encoder_attn = MvpAttention(
|
|
self.embed_dim,
|
|
config.decoder_attention_heads,
|
|
dropout=config.attention_dropout,
|
|
is_decoder=True,
|
|
layer_idx=layer_idx,
|
|
)
|
|
self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim)
|
|
self.fc1 = nn.Linear(self.embed_dim, config.decoder_ffn_dim)
|
|
self.fc2 = nn.Linear(config.decoder_ffn_dim, self.embed_dim)
|
|
self.final_layer_norm = nn.LayerNorm(self.embed_dim)
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.Tensor,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
encoder_hidden_states: Optional[torch.Tensor] = None,
|
|
encoder_attention_mask: Optional[torch.Tensor] = None,
|
|
layer_head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_layer_head_mask: Optional[torch.Tensor] = None,
|
|
self_attn_prompt: Optional[torch.Tensor] = None,
|
|
cross_attn_prompt: Optional[torch.Tensor] = None,
|
|
past_key_value: Optional[Cache] = None,
|
|
output_attentions: Optional[bool] = False,
|
|
use_cache: Optional[bool] = True,
|
|
cache_position: Optional[torch.Tensor] = None,
|
|
) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
|
|
"""
|
|
Args:
|
|
hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
|
|
attention_mask (`torch.FloatTensor`): attention mask of size
|
|
`(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
|
|
encoder_hidden_states (`torch.FloatTensor`):
|
|
cross attention input to the layer of shape `(batch, seq_len, embed_dim)`
|
|
encoder_attention_mask (`torch.FloatTensor`): encoder attention mask of size
|
|
`(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
|
|
layer_head_mask (`torch.FloatTensor`): mask for attention heads in a given layer of size
|
|
`(encoder_attention_heads,)`.
|
|
cross_attn_layer_head_mask (`torch.FloatTensor`): mask for cross-attention heads in a given layer of
|
|
size `(decoder_attention_heads,)`.
|
|
self_attn_prompt (`torch.FloatTensor`): prompt of self attention of shape
|
|
`(2, decoder_attention_heads, pro_len, head_dim)`.
|
|
cross_attn_prompt (`torch.FloatTensor`): prompt of cross attention of shape
|
|
`(2, decoder_attention_heads, pro_len, head_dim)`.
|
|
past_key_value (`Tuple(torch.FloatTensor)`): cached past key and value projection states
|
|
output_attentions (`bool`, *optional*):
|
|
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
|
returned tensors for more detail.
|
|
"""
|
|
residual = hidden_states
|
|
|
|
# Self Attention
|
|
hidden_states, self_attn_weights = self.self_attn(
|
|
hidden_states=hidden_states,
|
|
past_key_value=past_key_value,
|
|
attention_mask=attention_mask,
|
|
layer_head_mask=layer_head_mask,
|
|
attn_prompt=self_attn_prompt,
|
|
output_attentions=output_attentions,
|
|
cache_position=cache_position,
|
|
)
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
hidden_states = residual + hidden_states
|
|
hidden_states = self.self_attn_layer_norm(hidden_states)
|
|
|
|
# Cross-Attention Block
|
|
cross_attn_weights = None
|
|
if encoder_hidden_states is not None:
|
|
residual = hidden_states
|
|
|
|
hidden_states, cross_attn_weights = self.encoder_attn(
|
|
hidden_states=hidden_states,
|
|
key_value_states=encoder_hidden_states,
|
|
attention_mask=encoder_attention_mask,
|
|
layer_head_mask=cross_attn_layer_head_mask,
|
|
attn_prompt=cross_attn_prompt,
|
|
past_key_value=past_key_value,
|
|
output_attentions=output_attentions,
|
|
)
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
hidden_states = residual + hidden_states
|
|
hidden_states = self.encoder_attn_layer_norm(hidden_states)
|
|
|
|
# Fully Connected
|
|
residual = hidden_states
|
|
hidden_states = self.activation_fn(self.fc1(hidden_states))
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.activation_dropout, training=self.training)
|
|
hidden_states = self.fc2(hidden_states)
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
hidden_states = residual + hidden_states
|
|
hidden_states = self.final_layer_norm(hidden_states)
|
|
|
|
outputs = (hidden_states,)
|
|
|
|
if output_attentions:
|
|
outputs += (self_attn_weights, cross_attn_weights)
|
|
|
|
return outputs
|
|
|
|
|
|
# Copied from transformers.models.bart.modeling_bart.BartClassificationHead with Bart->MVP
|
|
class MvpClassificationHead(nn.Module):
|
|
"""Head for sentence-level classification tasks."""
|
|
|
|
def __init__(
|
|
self,
|
|
input_dim: int,
|
|
inner_dim: int,
|
|
num_classes: int,
|
|
pooler_dropout: float,
|
|
):
|
|
super().__init__()
|
|
self.dense = nn.Linear(input_dim, inner_dim)
|
|
self.dropout = nn.Dropout(p=pooler_dropout)
|
|
self.out_proj = nn.Linear(inner_dim, num_classes)
|
|
|
|
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
|
|
hidden_states = self.dropout(hidden_states)
|
|
hidden_states = self.dense(hidden_states)
|
|
hidden_states = torch.tanh(hidden_states)
|
|
hidden_states = self.dropout(hidden_states)
|
|
hidden_states = self.out_proj(hidden_states)
|
|
return hidden_states
|
|
|
|
|
|
class MvpPrompt(nn.Module):
|
|
"""Layer-wise prompt for encoder or decoder."""
|
|
|
|
def __init__(self, config, num_layers, num_heads):
|
|
super().__init__()
|
|
self.prompt_length = config.prompt_length
|
|
self.num_layers = num_layers
|
|
self.num_heads = num_heads
|
|
self.head_dim = config.d_model // num_heads
|
|
self.dropout = nn.Dropout(p=config.dropout)
|
|
self.prompt_embedding = nn.Embedding(config.prompt_length, config.d_model)
|
|
self.prompt_trans = nn.Sequential(
|
|
nn.Linear(config.d_model, config.prompt_mid_dim),
|
|
nn.GELU(),
|
|
nn.Linear(config.prompt_mid_dim, num_layers * 2 * config.d_model),
|
|
)
|
|
|
|
def forward(self, prompt_ids: torch.Tensor) -> tuple[torch.Tensor]:
|
|
prompt = self.prompt_trans(self.prompt_embedding(prompt_ids))
|
|
prompt = prompt.view(self.prompt_length, self.num_layers * 2, self.num_heads, self.head_dim)
|
|
prompt = self.dropout(prompt)
|
|
prompt = prompt.permute([1, 2, 0, 3]).split(2)
|
|
return prompt
|
|
|
|
|
|
@auto_docstring
|
|
class MvpPreTrainedModel(PreTrainedModel):
|
|
config: MvpConfig
|
|
base_model_prefix = "model"
|
|
supports_gradient_checkpointing = True
|
|
|
|
def _init_weights(self, module):
|
|
std = self.config.init_std
|
|
if isinstance(module, nn.Linear):
|
|
module.weight.data.normal_(mean=0.0, std=std)
|
|
if module.bias is not None:
|
|
module.bias.data.zero_()
|
|
elif isinstance(module, nn.Embedding):
|
|
module.weight.data.normal_(mean=0.0, std=std)
|
|
if module.padding_idx is not None:
|
|
module.weight.data[module.padding_idx].zero_()
|
|
|
|
@property
|
|
def dummy_inputs(self):
|
|
pad_token = self.config.pad_token_id
|
|
input_ids = torch.tensor([[0, 6, 10, 4, 2], [0, 8, 12, 2, pad_token]], device=self.device)
|
|
dummy_inputs = {
|
|
"attention_mask": input_ids.ne(pad_token),
|
|
"input_ids": input_ids,
|
|
}
|
|
return dummy_inputs
|
|
|
|
|
|
class MvpEncoder(MvpPreTrainedModel):
|
|
"""
|
|
Transformer encoder consisting of *config.encoder_layers* self attention layers. Each layer is a
|
|
[`MvpEncoderLayer`].
|
|
|
|
Args:
|
|
config: MvpConfig
|
|
embed_tokens (nn.Embedding): output embedding
|
|
use_prompt (bool): whether to use prompt
|
|
"""
|
|
|
|
def __init__(
|
|
self, config: MvpConfig, embed_tokens: Optional[nn.Embedding] = None, use_prompt: Optional[bool] = False
|
|
):
|
|
super().__init__(config)
|
|
|
|
self.dropout = config.dropout
|
|
self.layerdrop = config.encoder_layerdrop
|
|
|
|
embed_dim = config.d_model
|
|
self.padding_idx = config.pad_token_id
|
|
self.max_source_positions = config.max_position_embeddings
|
|
self.embed_scale = math.sqrt(embed_dim) if config.scale_embedding else 1.0
|
|
|
|
if embed_tokens is not None:
|
|
self.embed_tokens = embed_tokens
|
|
else:
|
|
self.embed_tokens = nn.Embedding(config.vocab_size, embed_dim, self.padding_idx)
|
|
|
|
self.embed_positions = MvpLearnedPositionalEmbedding(
|
|
config.max_position_embeddings,
|
|
embed_dim,
|
|
)
|
|
self.layers = nn.ModuleList([MvpEncoderLayer(config) for _ in range(config.encoder_layers)])
|
|
self.layernorm_embedding = nn.LayerNorm(embed_dim)
|
|
|
|
self.use_prompt = use_prompt
|
|
if use_prompt:
|
|
self.prompt_length = config.prompt_length
|
|
self.self_attn_prompt = MvpPrompt(
|
|
config,
|
|
config.encoder_layers,
|
|
config.encoder_attention_heads,
|
|
)
|
|
|
|
self.gradient_checkpointing = False
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.LongTensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
) -> Union[tuple, BaseModelOutput]:
|
|
r"""
|
|
Args:
|
|
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
|
|
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
|
|
provide it.
|
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
|
[`PreTrainedTokenizer.__call__`] for details.
|
|
|
|
[What are input IDs?](../glossary#input-ids)
|
|
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
|
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
|
|
|
- 1 for tokens that are **not masked**,
|
|
- 0 for tokens that are **masked**.
|
|
|
|
[What are attention masks?](../glossary#attention-mask)
|
|
head_mask (`torch.Tensor` of shape `(encoder_layers, encoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the attention modules. Mask values selected in `[0, 1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
|
|
inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
|
|
Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
|
|
This is useful if you want more control over how to convert `input_ids` indices into associated vectors
|
|
than the model's internal embedding lookup matrix.
|
|
output_attentions (`bool`, *optional*):
|
|
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
|
returned tensors for more detail.
|
|
output_hidden_states (`bool`, *optional*):
|
|
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
|
|
for more detail.
|
|
return_dict (`bool`, *optional*):
|
|
Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
|
|
"""
|
|
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 = return_dict if return_dict is not None else self.config.use_return_dict
|
|
|
|
# retrieve input_ids and inputs_embeds
|
|
if input_ids is not None and inputs_embeds is not None:
|
|
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
|
|
elif input_ids is not None:
|
|
input = input_ids
|
|
input_shape = input.shape
|
|
input_ids = input_ids.view(-1, input_shape[-1])
|
|
elif inputs_embeds is not None:
|
|
input_shape = inputs_embeds.size()[:-1]
|
|
input = inputs_embeds[:, :, -1]
|
|
else:
|
|
raise ValueError("You have to specify either input_ids or inputs_embeds")
|
|
|
|
if inputs_embeds is None:
|
|
inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale
|
|
|
|
embed_pos = self.embed_positions(input)
|
|
|
|
hidden_states = inputs_embeds + embed_pos
|
|
hidden_states = self.layernorm_embedding(hidden_states)
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
|
|
# layer-wise prompt
|
|
if self.use_prompt:
|
|
prompt_ids = torch.arange(self.prompt_length).to(self.device)
|
|
self_attn_prompt = self.self_attn_prompt(prompt_ids)
|
|
|
|
# expand attention_mask
|
|
if attention_mask is not None:
|
|
# [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
|
|
attention_mask = _prepare_4d_attention_mask(attention_mask, inputs_embeds.dtype)
|
|
|
|
encoder_states = () if output_hidden_states else None
|
|
all_attentions = () if output_attentions else None
|
|
|
|
# check if head_mask has a correct number of layers specified if desired
|
|
if head_mask is not None:
|
|
if head_mask.size()[0] != (len(self.layers)):
|
|
raise ValueError(
|
|
f"The head_mask should be specified for {len(self.layers)} layers, but it is for"
|
|
f" {head_mask.size()[0]}."
|
|
)
|
|
|
|
for idx, encoder_layer in enumerate(self.layers):
|
|
if output_hidden_states:
|
|
encoder_states = encoder_states + (hidden_states,)
|
|
# add LayerDrop (see https://huggingface.co/papers/1909.11556 for description)
|
|
to_drop = False
|
|
if self.training:
|
|
dropout_probability = torch.rand([])
|
|
if dropout_probability < self.layerdrop: # skip the layer
|
|
to_drop = True
|
|
|
|
if to_drop:
|
|
layer_outputs = (None, None)
|
|
else:
|
|
layer_outputs = encoder_layer(
|
|
hidden_states,
|
|
attention_mask,
|
|
layer_head_mask=(head_mask[idx] if head_mask is not None else None),
|
|
self_attn_prompt=(self_attn_prompt[idx] if self.use_prompt else None),
|
|
output_attentions=output_attentions,
|
|
)
|
|
|
|
hidden_states = layer_outputs[0]
|
|
|
|
if output_attentions:
|
|
all_attentions = all_attentions + (layer_outputs[1],)
|
|
|
|
if output_hidden_states:
|
|
encoder_states = encoder_states + (hidden_states,)
|
|
|
|
if not return_dict:
|
|
return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
|
|
return BaseModelOutput(
|
|
last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
|
|
)
|
|
|
|
|
|
class MvpDecoder(MvpPreTrainedModel):
|
|
"""
|
|
Transformer decoder consisting of *config.decoder_layers* layers. Each layer is a [`MvpDecoderLayer`]
|
|
|
|
Args:
|
|
config: MvpConfig
|
|
embed_tokens (nn.Embedding): output embedding
|
|
use_prompt (bool): whether to use prompt
|
|
"""
|
|
|
|
def __init__(
|
|
self, config: MvpConfig, embed_tokens: Optional[nn.Embedding] = None, use_prompt: Optional[bool] = False
|
|
):
|
|
super().__init__(config)
|
|
self.dropout = config.dropout
|
|
self.layerdrop = config.decoder_layerdrop
|
|
self.padding_idx = config.pad_token_id
|
|
self.max_target_positions = config.max_position_embeddings
|
|
self.embed_scale = math.sqrt(config.d_model) if config.scale_embedding else 1.0
|
|
|
|
if embed_tokens is not None:
|
|
self.embed_tokens = embed_tokens
|
|
else:
|
|
self.embed_tokens = nn.Embedding(config.vocab_size, config.d_model, self.padding_idx)
|
|
|
|
self.embed_positions = MvpLearnedPositionalEmbedding(
|
|
config.max_position_embeddings,
|
|
config.d_model,
|
|
)
|
|
self.layers = nn.ModuleList([MvpDecoderLayer(config, layer_idx=i) for i in range(config.decoder_layers)])
|
|
self.layernorm_embedding = nn.LayerNorm(config.d_model)
|
|
|
|
self.use_prompt = use_prompt
|
|
if use_prompt:
|
|
self.prompt_length = config.prompt_length
|
|
self.self_attn_prompt = MvpPrompt(
|
|
config,
|
|
config.decoder_layers,
|
|
config.decoder_attention_heads,
|
|
)
|
|
self.cross_attn_prompt = MvpPrompt(
|
|
config,
|
|
config.decoder_layers,
|
|
config.decoder_attention_heads,
|
|
)
|
|
|
|
self.gradient_checkpointing = False
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.LongTensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
|
encoder_attention_mask: Optional[torch.LongTensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
|
past_key_values: Optional[list[torch.FloatTensor]] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
use_cache: Optional[bool] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
cache_position: Optional[torch.Tensor] = None,
|
|
) -> Union[tuple, BaseModelOutputWithPastAndCrossAttentions]:
|
|
r"""
|
|
Args:
|
|
input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
|
|
Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you
|
|
provide it.
|
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
|
[`PreTrainedTokenizer.__call__`] for details.
|
|
|
|
[What are input IDs?](../glossary#input-ids)
|
|
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
|
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
|
|
|
- 1 for tokens that are **not masked**,
|
|
- 0 for tokens that are **masked**.
|
|
|
|
[What are attention masks?](../glossary#attention-mask)
|
|
encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, encoder_sequence_length, hidden_size)`, *optional*):
|
|
Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention
|
|
of the decoder.
|
|
encoder_attention_mask (`torch.LongTensor` of shape `(batch_size, encoder_sequence_length)`, *optional*):
|
|
Mask to avoid performing cross-attention on padding tokens indices of encoder input_ids. Mask values
|
|
selected in `[0, 1]`:
|
|
|
|
- 1 for tokens that are **not masked**,
|
|
- 0 for tokens that are **masked**.
|
|
|
|
[What are attention masks?](../glossary#attention-mask)
|
|
head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the attention modules. Mask values selected in `[0, 1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
|
|
cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the cross-attention modules in the decoder to avoid performing
|
|
cross-attention on hidden heads. Mask values selected in `[0, 1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
|
|
past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
|
|
Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
|
|
shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of
|
|
shape `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
|
|
|
|
Contains pre-computed hidden-states (key and values in the self-attention blocks and in the
|
|
cross-attention blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
|
|
|
|
If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
|
|
that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
|
|
all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
|
|
inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
|
|
Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
|
|
This is useful if you want more control over how to convert `input_ids` indices into associated vectors
|
|
than the model's internal embedding lookup matrix.
|
|
output_attentions (`bool`, *optional*):
|
|
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
|
returned tensors for more detail.
|
|
output_hidden_states (`bool`, *optional*):
|
|
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
|
|
for more detail.
|
|
return_dict (`bool`, *optional*):
|
|
Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
|
|
"""
|
|
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
|
|
)
|
|
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
|
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
|
|
|
# retrieve input_ids and inputs_embeds
|
|
if input_ids is not None and inputs_embeds is not None:
|
|
raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
|
|
elif input_ids is not None:
|
|
input = input_ids
|
|
input_shape = input_ids.shape
|
|
input_ids = input_ids.view(-1, input_shape[-1])
|
|
elif inputs_embeds is not None:
|
|
input_shape = inputs_embeds.size()[:-1]
|
|
input = inputs_embeds[:, :, -1]
|
|
else:
|
|
raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
|
|
|
|
if inputs_embeds is None:
|
|
inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale
|
|
|
|
if self.gradient_checkpointing and self.training:
|
|
if use_cache:
|
|
logger.warning_once(
|
|
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
|
|
)
|
|
use_cache = False
|
|
|
|
return_legacy_cache = False
|
|
if use_cache and not isinstance(past_key_values, Cache):
|
|
logger.warning_once(
|
|
"Passing a tuple of `past_key_values` is deprecated and will be removed in Transformers v4.58.0. "
|
|
"You should pass an instance of `EncoderDecoderCache` instead, e.g. "
|
|
"`past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values)`."
|
|
)
|
|
return_legacy_cache = True
|
|
past_key_values = EncoderDecoderCache.from_legacy_cache(past_key_values)
|
|
|
|
past_key_values_length = past_key_values.get_seq_length() if past_key_values is not None else 0
|
|
attention_mask = _prepare_4d_causal_attention_mask(
|
|
attention_mask, input_shape, inputs_embeds, past_key_values_length
|
|
)
|
|
|
|
# expand encoder attention mask
|
|
if encoder_hidden_states is not None and encoder_attention_mask is not None:
|
|
# [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
|
|
encoder_attention_mask = _prepare_4d_attention_mask(
|
|
encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
|
|
)
|
|
|
|
# embed positions
|
|
positions = self.embed_positions(input, past_key_values_length)
|
|
|
|
hidden_states = inputs_embeds + positions
|
|
hidden_states = self.layernorm_embedding(hidden_states)
|
|
|
|
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
|
|
|
|
# layer-wise prompt
|
|
if self.use_prompt:
|
|
prompt_ids = torch.arange(self.prompt_length).to(self.device)
|
|
self_attn_prompt = self.self_attn_prompt(prompt_ids)
|
|
cross_attn_prompt = self.cross_attn_prompt(prompt_ids)
|
|
|
|
# decoder layers
|
|
all_hidden_states = () if output_hidden_states else None
|
|
all_self_attns = () if output_attentions else None
|
|
all_cross_attentions = () if (output_attentions and encoder_hidden_states is not None) else None
|
|
|
|
# check if head_mask/cross_attn_head_mask has a correct number of layers specified if desired
|
|
for attn_mask, mask_name in zip([head_mask, cross_attn_head_mask], ["head_mask", "cross_attn_head_mask"]):
|
|
if attn_mask is not None:
|
|
if attn_mask.size()[0] != (len(self.layers)):
|
|
raise ValueError(
|
|
f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
|
|
f" {head_mask.size()[0]}."
|
|
)
|
|
|
|
for idx, decoder_layer in enumerate(self.layers):
|
|
# add LayerDrop (see https://huggingface.co/papers/1909.11556 for description)
|
|
if output_hidden_states:
|
|
all_hidden_states += (hidden_states,)
|
|
if self.training:
|
|
dropout_probability = torch.rand([])
|
|
if dropout_probability < self.layerdrop:
|
|
continue
|
|
|
|
layer_outputs = decoder_layer(
|
|
hidden_states,
|
|
attention_mask,
|
|
encoder_hidden_states, # as positional argument for gradient checkpointing
|
|
encoder_attention_mask=encoder_attention_mask,
|
|
layer_head_mask=(head_mask[idx] if head_mask is not None else None),
|
|
cross_attn_layer_head_mask=(cross_attn_head_mask[idx] if cross_attn_head_mask is not None else None),
|
|
self_attn_prompt=(self_attn_prompt[idx] if self.use_prompt else None),
|
|
cross_attn_prompt=(cross_attn_prompt[idx] if self.use_prompt else None),
|
|
past_key_value=past_key_values,
|
|
output_attentions=output_attentions,
|
|
use_cache=use_cache,
|
|
cache_position=cache_position,
|
|
)
|
|
hidden_states = layer_outputs[0]
|
|
if output_attentions:
|
|
all_self_attns += (layer_outputs[1],)
|
|
|
|
if encoder_hidden_states is not None:
|
|
all_cross_attentions += (layer_outputs[2],)
|
|
|
|
# add hidden states from the last decoder layer
|
|
if output_hidden_states:
|
|
all_hidden_states += (hidden_states,)
|
|
|
|
if return_legacy_cache:
|
|
past_key_values = past_key_values.to_legacy_cache()
|
|
|
|
if not return_dict:
|
|
return tuple(
|
|
v
|
|
for v in [hidden_states, past_key_values, all_hidden_states, all_self_attns, all_cross_attentions]
|
|
if v is not None
|
|
)
|
|
return BaseModelOutputWithPastAndCrossAttentions(
|
|
last_hidden_state=hidden_states,
|
|
past_key_values=past_key_values,
|
|
hidden_states=all_hidden_states,
|
|
attentions=all_self_attns,
|
|
cross_attentions=all_cross_attentions,
|
|
)
|
|
|
|
|
|
@auto_docstring
|
|
class MvpModel(MvpPreTrainedModel):
|
|
_keys_to_ignore_on_load_unexpected = ["final_logits_bias"]
|
|
_tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
|
|
|
|
def __init__(self, config: MvpConfig):
|
|
super().__init__(config)
|
|
|
|
padding_idx, vocab_size = config.pad_token_id, config.vocab_size
|
|
self.use_prompt = config.use_prompt
|
|
self.shared = nn.Embedding(vocab_size, config.d_model, padding_idx)
|
|
|
|
self.encoder = MvpEncoder(config, self.shared, config.use_prompt)
|
|
self.decoder = MvpDecoder(config, self.shared, config.use_prompt)
|
|
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def get_input_embeddings(self):
|
|
return self.shared
|
|
|
|
def set_input_embeddings(self, value):
|
|
self.shared = value
|
|
self.encoder.embed_tokens = self.shared
|
|
self.decoder.embed_tokens = self.shared
|
|
|
|
def get_encoder(self):
|
|
return self.encoder
|
|
|
|
def get_decoder(self):
|
|
return self.decoder
|
|
|
|
def set_lightweight_tuning(self):
|
|
assert self.use_prompt, "If you want to use lightweight tuning, make sure that `use_prompt=True`."
|
|
|
|
self.requires_grad_(False)
|
|
self.encoder.self_attn_prompt.requires_grad_(True)
|
|
self.decoder.self_attn_prompt.requires_grad_(True)
|
|
self.decoder.cross_attn_prompt.requires_grad_(True)
|
|
|
|
@auto_docstring
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.LongTensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
decoder_input_ids: Optional[torch.LongTensor] = None,
|
|
decoder_attention_mask: Optional[torch.LongTensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
decoder_head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
|
encoder_outputs: Optional[list[torch.FloatTensor]] = None,
|
|
past_key_values: Optional[list[torch.FloatTensor]] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
use_cache: Optional[bool] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
cache_position: Optional[torch.Tensor] = None,
|
|
) -> Union[tuple, Seq2SeqModelOutput]:
|
|
r"""
|
|
decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Indices of decoder input sequence tokens in the vocabulary.
|
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
|
[`PreTrainedTokenizer.__call__`] for details.
|
|
|
|
[What are decoder input IDs?](../glossary#decoder-input-ids)
|
|
|
|
Mvp uses the `eos_token_id` as the starting token for `decoder_input_ids` generation. If `past_key_values`
|
|
is used, optionally only the last `decoder_input_ids` have to be input (see `past_key_values`).
|
|
|
|
For translation and summarization training, `decoder_input_ids` should be provided. If no
|
|
`decoder_input_ids` is provided, the model will create this tensor by shifting the `input_ids` to the right
|
|
for denoising pre-training following the paper.
|
|
decoder_attention_mask (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also
|
|
be used by default.
|
|
|
|
If you want to change padding behavior, you should read [`modeling_mvp._prepare_decoder_attention_mask`]
|
|
and modify to your needs. See diagram 1 in [the paper](https://huggingface.co/papers/1910.13461) for more
|
|
information on the default strategy.
|
|
cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the cross-attention modules in the decoder. Mask values selected in `[0,
|
|
1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
"""
|
|
# different to other models, Mvp automatically creates decoder_input_ids from
|
|
# input_ids if no decoder_input_ids are provided
|
|
if decoder_input_ids is None and decoder_inputs_embeds is None:
|
|
if input_ids is None:
|
|
raise ValueError(
|
|
"If no `decoder_input_ids` or `decoder_inputs_embeds` are "
|
|
"passed, `input_ids` cannot be `None`. Please pass either "
|
|
"`input_ids` or `decoder_input_ids` or `decoder_inputs_embeds`."
|
|
)
|
|
|
|
decoder_input_ids = shift_tokens_right(
|
|
input_ids, self.config.pad_token_id, self.config.decoder_start_token_id
|
|
)
|
|
|
|
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
|
|
)
|
|
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
|
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
|
|
|
if encoder_outputs is None:
|
|
encoder_outputs = self.encoder(
|
|
input_ids=input_ids,
|
|
attention_mask=attention_mask,
|
|
head_mask=head_mask,
|
|
inputs_embeds=inputs_embeds,
|
|
output_attentions=output_attentions,
|
|
output_hidden_states=output_hidden_states,
|
|
return_dict=return_dict,
|
|
)
|
|
# If the user passed a tuple for encoder_outputs, we wrap it in a BaseModelOutput when return_dict=True
|
|
elif return_dict and not isinstance(encoder_outputs, BaseModelOutput):
|
|
encoder_outputs = BaseModelOutput(
|
|
last_hidden_state=encoder_outputs[0],
|
|
hidden_states=encoder_outputs[1] if len(encoder_outputs) > 1 else None,
|
|
attentions=encoder_outputs[2] if len(encoder_outputs) > 2 else None,
|
|
)
|
|
|
|
# decoder outputs consists of (dec_features, past_key_value, dec_hidden, dec_attn)
|
|
decoder_outputs = self.decoder(
|
|
input_ids=decoder_input_ids,
|
|
attention_mask=decoder_attention_mask,
|
|
encoder_hidden_states=encoder_outputs[0],
|
|
encoder_attention_mask=attention_mask,
|
|
head_mask=decoder_head_mask,
|
|
cross_attn_head_mask=cross_attn_head_mask,
|
|
past_key_values=past_key_values,
|
|
inputs_embeds=decoder_inputs_embeds,
|
|
use_cache=use_cache,
|
|
output_attentions=output_attentions,
|
|
output_hidden_states=output_hidden_states,
|
|
return_dict=return_dict,
|
|
cache_position=cache_position,
|
|
)
|
|
|
|
if not return_dict:
|
|
return decoder_outputs + encoder_outputs
|
|
|
|
return Seq2SeqModelOutput(
|
|
last_hidden_state=decoder_outputs.last_hidden_state,
|
|
past_key_values=decoder_outputs.past_key_values,
|
|
decoder_hidden_states=decoder_outputs.hidden_states,
|
|
decoder_attentions=decoder_outputs.attentions,
|
|
cross_attentions=decoder_outputs.cross_attentions,
|
|
encoder_last_hidden_state=encoder_outputs.last_hidden_state,
|
|
encoder_hidden_states=encoder_outputs.hidden_states,
|
|
encoder_attentions=encoder_outputs.attentions,
|
|
)
|
|
|
|
|
|
@auto_docstring(
|
|
custom_intro="""
|
|
The MVP Model with a language modeling head. Can be used for various text generation tasks.
|
|
"""
|
|
)
|
|
class MvpForConditionalGeneration(MvpPreTrainedModel, GenerationMixin):
|
|
_tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight", "lm_head.weight"]
|
|
|
|
def __init__(self, config: MvpConfig):
|
|
super().__init__(config)
|
|
self.model = MvpModel(config)
|
|
self.register_buffer("final_logits_bias", torch.zeros((1, self.model.shared.num_embeddings)))
|
|
self.lm_head = nn.Linear(config.d_model, self.model.shared.num_embeddings, bias=False)
|
|
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def get_encoder(self):
|
|
return self.model.get_encoder()
|
|
|
|
def get_decoder(self):
|
|
return self.model.get_decoder()
|
|
|
|
def resize_token_embeddings(
|
|
self, new_num_tokens: int, pad_to_multiple_of: Optional[int] = None, mean_resizing: bool = True
|
|
) -> nn.Embedding:
|
|
new_embeddings = super().resize_token_embeddings(new_num_tokens, pad_to_multiple_of, mean_resizing)
|
|
self._resize_final_logits_bias(new_num_tokens)
|
|
return new_embeddings
|
|
|
|
def _resize_final_logits_bias(self, new_num_tokens: int) -> None:
|
|
old_num_tokens = self.final_logits_bias.shape[-1]
|
|
if new_num_tokens <= old_num_tokens:
|
|
new_bias = self.final_logits_bias[:, :new_num_tokens]
|
|
else:
|
|
extra_bias = torch.zeros((1, new_num_tokens - old_num_tokens), device=self.final_logits_bias.device)
|
|
new_bias = torch.cat([self.final_logits_bias, extra_bias], dim=1)
|
|
self.register_buffer("final_logits_bias", new_bias)
|
|
|
|
def set_lightweight_tuning(self):
|
|
self.model.set_lightweight_tuning()
|
|
self.lm_head.requires_grad_(False)
|
|
|
|
@auto_docstring
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.LongTensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
decoder_input_ids: Optional[torch.LongTensor] = None,
|
|
decoder_attention_mask: Optional[torch.LongTensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
decoder_head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
|
encoder_outputs: Optional[list[torch.FloatTensor]] = None,
|
|
past_key_values: Optional[list[torch.FloatTensor]] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
labels: Optional[torch.LongTensor] = None,
|
|
use_cache: Optional[bool] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
cache_position: Optional[torch.Tensor] = None,
|
|
) -> Union[tuple, Seq2SeqLMOutput]:
|
|
r"""
|
|
decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Indices of decoder input sequence tokens in the vocabulary.
|
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
|
[`PreTrainedTokenizer.__call__`] for details.
|
|
|
|
[What are decoder input IDs?](../glossary#decoder-input-ids)
|
|
|
|
Mvp uses the `eos_token_id` as the starting token for `decoder_input_ids` generation. If `past_key_values`
|
|
is used, optionally only the last `decoder_input_ids` have to be input (see `past_key_values`).
|
|
|
|
For translation and summarization training, `decoder_input_ids` should be provided. If no
|
|
`decoder_input_ids` is provided, the model will create this tensor by shifting the `input_ids` to the right
|
|
for denoising pre-training following the paper.
|
|
decoder_attention_mask (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also
|
|
be used by default.
|
|
|
|
If you want to change padding behavior, you should read [`modeling_mvp._prepare_decoder_attention_mask`]
|
|
and modify to your needs. See diagram 1 in [the paper](https://huggingface.co/papers/1910.13461) for more
|
|
information on the default strategy.
|
|
cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the cross-attention modules in the decoder. Mask values selected in `[0,
|
|
1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
|
Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
|
|
config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
|
|
(masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
|
|
|
|
Example of summarization:
|
|
|
|
Fine-tuning a model
|
|
```python
|
|
>>> import torch
|
|
>>> from transformers import AutoTokenizer, MvpForConditionalGeneration
|
|
|
|
>>> tokenizer = AutoTokenizer.from_pretrained("RUCAIBox/mvp")
|
|
>>> model = MvpForConditionalGeneration.from_pretrained("RUCAIBox/mvp")
|
|
|
|
>>> inputs = tokenizer(
|
|
... "Summarize: You may want to stick it to your boss and leave your job, but don't do it if these are your reasons.",
|
|
... return_tensors="pt",
|
|
... )
|
|
>>> labels = tokenizer("Bad Reasons To Quit Your Job", return_tensors="pt")["input_ids"]
|
|
|
|
>>> loss = model(**inputs, labels=labels).loss
|
|
>>> loss.backward()
|
|
```
|
|
|
|
Inference after the model fine-tuned
|
|
```python
|
|
>>> with torch.no_grad():
|
|
... generated_ids = model.generate(**inputs)
|
|
|
|
>>> generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
|
|
```
|
|
"""
|
|
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
|
|
|
if labels is not None:
|
|
if use_cache:
|
|
logger.warning("The `use_cache` argument is changed to `False` since `labels` is provided.")
|
|
use_cache = False
|
|
if decoder_input_ids is None and decoder_inputs_embeds is None:
|
|
decoder_input_ids = shift_tokens_right(
|
|
labels, self.config.pad_token_id, self.config.decoder_start_token_id
|
|
)
|
|
|
|
outputs = self.model(
|
|
input_ids,
|
|
attention_mask=attention_mask,
|
|
decoder_input_ids=decoder_input_ids,
|
|
encoder_outputs=encoder_outputs,
|
|
decoder_attention_mask=decoder_attention_mask,
|
|
head_mask=head_mask,
|
|
decoder_head_mask=decoder_head_mask,
|
|
cross_attn_head_mask=cross_attn_head_mask,
|
|
past_key_values=past_key_values,
|
|
inputs_embeds=inputs_embeds,
|
|
decoder_inputs_embeds=decoder_inputs_embeds,
|
|
use_cache=use_cache,
|
|
output_attentions=output_attentions,
|
|
output_hidden_states=output_hidden_states,
|
|
return_dict=return_dict,
|
|
cache_position=cache_position,
|
|
)
|
|
lm_logits = self.lm_head(outputs[0]) + self.final_logits_bias
|
|
|
|
masked_lm_loss = None
|
|
if labels is not None:
|
|
loss_fct = CrossEntropyLoss()
|
|
masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.vocab_size), labels.view(-1))
|
|
|
|
if not return_dict:
|
|
output = (lm_logits,) + outputs[1:]
|
|
return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
|
|
|
|
return Seq2SeqLMOutput(
|
|
loss=masked_lm_loss,
|
|
logits=lm_logits,
|
|
past_key_values=outputs.past_key_values,
|
|
decoder_hidden_states=outputs.decoder_hidden_states,
|
|
decoder_attentions=outputs.decoder_attentions,
|
|
cross_attentions=outputs.cross_attentions,
|
|
encoder_last_hidden_state=outputs.encoder_last_hidden_state,
|
|
encoder_hidden_states=outputs.encoder_hidden_states,
|
|
encoder_attentions=outputs.encoder_attentions,
|
|
)
|
|
|
|
def prepare_decoder_input_ids_from_labels(self, labels: torch.Tensor):
|
|
return shift_tokens_right(labels, self.config.pad_token_id, self.config.decoder_start_token_id)
|
|
|
|
|
|
@auto_docstring(
|
|
custom_intro="""
|
|
Mvp model with a sequence classification/head on top (a linear layer on top of the pooled output) e.g. for GLUE
|
|
tasks.
|
|
"""
|
|
)
|
|
class MvpForSequenceClassification(MvpPreTrainedModel):
|
|
_tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
|
|
|
|
def __init__(self, config: MvpConfig, **kwargs):
|
|
super().__init__(config, **kwargs)
|
|
self.model = MvpModel(config)
|
|
self.classification_head = MvpClassificationHead(
|
|
config.d_model,
|
|
config.d_model,
|
|
config.num_labels,
|
|
config.classifier_dropout,
|
|
)
|
|
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def set_lightweight_tuning(self):
|
|
self.model.set_lightweight_tuning()
|
|
self.classification_head.requires_grad_(False)
|
|
|
|
@auto_docstring
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.LongTensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
decoder_input_ids: Optional[torch.LongTensor] = None,
|
|
decoder_attention_mask: Optional[torch.LongTensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
decoder_head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
|
encoder_outputs: Optional[list[torch.FloatTensor]] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
labels: Optional[torch.LongTensor] = None,
|
|
use_cache: Optional[bool] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
) -> Union[tuple, Seq2SeqSequenceClassifierOutput]:
|
|
r"""
|
|
decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Indices of decoder input sequence tokens in the vocabulary.
|
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
|
[`PreTrainedTokenizer.__call__`] for details.
|
|
|
|
[What are decoder input IDs?](../glossary#decoder-input-ids)
|
|
|
|
Mvp uses the `eos_token_id` as the starting token for `decoder_input_ids` generation. If `past_key_values`
|
|
is used, optionally only the last `decoder_input_ids` have to be input (see `past_key_values`).
|
|
|
|
For translation and summarization training, `decoder_input_ids` should be provided. If no
|
|
`decoder_input_ids` is provided, the model will create this tensor by shifting the `input_ids` to the right
|
|
for denoising pre-training following the paper.
|
|
decoder_attention_mask (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also
|
|
be used by default.
|
|
|
|
If you want to change padding behavior, you should read [`modeling_mvp._prepare_decoder_attention_mask`]
|
|
and modify to your needs. See diagram 1 in [the paper](https://huggingface.co/papers/1910.13461) for more
|
|
information on the default strategy.
|
|
cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the cross-attention modules in the decoder. Mask values selected in `[0,
|
|
1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
|
|
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
|
|
config.num_labels - 1]`. If `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
|
|
|
|
Example of single-label classification:
|
|
|
|
Fine-tuning a model on `num_labels` classes
|
|
```python
|
|
>>> import torch
|
|
>>> from transformers import AutoTokenizer, MvpForSequenceClassification
|
|
|
|
>>> num_labels = 2 # for example, this is a binary classification task
|
|
>>> tokenizer = AutoTokenizer.from_pretrained("RUCAIBox/mvp")
|
|
>>> model = MvpForSequenceClassification.from_pretrained("RUCAIBox/mvp", num_labels=num_labels)
|
|
|
|
>>> inputs = tokenizer("Classify: Hello, my dog is cute", return_tensors="pt")
|
|
>>> labels = torch.tensor(1) # the real label for inputs
|
|
|
|
>>> loss = model(**inputs, labels=labels).loss
|
|
>>> loss.backward()
|
|
```
|
|
|
|
Inference after the model fine-tuned
|
|
```python
|
|
>>> with torch.no_grad():
|
|
... logits = model(**inputs).logits
|
|
|
|
>>> predicted_class_id = logits.argmax()
|
|
```
|
|
"""
|
|
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
|
if labels is not None:
|
|
use_cache = False
|
|
|
|
if input_ids is None and inputs_embeds is not None:
|
|
raise NotImplementedError(
|
|
f"Passing input embeddings is currently not supported for {self.__class__.__name__}"
|
|
)
|
|
|
|
outputs = self.model(
|
|
input_ids,
|
|
attention_mask=attention_mask,
|
|
decoder_input_ids=decoder_input_ids,
|
|
decoder_attention_mask=decoder_attention_mask,
|
|
head_mask=head_mask,
|
|
decoder_head_mask=decoder_head_mask,
|
|
cross_attn_head_mask=cross_attn_head_mask,
|
|
encoder_outputs=encoder_outputs,
|
|
inputs_embeds=inputs_embeds,
|
|
decoder_inputs_embeds=decoder_inputs_embeds,
|
|
use_cache=use_cache,
|
|
output_attentions=output_attentions,
|
|
output_hidden_states=output_hidden_states,
|
|
return_dict=return_dict,
|
|
)
|
|
hidden_states = outputs[0] # last hidden state
|
|
|
|
eos_mask = input_ids.eq(self.config.eos_token_id).to(hidden_states.device)
|
|
|
|
if len(torch.unique_consecutive(eos_mask.sum(1))) > 1:
|
|
raise ValueError("All examples must have the same number of <eos> tokens.")
|
|
sentence_representation = hidden_states[eos_mask, :].view(hidden_states.size(0), -1, hidden_states.size(-1))[
|
|
:, -1, :
|
|
]
|
|
logits = self.classification_head(sentence_representation)
|
|
|
|
loss = None
|
|
if labels is not None:
|
|
if self.config.problem_type is None:
|
|
if self.config.num_labels == 1:
|
|
self.config.problem_type = "regression"
|
|
elif self.config.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
|
|
self.config.problem_type = "single_label_classification"
|
|
else:
|
|
self.config.problem_type = "multi_label_classification"
|
|
|
|
if self.config.problem_type == "regression":
|
|
loss_fct = MSELoss()
|
|
if self.config.num_labels == 1:
|
|
loss = loss_fct(logits.squeeze(), labels.squeeze())
|
|
else:
|
|
loss = loss_fct(logits, labels)
|
|
elif self.config.problem_type == "single_label_classification":
|
|
loss_fct = CrossEntropyLoss()
|
|
loss = loss_fct(logits.view(-1, self.config.num_labels), labels.view(-1))
|
|
elif self.config.problem_type == "multi_label_classification":
|
|
loss_fct = BCEWithLogitsLoss()
|
|
loss = loss_fct(logits, labels)
|
|
if not return_dict:
|
|
output = (logits,) + outputs[1:]
|
|
return ((loss,) + output) if loss is not None else output
|
|
|
|
return Seq2SeqSequenceClassifierOutput(
|
|
loss=loss,
|
|
logits=logits,
|
|
past_key_values=outputs.past_key_values,
|
|
decoder_hidden_states=outputs.decoder_hidden_states,
|
|
decoder_attentions=outputs.decoder_attentions,
|
|
cross_attentions=outputs.cross_attentions,
|
|
encoder_last_hidden_state=outputs.encoder_last_hidden_state,
|
|
encoder_hidden_states=outputs.encoder_hidden_states,
|
|
encoder_attentions=outputs.encoder_attentions,
|
|
)
|
|
|
|
|
|
@auto_docstring
|
|
class MvpForQuestionAnswering(MvpPreTrainedModel):
|
|
_tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
|
|
|
|
def __init__(self, config):
|
|
super().__init__(config)
|
|
|
|
config.num_labels = 2
|
|
self.num_labels = config.num_labels
|
|
|
|
self.model = MvpModel(config)
|
|
self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
|
|
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def set_lightweight_tuning(self):
|
|
self.model.set_lightweight_tuning()
|
|
self.qa_outputs.requires_grad_(False)
|
|
|
|
@auto_docstring
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.Tensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
decoder_input_ids: Optional[torch.LongTensor] = None,
|
|
decoder_attention_mask: Optional[torch.LongTensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
decoder_head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
|
encoder_outputs: Optional[list[torch.FloatTensor]] = None,
|
|
start_positions: Optional[torch.LongTensor] = None,
|
|
end_positions: Optional[torch.LongTensor] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
decoder_inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
use_cache: Optional[bool] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
) -> Union[tuple, Seq2SeqQuestionAnsweringModelOutput]:
|
|
r"""
|
|
decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Indices of decoder input sequence tokens in the vocabulary.
|
|
|
|
Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
|
|
[`PreTrainedTokenizer.__call__`] for details.
|
|
|
|
[What are decoder input IDs?](../glossary#decoder-input-ids)
|
|
|
|
Mvp uses the `eos_token_id` as the starting token for `decoder_input_ids` generation. If `past_key_values`
|
|
is used, optionally only the last `decoder_input_ids` have to be input (see `past_key_values`).
|
|
|
|
For translation and summarization training, `decoder_input_ids` should be provided. If no
|
|
`decoder_input_ids` is provided, the model will create this tensor by shifting the `input_ids` to the right
|
|
for denoising pre-training following the paper.
|
|
decoder_attention_mask (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
|
|
Default behavior: generate a tensor that ignores pad tokens in `decoder_input_ids`. Causal mask will also
|
|
be used by default.
|
|
|
|
If you want to change padding behavior, you should read [`modeling_mvp._prepare_decoder_attention_mask`]
|
|
and modify to your needs. See diagram 1 in [the paper](https://huggingface.co/papers/1910.13461) for more
|
|
information on the default strategy.
|
|
cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the cross-attention modules in the decoder. Mask values selected in `[0,
|
|
1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
|
|
Example:
|
|
|
|
Fine-tuning a model for extrative question answering, and our model also supports generative question answering
|
|
using `BartForConditionalGeneration`
|
|
```python
|
|
>>> import torch
|
|
>>> from transformers import AutoTokenizer, MvpForQuestionAnswering
|
|
|
|
>>> tokenizer = AutoTokenizer.from_pretrained("RUCAIBox/mvp")
|
|
>>> model = MvpForQuestionAnswering.from_pretrained("RUCAIBox/mvp")
|
|
|
|
>>> inputs = tokenizer(
|
|
... "Answer the following question: Who was Jim Henson? [SEP] Jim Henson was a nice puppet",
|
|
... return_tensors="pt",
|
|
... )
|
|
>>> target_start_index = torch.tensor([18])
|
|
>>> target_end_index = torch.tensor([19])
|
|
|
|
>>> loss = model(**inputs, start_positions=target_start_index, end_positions=target_end_index).loss
|
|
>>> loss.backward()
|
|
```
|
|
|
|
Inference after the model fine-tuned
|
|
```python
|
|
>>> with torch.no_grad():
|
|
... outputs = model(**inputs)
|
|
|
|
>>> answer_start_index = outputs.start_logits.argmax()
|
|
>>> answer_end_index = outputs.end_logits.argmax()
|
|
|
|
>>> predict_answer_tokens = inputs.input_ids[0, answer_start_index : answer_end_index + 1]
|
|
>>> predict_answer = tokenizer.decode(predict_answer_tokens)
|
|
```
|
|
"""
|
|
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
|
if start_positions is not None and end_positions is not None:
|
|
use_cache = False
|
|
|
|
outputs = self.model(
|
|
input_ids,
|
|
attention_mask=attention_mask,
|
|
decoder_input_ids=decoder_input_ids,
|
|
decoder_attention_mask=decoder_attention_mask,
|
|
head_mask=head_mask,
|
|
decoder_head_mask=decoder_head_mask,
|
|
cross_attn_head_mask=cross_attn_head_mask,
|
|
encoder_outputs=encoder_outputs,
|
|
inputs_embeds=inputs_embeds,
|
|
decoder_inputs_embeds=decoder_inputs_embeds,
|
|
use_cache=use_cache,
|
|
output_attentions=output_attentions,
|
|
output_hidden_states=output_hidden_states,
|
|
return_dict=return_dict,
|
|
)
|
|
|
|
sequence_output = outputs[0]
|
|
|
|
logits = self.qa_outputs(sequence_output)
|
|
start_logits, end_logits = logits.split(1, dim=-1)
|
|
start_logits = start_logits.squeeze(-1).contiguous()
|
|
end_logits = end_logits.squeeze(-1).contiguous()
|
|
|
|
total_loss = None
|
|
if start_positions is not None and end_positions is not None:
|
|
# If we are on multi-GPU, split add a dimension
|
|
if len(start_positions.size()) > 1:
|
|
start_positions = start_positions.squeeze(-1)
|
|
if len(end_positions.size()) > 1:
|
|
end_positions = end_positions.squeeze(-1)
|
|
# sometimes the start/end positions are outside our model inputs, we ignore these terms
|
|
ignored_index = start_logits.size(1)
|
|
start_positions = start_positions.clamp(0, ignored_index)
|
|
end_positions = end_positions.clamp(0, ignored_index)
|
|
|
|
loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
|
|
start_loss = loss_fct(start_logits, start_positions)
|
|
end_loss = loss_fct(end_logits, end_positions)
|
|
total_loss = (start_loss + end_loss) / 2
|
|
|
|
if not return_dict:
|
|
output = (
|
|
start_logits,
|
|
end_logits,
|
|
) + outputs[1:]
|
|
return ((total_loss,) + output) if total_loss is not None else output
|
|
|
|
return Seq2SeqQuestionAnsweringModelOutput(
|
|
loss=total_loss,
|
|
start_logits=start_logits,
|
|
end_logits=end_logits,
|
|
past_key_values=outputs.past_key_values,
|
|
decoder_hidden_states=outputs.decoder_hidden_states,
|
|
decoder_attentions=outputs.decoder_attentions,
|
|
cross_attentions=outputs.cross_attentions,
|
|
encoder_last_hidden_state=outputs.encoder_last_hidden_state,
|
|
encoder_hidden_states=outputs.encoder_hidden_states,
|
|
encoder_attentions=outputs.encoder_attentions,
|
|
)
|
|
|
|
|
|
# Copied from transformers.models.bart.modeling_bart.BartDecoderWrapper with Bart->Mvp
|
|
class MvpDecoderWrapper(MvpPreTrainedModel):
|
|
"""
|
|
This wrapper class is a helper class to correctly load pretrained checkpoints when the causal language model is
|
|
used in combination with the [`EncoderDecoderModel`] framework.
|
|
"""
|
|
|
|
def __init__(self, config):
|
|
super().__init__(config)
|
|
self.decoder = MvpDecoder(config)
|
|
|
|
def forward(self, *args, **kwargs):
|
|
return self.decoder(*args, **kwargs)
|
|
|
|
|
|
class MvpForCausalLM(MvpPreTrainedModel, GenerationMixin):
|
|
_tied_weights_keys = ["lm_head.weight"]
|
|
|
|
def __init__(self, config):
|
|
config.is_decoder = True
|
|
config.is_encoder_decoder = False
|
|
super().__init__(config)
|
|
self.model = MvpDecoderWrapper(config)
|
|
|
|
self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
|
|
|
|
# Initialize weights and apply final processing
|
|
self.post_init()
|
|
|
|
def get_input_embeddings(self):
|
|
return self.model.decoder.embed_tokens
|
|
|
|
def set_input_embeddings(self, value):
|
|
self.model.decoder.embed_tokens = value
|
|
|
|
def set_decoder(self, decoder):
|
|
self.model.decoder = decoder
|
|
|
|
def get_decoder(self):
|
|
return self.model.decoder
|
|
|
|
def set_lightweight_tuning(self):
|
|
self.model.set_lightweight_tuning()
|
|
self.lm_head.requires_grad_(False)
|
|
|
|
@auto_docstring
|
|
def forward(
|
|
self,
|
|
input_ids: Optional[torch.LongTensor] = None,
|
|
attention_mask: Optional[torch.Tensor] = None,
|
|
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
|
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
|
head_mask: Optional[torch.Tensor] = None,
|
|
cross_attn_head_mask: Optional[torch.Tensor] = None,
|
|
past_key_values: Optional[list[torch.FloatTensor]] = None,
|
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
labels: Optional[torch.LongTensor] = None,
|
|
use_cache: Optional[bool] = None,
|
|
output_attentions: Optional[bool] = None,
|
|
output_hidden_states: Optional[bool] = None,
|
|
return_dict: Optional[bool] = None,
|
|
cache_position: Optional[torch.Tensor] = None,
|
|
) -> Union[tuple, CausalLMOutputWithCrossAttentions]:
|
|
r"""
|
|
cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*):
|
|
Mask to nullify selected heads of the cross-attention modules. Mask values selected in `[0, 1]`:
|
|
|
|
- 1 indicates the head is **not masked**,
|
|
- 0 indicates the head is **masked**.
|
|
labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
|
|
Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
|
|
config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
|
|
(masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
|
|
|
|
Example:
|
|
|
|
```python
|
|
>>> from transformers import AutoTokenizer, MvpForCausalLM
|
|
|
|
>>> tokenizer = AutoTokenizer.from_pretrained("RUCAIBox/mvp")
|
|
>>> model = MvpForCausalLM.from_pretrained("RUCAIBox/mvp", add_cross_attention=False)
|
|
|
|
>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="pt")
|
|
>>> outputs = model(**inputs)
|
|
|
|
>>> logits = outputs.logits
|
|
>>> list(logits.shape)
|
|
[1, 8, 50267]
|
|
```"""
|
|
|
|
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 = return_dict if return_dict is not None else self.config.use_return_dict
|
|
|
|
# decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
|
|
outputs = self.model.decoder(
|
|
input_ids=input_ids,
|
|
attention_mask=attention_mask,
|
|
encoder_hidden_states=encoder_hidden_states,
|
|
encoder_attention_mask=encoder_attention_mask,
|
|
head_mask=head_mask,
|
|
cross_attn_head_mask=cross_attn_head_mask,
|
|
past_key_values=past_key_values,
|
|
inputs_embeds=inputs_embeds,
|
|
use_cache=use_cache,
|
|
output_attentions=output_attentions,
|
|
output_hidden_states=output_hidden_states,
|
|
return_dict=return_dict,
|
|
)
|
|
|
|
logits = self.lm_head(outputs[0])
|
|
|
|
loss = None
|
|
if labels is not None:
|
|
loss_fct = CrossEntropyLoss()
|
|
loss = loss_fct(logits.view(-1, self.config.vocab_size), labels.view(-1))
|
|
|
|
if not return_dict:
|
|
output = (logits,) + outputs[1:]
|
|
return (loss,) + output if loss is not None else output
|
|
|
|
return CausalLMOutputWithCrossAttentions(
|
|
loss=loss,
|
|
logits=logits,
|
|
past_key_values=outputs.past_key_values,
|
|
hidden_states=outputs.hidden_states,
|
|
attentions=outputs.attentions,
|
|
cross_attentions=outputs.cross_attentions,
|
|
)
|
|
|
|
|
|
__all__ = [
|
|
"MvpForCausalLM",
|
|
"MvpForConditionalGeneration",
|
|
"MvpForQuestionAnswering",
|
|
"MvpForSequenceClassification",
|
|
"MvpModel",
|
|
"MvpPreTrainedModel",
|
|
]
|