# coding=utf-8 # This file has been modified from the modeling_roformer.py file in the transformers library. The original RoPE implementation has been replaced with the LLaMA style RoPE implementation. # Copyright 2021 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 E5Rope model.""" import math import random import os from typing import Optional, Tuple, Union import numpy as np import torch import torch.utils.checkpoint import xformers.ops as xops from torch import nn from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss from transformers.activations import ACT2FN from transformers.modeling_outputs import ( BaseModelOutputWithPastAndCrossAttentions, CausalLMOutputWithCrossAttentions, MaskedLMOutput, MultipleChoiceModelOutput, QuestionAnsweringModelOutput, SequenceClassifierOutput, TokenClassifierOutput, ) from transformers.modeling_utils import PreTrainedModel, SequenceSummary from transformers.pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer from transformers.utils import ( add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging, replace_return_docstrings, ) from .configuration_e5rope import E5RopeConfig logger = logging.get_logger(__name__) class E5RopeRotaryEmbedding(torch.nn.Module): def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None): super().__init__() self.dim = dim self.max_position_embeddings = max_position_embeddings self.base = base inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim)) self.register_buffer("inv_freq", inv_freq, persistent=False) # Build here to make `torch.jit.trace` work. self._set_cos_sin_cache( seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype() ) def _set_cos_sin_cache(self, seq_len, device, dtype): self.max_seq_len_cached = seq_len # t = torch.arange(self.max_seq_len_cached, device=device, dtype=self.inv_freq.dtype) t = np.arange(self.max_seq_len_cached, dtype=np.float64) t = torch.tensor(t, device=self.inv_freq.device, dtype=torch.float64) # freqs = torch.einsum("i,j->ij", t, self.inv_freq) freqs = torch.outer(t, self.inv_freq.to(device=t.device).to(t.dtype)) # Different from paper, but it uses a different permutation in order to obtain the same calculation emb = torch.cat((freqs, freqs), dim=-1) self.register_buffer("cos_cached", emb.cos()[None, None, :, :].to(dtype), persistent=False) self.register_buffer("sin_cached", emb.sin()[None, None, :, :].to(dtype), persistent=False) def forward(self, x, seq_len=None): # x: [bs, num_attention_heads, seq_len, head_size] if seq_len > self.max_seq_len_cached: self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype) return ( self.cos_cached[:, :, :, ...].to(dtype=x.dtype), self.sin_cached[:, :, :, ...].to(dtype=x.dtype), ) def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor: """ This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim) """ batch, num_key_value_heads, slen, head_dim = hidden_states.shape if n_rep == 1: return hidden_states hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim) return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim) def rotate_half(x): """Rotates half the hidden dims of the input.""" x1 = x[..., : x.shape[-1] // 2] x2 = x[..., x.shape[-1] // 2 :] return torch.cat((-x2, x1), dim=-1) def apply_rotary_pos_emb(q, k, cos, sin, position_ids): # The first two dimensions of cos and sin are always 1, so we can `squeeze` them. cos = cos.squeeze(1).squeeze(0) # [seq_len, dim] sin = sin.squeeze(1).squeeze(0) # [seq_len, dim] cos = cos[position_ids].unsqueeze(1) # [bs, 1, seq_len, dim] sin = sin[position_ids].unsqueeze(1) # [bs, 1, seq_len, dim] q_embed = (q * cos) + (rotate_half(q) * sin) k_embed = (k * cos) + (rotate_half(k) * sin) return q_embed, k_embed def load_tf_weights_in_e5rope(model, config, tf_checkpoint_path): """Load tf checkpoints in a pytorch model.""" try: import re import numpy as np import tensorflow as tf except ImportError: logger.error( "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see " "https://www.tensorflow.org/install/ for installation instructions." ) raise tf_path = os.path.abspath(tf_checkpoint_path) logger.info(f"Converting TensorFlow checkpoint from {tf_path}") # Load weights from TF model init_vars = tf.train.list_variables(tf_path) names = [] arrays = [] for name, shape in init_vars: logger.info(f"Loading TF weight {name} with shape {shape}") array = tf.train.load_variable(tf_path, name) names.append(name.replace("bert", "e5rope")) arrays.append(array) for name, array in zip(names, arrays): name = name.split("/") # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v # which are not required for using pretrained model if any( n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"] for n in name ): logger.info(f"Skipping {'/'.join(name)}") continue pointer = model for m_name in name: if re.fullmatch(r"[A-Za-z]+_\d+", m_name): scope_names = re.split(r"_(\d+)", m_name) else: scope_names = [m_name] if scope_names[0] == "kernel" or scope_names[0] == "gamma": pointer = getattr(pointer, "weight") elif scope_names[0] == "output_bias" or scope_names[0] == "beta": pointer = getattr(pointer, "bias") elif scope_names[0] == "output_weights": pointer = getattr(pointer, "weight") elif scope_names[0] == "squad": pointer = getattr(pointer, "classifier") else: try: pointer = getattr(pointer, scope_names[0]) except AttributeError: logger.info(f"Skipping {'/'.join(name)}") continue if len(scope_names) >= 2: num = int(scope_names[1]) pointer = pointer[num] if m_name[-11:] == "_embeddings": pointer = getattr(pointer, "weight") elif m_name == "kernel": array = np.transpose(array) try: if not pointer.shape == array.shape: raise ValueError(f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched") except AssertionError as e: e.args += (pointer.shape, array.shape) raise logger.info(f"Initialize PyTorch weight {name}") pointer.data = torch.from_numpy(array) return model class E5RopeEmbeddings(nn.Module): """Construct the embeddings from word and token_type embeddings.""" def __init__(self, config): super().__init__() self.word_embeddings = nn.Embedding(config.vocab_size, config.embedding_size, padding_idx=config.pad_token_id) self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.embedding_size) # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load # any TensorFlow checkpoint file self.LayerNorm = nn.LayerNorm(config.embedding_size, eps=config.layer_norm_eps) self.dropout = nn.Dropout(config.hidden_dropout_prob) def forward(self, input_ids=None, token_type_ids=None, inputs_embeds=None): if input_ids is not None: input_shape = input_ids.size() else: input_shape = inputs_embeds.size()[:-1] if inputs_embeds is None: inputs_embeds = self.word_embeddings(input_ids) if token_type_ids is None: token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=inputs_embeds.device) token_type_embeddings = self.token_type_embeddings(token_type_ids) embeddings = inputs_embeds + token_type_embeddings embeddings = self.LayerNorm(embeddings) embeddings = self.dropout(embeddings) return embeddings class E5RopeSelfAttention(nn.Module): def __init__(self, config): super().__init__() if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"): raise ValueError( f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention " f"heads ({config.num_attention_heads})" ) self.num_attention_heads = config.num_attention_heads self.attention_head_size = int(config.hidden_size / config.num_attention_heads) self.all_head_size = self.num_attention_heads * self.attention_head_size self.query = nn.Linear(config.hidden_size, self.all_head_size) self.key = nn.Linear(config.hidden_size, self.all_head_size) self.value = nn.Linear(config.hidden_size, self.all_head_size) self.dropout = nn.Dropout(config.attention_probs_dropout_prob) self.is_decoder = config.is_decoder self.config = config self.max_position_embeddings = config.max_position_embeddings self.rope_theta = config.rope_theta self.rotary_emb = E5RopeRotaryEmbedding( self.attention_head_size, max_position_embeddings=self.max_position_embeddings, base=self.rope_theta, ) # self.forward = self.normal_forward def transpose_for_scores(self, x): new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size) x = x.view(*new_x_shape) return x.permute(0, 2, 1, 3) def forward( self, hidden_states, attention_mask=None, position_ids=None, head_mask=None, encoder_hidden_states=None, encoder_attention_mask=None, past_key_value=None, output_attentions=False, ): mixed_query_layer = self.query(hidden_states) query_layer = self.transpose_for_scores(mixed_query_layer) # If this is instantiated as a cross-attention module, the keys # and values come from an encoder; the attention mask needs to be # such that the encoder's padding tokens are not attended to. is_cross_attention = encoder_hidden_states is not None if is_cross_attention and past_key_value is not None: # reuse k,v, cross_attentions key_layer = past_key_value[0] value_layer = past_key_value[1] attention_mask = encoder_attention_mask elif is_cross_attention: key_layer = self.transpose_for_scores(self.key(encoder_hidden_states)) value_layer = self.transpose_for_scores(self.value(encoder_hidden_states)) attention_mask = encoder_attention_mask else: key_layer = self.transpose_for_scores(self.key(hidden_states)) value_layer = self.transpose_for_scores(self.value(hidden_states)) kv_seq_len = key_layer.shape[-2] if past_key_value is not None: kv_seq_len += past_key_value[0].shape[-2] cos, sin = self.rotary_emb(value_layer, seq_len=kv_seq_len) query_layer, key_layer = apply_rotary_pos_emb(query_layer, key_layer, cos, sin, position_ids) if past_key_value is not None: key_layer = torch.cat([past_key_value[0], key_layer], dim=2) value_layer = torch.cat([past_key_value[1], value_layer], dim=2) if self.is_decoder: # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states. # Further calls to cross_attention layer can then reuse all cross-attention # key/value_states (first "if" case) # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of # all previous decoder key/value_states. Further calls to uni-directional self-attention # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case) # if encoder bi-directional self-attention `past_key_value` is always `None` past_key_value = (key_layer, value_layer) bsz, n_heads, seq_len, head_dim = query_layer.shape # get each seq len tmp_attention_mask = attention_mask.squeeze() if tmp_attention_mask.dim() == 1: tmp_attention_mask = tmp_attention_mask.unsqueeze(0) each_seq_len = torch.sum(tmp_attention_mask == 0, dim=-1) original_len = torch.tensor(512) ### attention scaling for better length extrapolation ### ### https://arxiv.org/abs/2202.12172 ; https://kexue.fm/archives/8823 ### attn_factors = torch.log(each_seq_len) / torch.log(original_len) attn_factors = torch.clamp(attn_factors, min=1.0) # Ensure a minimum value of 1 attn_factors = attn_factors.view(-1, 1, 1, 1) query_layer *= attn_factors attention_mask = attention_mask.expand(bsz, n_heads, seq_len, seq_len).to(dtype=query_layer.dtype) attn_output = xops.memory_efficient_attention( query_layer.transpose(1, 2), key_layer.transpose(1, 2), value_layer.transpose(1, 2), attn_bias=attention_mask, p=(self.dropout.p if self.training else 0) ).reshape(bsz, seq_len, n_heads * head_dim) if output_attentions is True: raise NotImplementedError('output_attentions is not supported for xformers attention') return (attn_output,) def normal_forward( self, hidden_states, attention_mask=None, position_ids=None, head_mask=None, encoder_hidden_states=None, encoder_attention_mask=None, past_key_value=None, output_attentions=False, ): mixed_query_layer = self.query(hidden_states) query_layer = self.transpose_for_scores(mixed_query_layer) # If this is instantiated as a cross-attention module, the keys # and values come from an encoder; the attention mask needs to be # such that the encoder's padding tokens are not attended to. is_cross_attention = encoder_hidden_states is not None if is_cross_attention and past_key_value is not None: # reuse k,v, cross_attentions key_layer = past_key_value[0] value_layer = past_key_value[1] attention_mask = encoder_attention_mask elif is_cross_attention: key_layer = self.transpose_for_scores(self.key(encoder_hidden_states)) value_layer = self.transpose_for_scores(self.value(encoder_hidden_states)) attention_mask = encoder_attention_mask else: key_layer = self.transpose_for_scores(self.key(hidden_states)) value_layer = self.transpose_for_scores(self.value(hidden_states)) kv_seq_len = key_layer.shape[-2] if past_key_value is not None: kv_seq_len += past_key_value[0].shape[-2] cos, sin = self.rotary_emb(value_layer, seq_len=kv_seq_len) query_layer, key_layer = apply_rotary_pos_emb(query_layer, key_layer, cos, sin, position_ids) if past_key_value is not None: key_layer = torch.cat([past_key_value[0], key_layer], dim=2) value_layer = torch.cat([past_key_value[1], value_layer], dim=2) attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2)) attention_scores = attention_scores / math.sqrt(self.attention_head_size) if attention_mask is not None: # Apply the attention mask is (precomputed for all layers in BertModel forward() function) attention_scores = attention_scores + attention_mask attention_probs = nn.functional.softmax(attention_scores, dim=-1) # This is actually dropping out entire tokens to attend to, which might # seem a bit unusual, but is taken from the original Transformer paper. attention_probs = self.dropout(attention_probs) context_layer = torch.matmul(attention_probs, value_layer) context_layer = context_layer.permute(0, 2, 1, 3).contiguous() new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,) context_layer = context_layer.view(*new_context_layer_shape) outputs = (context_layer, attention_probs) if output_attentions else (context_layer,) if self.is_decoder: outputs = outputs + (past_key_value,) return outputs # Copied from transformers.models.bert.modeling_bert.BertSelfOutput with Bert->E5Rope class E5RopeSelfOutput(nn.Module): def __init__(self, config): super().__init__() self.dense = nn.Linear(config.hidden_size, config.hidden_size) self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) self.dropout = nn.Dropout(config.hidden_dropout_prob) def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor: hidden_states = self.dense(hidden_states) hidden_states = self.dropout(hidden_states) hidden_states = self.LayerNorm(hidden_states + input_tensor) return hidden_states class E5RopeAttention(nn.Module): def __init__(self, config): super().__init__() self.self = E5RopeSelfAttention(config) self.output = E5RopeSelfOutput(config) self.pruned_heads = set() # Copied from transformers.models.bert.modeling_bert.BertAttention.prune_heads def prune_heads(self, heads): if len(heads) == 0: return heads, index = find_pruneable_heads_and_indices( heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads ) # Prune linear layers self.self.query = prune_linear_layer(self.self.query, index) self.self.key = prune_linear_layer(self.self.key, index) self.self.value = prune_linear_layer(self.self.value, index) self.output.dense = prune_linear_layer(self.output.dense, index, dim=1) # Update hyper params and store pruned heads self.self.num_attention_heads = self.self.num_attention_heads - len(heads) self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads self.pruned_heads = self.pruned_heads.union(heads) # End Copy def forward( self, hidden_states, attention_mask=None, position_ids=None, head_mask=None, encoder_hidden_states=None, encoder_attention_mask=None, past_key_value=None, output_attentions=False, ): self_outputs = self.self( hidden_states, attention_mask, position_ids, head_mask, encoder_hidden_states, encoder_attention_mask, past_key_value, output_attentions, ) attention_output = self.output(self_outputs[0], hidden_states) outputs = (attention_output,) + self_outputs[1:] # add attentions if we output them return outputs # Copied from transformers.models.bert.modeling_bert.BertIntermediate with Bert->E5Rope class E5RopeIntermediate(nn.Module): def __init__(self, config): super().__init__() self.dense = nn.Linear(config.hidden_size, config.intermediate_size) if isinstance(config.hidden_act, str): self.intermediate_act_fn = ACT2FN[config.hidden_act] else: self.intermediate_act_fn = config.hidden_act def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: hidden_states = self.dense(hidden_states) hidden_states = self.intermediate_act_fn(hidden_states) return hidden_states # Copied from transformers.models.bert.modeling_bert.BertOutput with Bert->E5Rope class E5RopeOutput(nn.Module): def __init__(self, config): super().__init__() self.dense = nn.Linear(config.intermediate_size, config.hidden_size) self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps) self.dropout = nn.Dropout(config.hidden_dropout_prob) def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor: hidden_states = self.dense(hidden_states) hidden_states = self.dropout(hidden_states) hidden_states = self.LayerNorm(hidden_states + input_tensor) return hidden_states class E5RopeLayer(nn.Module): def __init__(self, config): super().__init__() self.chunk_size_feed_forward = config.chunk_size_feed_forward self.seq_len_dim = 1 self.attention = E5RopeAttention(config) self.is_decoder = config.is_decoder self.add_cross_attention = config.add_cross_attention if self.add_cross_attention: if not self.is_decoder: raise ValueError(f"{self} should be used as a decoder model if cross attention is added") self.crossattention = E5RopeAttention(config) self.intermediate = E5RopeIntermediate(config) self.output = E5RopeOutput(config) def forward( self, hidden_states, attention_mask=None, position_ids=None, head_mask=None, encoder_hidden_states=None, encoder_attention_mask=None, past_key_value=None, output_attentions=False, ): # decoder uni-directional self-attention cached key/values tuple is at positions 1,2 self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None self_attention_outputs = self.attention( hidden_states, attention_mask, position_ids, head_mask, output_attentions=output_attentions, past_key_value=self_attn_past_key_value, ) attention_output = self_attention_outputs[0] # if decoder, the last output is tuple of self-attn cache if self.is_decoder: outputs = self_attention_outputs[1:-1] present_key_value = self_attention_outputs[-1] else: outputs = self_attention_outputs[1:] # add self attentions if we output attention weights cross_attn_present_key_value = None if self.is_decoder and encoder_hidden_states is not None: if not hasattr(self, "crossattention"): raise ValueError( f"If `encoder_hidden_states` are passed, {self} has to be instantiated with cross-attention " "layers by setting `config.add_cross_attention=True`" ) # cross_attn cached key/values tuple is at positions 3,4 of past_key_value tuple cross_attn_past_key_value = past_key_value[-2:] if past_key_value is not None else None cross_attention_outputs = self.crossattention( attention_output, attention_mask, position_ids, head_mask, encoder_hidden_states, encoder_attention_mask, cross_attn_past_key_value, output_attentions, ) attention_output = cross_attention_outputs[0] outputs = outputs + cross_attention_outputs[1:-1] # add cross attentions if we output attention weights # add cross-attn cache to positions 3,4 of present_key_value tuple cross_attn_present_key_value = cross_attention_outputs[-1] present_key_value = present_key_value + cross_attn_present_key_value layer_output = apply_chunking_to_forward( self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output ) outputs = (layer_output,) + outputs # if decoder, return the attn key/values as the last output if self.is_decoder: outputs = outputs + (present_key_value,) return outputs def feed_forward_chunk(self, attention_output): intermediate_output = self.intermediate(attention_output) layer_output = self.output(intermediate_output, attention_output) return layer_output class E5RopeEncoder(nn.Module): def __init__(self, config): super().__init__() self.config = config self.layer = nn.ModuleList([E5RopeLayer(config) for _ in range(config.num_hidden_layers)]) self.gradient_checkpointing = False def forward( self, hidden_states, attention_mask=None, position_ids=None, head_mask=None, encoder_hidden_states=None, encoder_attention_mask=None, past_key_values=None, use_cache=None, output_attentions=False, output_hidden_states=False, return_dict=True, ): 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 all_hidden_states = () if output_hidden_states else None all_self_attentions = () if output_attentions else None all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0 # [sequence_length, embed_size_per_head] -> [batch_size, num_heads, sequence_length, embed_size_per_head] # sinusoidal_pos = self.embed_positions(hidden_states.shape[:-1], past_key_values_length)[None, None, :, :] next_decoder_cache = () if use_cache else None for i, layer_module in enumerate(self.layer): if output_hidden_states: all_hidden_states = all_hidden_states + (hidden_states,) layer_head_mask = head_mask[i] if head_mask is not None else None past_key_value = past_key_values[i] if past_key_values is not None else None if self.gradient_checkpointing and self.training: def create_custom_forward(module): def custom_forward(*inputs): return module(*inputs, past_key_value, output_attentions) return custom_forward layer_outputs = torch.utils.checkpoint.checkpoint( create_custom_forward(layer_module), hidden_states, attention_mask, position_ids, layer_head_mask, encoder_hidden_states, encoder_attention_mask, ) else: layer_outputs = layer_module( hidden_states, attention_mask, position_ids, layer_head_mask, encoder_hidden_states, encoder_attention_mask, past_key_value, output_attentions, ) hidden_states = layer_outputs[0] if use_cache: next_decoder_cache += (layer_outputs[-1],) if output_attentions: all_self_attentions = all_self_attentions + (layer_outputs[1],) if self.config.add_cross_attention: all_cross_attentions = all_cross_attentions + (layer_outputs[2],) if output_hidden_states: all_hidden_states = all_hidden_states + (hidden_states,) if not return_dict: return tuple( v for v in [ hidden_states, next_decoder_cache, all_hidden_states, all_self_attentions, all_cross_attentions, ] if v is not None ) return BaseModelOutputWithPastAndCrossAttentions( last_hidden_state=hidden_states, past_key_values=next_decoder_cache, hidden_states=all_hidden_states, attentions=all_self_attentions, cross_attentions=all_cross_attentions, ) class E5RopePredictionHeadTransform(nn.Module): def __init__(self, config): super().__init__() self.dense = nn.Linear(config.hidden_size, config.embedding_size) if isinstance(config.hidden_act, str): self.transform_act_fn = ACT2FN[config.hidden_act] else: self.transform_act_fn = config.hidden_act self.LayerNorm = nn.LayerNorm(config.embedding_size, eps=config.layer_norm_eps) def forward(self, hidden_states): hidden_states = self.dense(hidden_states) hidden_states = self.transform_act_fn(hidden_states) hidden_states = self.LayerNorm(hidden_states) return hidden_states class E5RopeLMPredictionHead(nn.Module): def __init__(self, config): super().__init__() self.transform = E5RopePredictionHeadTransform(config) # The output weights are the same as the input embeddings, but there is # an output-only bias for each token. self.decoder = nn.Linear(config.embedding_size, config.vocab_size, bias=False) self.bias = nn.Parameter(torch.zeros(config.vocab_size)) # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings` self.decoder.bias = self.bias def forward(self, hidden_states): hidden_states = self.transform(hidden_states) hidden_states = self.decoder(hidden_states) return hidden_states # Copied from transformers.models.bert.modeling_bert.BertOnlyMLMHead with Bert->E5Rope class E5RopeOnlyMLMHead(nn.Module): def __init__(self, config): super().__init__() self.predictions = E5RopeLMPredictionHead(config) def forward(self, sequence_output: torch.Tensor) -> torch.Tensor: prediction_scores = self.predictions(sequence_output) return prediction_scores class E5RopePreTrainedModel(PreTrainedModel): """ An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. """ config_class = E5RopeConfig load_tf_weights = load_tf_weights_in_e5rope base_model_prefix = "e5rope" supports_gradient_checkpointing = True def _init_weights(self, module): """Initialize the weights""" if isinstance(module, nn.Linear): # Slightly different from the TF version which uses truncated_normal for initialization # cf https://github.com/pytorch/pytorch/pull/5617 module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) if module.bias is not None: module.bias.data.zero_() elif isinstance(module, E5RopeRotaryEmbedding): pass elif isinstance(module, nn.Embedding): module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) if module.padding_idx is not None: module.weight.data[module.padding_idx].zero_() elif isinstance(module, nn.LayerNorm): module.bias.data.zero_() module.weight.data.fill_(1.0) def _set_gradient_checkpointing(self, module, value=False): if isinstance(module, E5RopeEncoder): module.gradient_checkpointing = value E5ROPE_START_DOCSTRING = r""" This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`E5RopeConfig`]): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights. """ E5ROPE_INPUTS_DOCSTRING = r""" Args: input_ids (`torch.LongTensor` of shape `({0})`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) attention_mask (`torch.FloatTensor` of shape `({0})`, *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) token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*): Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *sentence A* token, - 1 corresponds to a *sentence B* token. [What are token type IDs?](../glossary#token-type-ids) head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*): Mask to nullify selected heads of the self-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 `({0}, 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. """ @add_start_docstrings( "The bare E5Rope Model transformer outputting raw hidden-states without any specific head on top.", E5ROPE_START_DOCSTRING, ) class E5RopeModel(E5RopePreTrainedModel): """ The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added between the self-attention layers, following the architecture described in [Attention is all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin. To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder` argument and `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass. """ def __init__(self, config): super().__init__(config) self.config = config self.embeddings = E5RopeEmbeddings(config) if config.embedding_size != config.hidden_size: self.embeddings_project = nn.Linear(config.embedding_size, config.hidden_size) self.encoder = E5RopeEncoder(config) # Initialize weights and apply final processing self.post_init() def get_input_embeddings(self): return self.embeddings.word_embeddings def set_input_embeddings(self, value): self.embeddings.word_embeddings = value def _prune_heads(self, heads_to_prune): """ Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base class PreTrainedModel """ for layer, heads in heads_to_prune.items(): self.encoder.layer[layer].attention.prune_heads(heads) @add_start_docstrings_to_model_forward(E5ROPE_INPUTS_DOCSTRING.format("batch_size, sequence_length")) def forward( self, input_ids: Optional[torch.LongTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, position_ids: Optional[torch.LongTensor] = None, token_type_ids: Optional[torch.LongTensor] = None, head_mask: Optional[torch.FloatTensor] = None, inputs_embeds: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, past_key_values: Optional[Tuple[Tuple[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[BaseModelOutputWithPastAndCrossAttentions, Tuple[torch.Tensor]]: r""" encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder. encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up 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)`. use_cache (`bool`, *optional*): If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see `past_key_values`). """ 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 if self.config.is_decoder: use_cache = use_cache if use_cache is not None else self.config.use_cache else: use_cache = False 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: self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask) input_shape = input_ids.size() elif inputs_embeds is not None: input_shape = inputs_embeds.size()[:-1] else: raise ValueError("You have to specify either input_ids or inputs_embeds") batch_size, seq_length = input_shape device = input_ids.device if input_ids is not None else inputs_embeds.device # past_key_values_length past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0 if attention_mask is None: attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device) if token_type_ids is None: token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device) if position_ids is None: position_ids = torch.arange( past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device ) position_ids = position_ids.unsqueeze(0).view(-1, seq_length) ### inserted code for positional skip-wise training ### ### https://arxiv.org/abs/2309.10400 ### if self.config.use_pose == True and self.training: pos_list = [] for i in range(batch_size): bias = random.randint(-seq_length, self.config.pose_target_len) bias = min(bias, self.config.pose_target_len - seq_length) bias = max(bias, 0) pos = torch.arange( past_key_values_length, past_key_values_length + seq_length, dtype=torch.long, device=device ) bias_st_ids = random.randint(min(64, seq_length-1), seq_length - 1) # do not skip very short sequences pos[bias_st_ids:] += bias pos_list.append(pos) position_ids = torch.stack(pos_list, dim=0) ####################################################### else: position_ids = position_ids.view(-1, seq_length).long() # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length] # ourselves in which case we just need to make it broadcastable to all heads. extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape) # If a 2D or 3D attention mask is provided for the cross-attention # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length] if self.config.is_decoder and encoder_hidden_states is not None: encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size() encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length) if encoder_attention_mask is None: encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device) encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask) else: encoder_extended_attention_mask = None # Prepare head mask if needed # 1.0 in head_mask indicate we keep the head # attention_probs has shape bsz x n_heads x N x N # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads] # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length] head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers) embedding_output = self.embeddings( input_ids=input_ids, token_type_ids=token_type_ids, inputs_embeds=inputs_embeds ) if hasattr(self, "embeddings_project"): embedding_output = self.embeddings_project(embedding_output) encoder_outputs = self.encoder( embedding_output, attention_mask=extended_attention_mask, position_ids=position_ids, head_mask=head_mask, encoder_hidden_states=encoder_hidden_states, encoder_attention_mask=encoder_extended_attention_mask, past_key_values=past_key_values, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) sequence_output = encoder_outputs[0] if not return_dict: return (sequence_output,) + encoder_outputs[1:] return BaseModelOutputWithPastAndCrossAttentions( last_hidden_state=sequence_output, past_key_values=encoder_outputs.past_key_values, hidden_states=encoder_outputs.hidden_states, attentions=encoder_outputs.attentions, cross_attentions=encoder_outputs.cross_attentions, ) @add_start_docstrings("""E5Rope Model with a `language modeling` head on top.""", E5ROPE_START_DOCSTRING) class E5RopeForMaskedLM(E5RopePreTrainedModel): _tied_weights_keys = ["cls.predictions.decoder.bias", "cls.predictions.decoder.weight"] def __init__(self, config): super().__init__(config) if config.is_decoder: logger.warning( "If you want to use `E5RopeForMaskedLM` make sure `config.is_decoder=False` for " "bi-directional self-attention." ) self.e5rope = E5RopeModel(config) self.cls = E5RopeOnlyMLMHead(config) # Initialize weights and apply final processing self.post_init() def get_output_embeddings(self): return self.cls.predictions.decoder def set_output_embeddings(self, new_embeddings): self.cls.predictions.decoder = new_embeddings @add_start_docstrings_to_model_forward(E5ROPE_INPUTS_DOCSTRING.format("batch_size, sequence_length")) def forward( self, input_ids: Optional[torch.LongTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, token_type_ids: Optional[torch.LongTensor] = None, head_mask: Optional[torch.FloatTensor] = None, inputs_embeds: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, labels: Optional[torch.LongTensor] = None, output_attentions: Optional[bool] = None, output_hidden_states: Optional[bool] = None, return_dict: Optional[bool] = None, ) -> Union[MaskedLMOutput, Tuple[torch.Tensor]]: r""" labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ..., config.vocab_size]` (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]`. """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict outputs = self.e5rope( input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, encoder_hidden_states=encoder_hidden_states, encoder_attention_mask=encoder_attention_mask, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) sequence_output = outputs[0] prediction_scores = self.cls(sequence_output) masked_lm_loss = None if labels is not None: loss_fct = CrossEntropyLoss() # -100 index = padding token masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1)) if not return_dict: output = (prediction_scores,) + outputs[1:] return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output return MaskedLMOutput( loss=masked_lm_loss, logits=prediction_scores, hidden_states=outputs.hidden_states, attentions=outputs.attentions, ) def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **model_kwargs): input_shape = input_ids.shape effective_batch_size = input_shape[0] # add a dummy token assert self.config.pad_token_id is not None, "The PAD token should be defined for generation" attention_mask = torch.cat([attention_mask, attention_mask.new_zeros((attention_mask.shape[0], 1))], dim=-1) dummy_token = torch.full( (effective_batch_size, 1), self.config.pad_token_id, dtype=torch.long, device=input_ids.device ) input_ids = torch.cat([input_ids, dummy_token], dim=1) return {"input_ids": input_ids, "attention_mask": attention_mask} @add_start_docstrings( """E5Rope Model with a `language modeling` head on top for CLM fine-tuning.""", E5ROPE_START_DOCSTRING ) class E5RopeForCausalLM(E5RopePreTrainedModel): _tied_weights_keys = ["cls.predictions.decoder.bias", "cls.predictions.decoder.weight"] def __init__(self, config): super().__init__(config) if not config.is_decoder: logger.warning("If you want to use `E5RopeForCausalLM` as a standalone, add `is_decoder=True.`") self.e5rope = E5RopeModel(config) self.cls = E5RopeOnlyMLMHead(config) # Initialize weights and apply final processing self.post_init() def get_output_embeddings(self): return self.cls.predictions.decoder def set_output_embeddings(self, new_embeddings): self.cls.predictions.decoder = new_embeddings @add_start_docstrings_to_model_forward(E5ROPE_INPUTS_DOCSTRING.format("batch_size, sequence_length")) def forward( self, input_ids: Optional[torch.LongTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, token_type_ids: Optional[torch.LongTensor] = None, inputs_embeds: Optional[torch.FloatTensor] = None, encoder_hidden_states: Optional[torch.FloatTensor] = None, encoder_attention_mask: Optional[torch.FloatTensor] = None, head_mask: Optional[torch.FloatTensor] = None, cross_attn_head_mask: Optional[torch.Tensor] = None, past_key_values: Optional[Tuple[Tuple[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[CausalLMOutputWithCrossAttentions, Tuple[torch.Tensor]]: r""" encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder. encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`): Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up 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)`. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels n `[0, ..., config.vocab_size]`. use_cache (`bool`, *optional*): If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see `past_key_values`). Returns: Example: ```python >>> from transformers import AutoTokenizer, E5RopeForCausalLM, E5RopeConfig >>> import torch >>> tokenizer = AutoTokenizer.from_pretrained("junnyu/e5rope_chinese_base") >>> config = E5RopeConfig.from_pretrained("junnyu/e5rope_chinese_base") >>> config.is_decoder = True >>> model = E5RopeForCausalLM.from_pretrained("junnyu/e5rope_chinese_base", config=config) >>> inputs = tokenizer("今天天气非常好。", return_tensors="pt") >>> outputs = model(**inputs) >>> prediction_logits = outputs.logits ```""" return_dict = return_dict if return_dict is not None else self.config.use_return_dict outputs = self.e5rope( input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, encoder_hidden_states=encoder_hidden_states, encoder_attention_mask=encoder_attention_mask, past_key_values=past_key_values, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) sequence_output = outputs[0] prediction_scores = self.cls(sequence_output) lm_loss = None if labels is not None: # we are doing next-token prediction; shift prediction scores and input ids by one shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous() labels = labels[:, 1:].contiguous() loss_fct = CrossEntropyLoss() lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1)) if not return_dict: output = (prediction_scores,) + outputs[1:] return ((lm_loss,) + output) if lm_loss is not None else output return CausalLMOutputWithCrossAttentions( loss=lm_loss, logits=prediction_scores, past_key_values=outputs.past_key_values, hidden_states=outputs.hidden_states, attentions=outputs.attentions, cross_attentions=outputs.cross_attentions, ) def prepare_inputs_for_generation(self, input_ids, past_key_values=None, attention_mask=None, **model_kwargs): input_shape = input_ids.shape # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly if attention_mask is None: attention_mask = input_ids.new_ones(input_shape) # cut decoder_input_ids if past is used if past_key_values is not None: input_ids = input_ids[:, -1:] return {"input_ids": input_ids, "attention_mask": attention_mask, "past_key_values": past_key_values} def _reorder_cache(self, past_key_values, beam_idx): reordered_past = () for layer_past in past_key_values: reordered_past += ( tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past[:2]) + layer_past[2:], ) return reordered_past class E5RopeClassificationHead(nn.Module): """Head for sentence-level classification tasks.""" def __init__(self, config): super().__init__() self.dense = nn.Linear(config.hidden_size, config.hidden_size) self.dropout = nn.Dropout(config.hidden_dropout_prob) self.out_proj = nn.Linear(config.hidden_size, config.num_labels) self.config = config def forward(self, features, **kwargs): x = features[:, 0, :] # take token (equiv. to [CLS]) x = self.dropout(x) x = self.dense(x) x = ACT2FN[self.config.hidden_act](x) x = self.dropout(x) x = self.out_proj(x) return x