# coding=utf-8 # Copyright 2021 The Eleuther AI and 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 GPT Neo model.""" import os from typing import Optional, Tuple, Union import torch import torch.nn.functional as F import torch.utils.checkpoint from torch import nn from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss from ...activations import ACT2FN from ...modeling_attn_mask_utils import _prepare_4d_causal_attention_mask from ...modeling_outputs import ( BaseModelOutputWithPast, BaseModelOutputWithPastAndCrossAttentions, CausalLMOutputWithCrossAttentions, CausalLMOutputWithPast, QuestionAnsweringModelOutput, SequenceClassifierOutputWithPast, TokenClassifierOutput, ) from ...modeling_utils import PreTrainedModel from ...pytorch_utils import is_torch_greater_or_equal_than_1_13 from ...utils import ( add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, is_flash_attn_2_available, is_flash_attn_greater_or_equal_2_10, is_torch_fx_available, logging, ) from .configuration_gpt_neo import GPTNeoConfig if is_flash_attn_2_available(): from flash_attn import flash_attn_func, flash_attn_varlen_func from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input # noqa # This makes `_prepare_4d_causal_attention_mask` a leaf function in the FX graph. # It means that the function will not be traced through and simply appear as a node in the graph. if is_torch_fx_available(): if not is_torch_greater_or_equal_than_1_13: import torch.fx _prepare_4d_causal_attention_mask = torch.fx.wrap(_prepare_4d_causal_attention_mask) logger = logging.get_logger(__name__) _CONFIG_FOR_DOC = "GPTNeoConfig" GPT_NEO_PRETRAINED_MODEL_ARCHIVE_LIST = [ "EleutherAI/gpt-neo-1.3B", # See all GPTNeo models at https://huggingface.co/models?filter=gpt_neo ] _CHECKPOINT_FOR_DOC = "EleutherAI/gpt-neo-1.3B" # Copied from transformers.models.llama.modeling_llama._get_unpad_data def _get_unpad_data(attention_mask): seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32) indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten() max_seqlen_in_batch = seqlens_in_batch.max().item() cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0)) return ( indices, cu_seqlens, max_seqlen_in_batch, ) def load_tf_weights_in_gpt_neo(model, config, gpt_neo_checkpoint_path): """Load tf checkpoints in a pytorch model""" try: import re 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(gpt_neo_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: if "global_step" not in name and "adam" not in name: array = tf.train.load_variable(tf_path, name) array = tf.dtypes.cast(array.squeeze(), tf.float32).numpy() name = name.replace("attn/q", "attn/attention/q_proj/w") name = name.replace("attn/k", "attn/attention/k_proj/w") name = name.replace("attn/v", "attn/attention/v_proj/w") name = name.replace("attn/o", "attn/attention/out_proj/w") name = name.replace("norm_1", "ln_1") name = name.replace("norm_2", "ln_2") name = name.replace("attn/compute_output_bias/o_b", "attn/attention/out_proj/b") name = name.replace("conv1d_main/c_fc/kernel", "c_fc/w") name = name.replace("conv1d_main/c_fc/bias", "c_fc/b") name = name.replace("conv1d_main/c_proj/kernel", "c_proj/w") name = name.replace("conv1d_main/c_proj/bias", "c_proj/b") names.append(name) arrays.append(array) for name, array in zip(names, arrays): name = name[5:] # skip "gpt2/" name = name.split("/") pointer = model.transformer 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] == "w" or scope_names[0] == "g": pointer = getattr(pointer, "weight") elif scope_names[0] == "b": pointer = getattr(pointer, "bias") elif scope_names[0] == "wpe" or scope_names[0] == "wte": pointer = getattr(pointer, scope_names[0]) pointer = getattr(pointer, "weight") else: pointer = getattr(pointer, scope_names[0]) if len(scope_names) >= 2: num = int(scope_names[1]) pointer = pointer[num] if name[-1] == "w" and name[-2] in ["out_proj", "k_proj", "q_proj", "v_proj", "c_proj", "c_fc"]: array = array.transpose() if name == ["wte"]: # if vocab is padded, then trim off the padding embeddings array = array[: config.vocab_size] if pointer.shape != array.shape: raise ValueError(f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched {name}") print(f"Initialize PyTorch weight {name}") pointer.data = torch.from_numpy(array) # init the final linear layer using word embeddings embs = model.transformer.wte.weight lin = nn.Linear(embs.size()[1], embs.size()[0], bias=False) lin.weight = embs model.set_output_embeddings(lin) return model class GPTNeoSelfAttention(nn.Module): def __init__(self, config, attention_type): super().__init__() self.config = config max_positions = config.max_position_embeddings bias = torch.tril(torch.ones((max_positions, max_positions), dtype=bool)).view( 1, 1, max_positions, max_positions ) # local causal self attention is a sliding window where each token can only attend to the previous # window_size tokens. This is implemented by updating the causal mask such that for each token # all other tokens are masked except the previous window_size tokens. if attention_type == "local": bias = torch.bitwise_xor(bias, torch.tril(bias, -config.window_size)) self.register_buffer("bias", bias, persistent=False) self.register_buffer("masked_bias", torch.tensor(-1e9), persistent=False) self.attn_dropout = nn.Dropout(float(config.attention_dropout)) self.resid_dropout = nn.Dropout(float(config.resid_dropout)) self.is_causal = True self.embed_dim = config.hidden_size self.num_heads = config.num_heads self.head_dim = self.embed_dim // self.num_heads if self.head_dim * self.num_heads != self.embed_dim: raise ValueError( f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:" f" {self.num_heads})." ) self.k_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False) self.v_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False) self.q_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False) self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=True) def _split_heads(self, tensor, num_heads, attn_head_size): """ Splits hidden_size dim into attn_head_size and num_heads """ new_shape = tensor.size()[:-1] + (num_heads, attn_head_size) tensor = tensor.view(new_shape) return tensor.permute(0, 2, 1, 3) # (batch, head, seq_length, head_features) def _merge_heads(self, tensor, num_heads, attn_head_size): """ Merges attn_head_size dim and num_attn_heads dim into hidden_size """ tensor = tensor.permute(0, 2, 1, 3).contiguous() new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,) return tensor.view(new_shape) def _attn(self, query, key, value, attention_mask=None, head_mask=None): # Keep the attention weights computation in fp32 to avoid overflow issues query = query.to(torch.float32) key = key.to(torch.float32) attn_weights = torch.matmul(query, key.transpose(-1, -2)) query_length, key_length = query.size(-2), key.size(-2) causal_mask = self.bias[:, :, key_length - query_length : key_length, :key_length] mask_value = torch.finfo(attn_weights.dtype).min # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`. # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device` mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(attn_weights.device) attn_weights = torch.where(causal_mask, attn_weights, mask_value) if attention_mask is not None: # Apply the attention mask attn_weights = attn_weights + attention_mask attn_weights = nn.functional.softmax(attn_weights, dim=-1) attn_weights = attn_weights.to(value.dtype) attn_weights = self.attn_dropout(attn_weights) # Mask heads if we want to if head_mask is not None: attn_weights = attn_weights * head_mask attn_output = torch.matmul(attn_weights, value) return attn_output, attn_weights def forward( self, hidden_states, attention_mask=None, layer_past=None, head_mask=None, use_cache=False, output_attentions=False, ): query = self.q_proj(hidden_states) key = self.k_proj(hidden_states) value = self.v_proj(hidden_states) query = self._split_heads(query, self.num_heads, self.head_dim) key = self._split_heads(key, self.num_heads, self.head_dim) value = self._split_heads(value, self.num_heads, self.head_dim) if layer_past is not None: past_key = layer_past[0] past_value = layer_past[1] key = torch.cat((past_key, key), dim=-2) value = torch.cat((past_value, value), dim=-2) if use_cache is True: present = (key, value) else: present = None attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask) attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim) attn_output = self.out_proj(attn_output) attn_output = self.resid_dropout(attn_output) outputs = (attn_output, present) if output_attentions: outputs += (attn_weights,) return outputs # a, present, (attentions) class GPTNeoFlashAttention2(GPTNeoSelfAttention): """ GPTNeo flash attention module. This module inherits from `GPTNeoSelfAttention` as the weights of the module stays untouched. The only required change would be on the forward pass where it needs to correctly call the public API of flash attention and deal with padding tokens in case the input contains any of them. """ # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__ def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs) # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1. # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0. # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left). self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10() def forward( self, hidden_states, attention_mask=None, layer_past=None, head_mask=None, use_cache=False, output_attentions=False, ): bsz, _, _ = hidden_states.size() query = self.q_proj(hidden_states) key = self.k_proj(hidden_states) value = self.v_proj(hidden_states) query = self._split_heads(query, self.num_heads, self.head_dim) key = self._split_heads(key, self.num_heads, self.head_dim) value = self._split_heads(value, self.num_heads, self.head_dim) if layer_past is not None: past_key = layer_past[0] past_value = layer_past[1] key = torch.cat((past_key, key), dim=-2) value = torch.cat((past_value, value), dim=-2) if use_cache is True: present = (key, value) else: present = None query_length = query.shape[2] tgt_len = key.shape[2] # Flash attention requires the input to have the shape # batch_size x seq_length x head_dim x hidden_dim query = query.transpose(1, 2).view(bsz, query_length, self.num_heads, self.head_dim) key = key.transpose(1, 2).view(bsz, tgt_len, self.num_heads, self.head_dim) value = value.transpose(1, 2).view(bsz, tgt_len, self.num_heads, self.head_dim) attn_dropout = self.config.attention_dropout if self.training else 0.0 # In PEFT, usually we cast the layer norms in float32 for training stability reasons # therefore the input hidden states gets silently casted in float32. Hence, we need # cast them back in the correct dtype just to be sure everything works as expected. # This might slowdown training & inference so it is recommended to not cast the LayerNorms # in fp32. (LlamaRMSNorm handles it correctly) if query.dtype == torch.float32: # Handle the case where the model is quantized if hasattr(self.config, "_pre_quantization_dtype"): target_dtype = self.config._pre_quantization_dtype else: target_dtype = self.q_proj.weight.dtype logger.warning_once( f"The input hidden states seems to be silently casted in float32, this might be related to" f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in" f" {target_dtype}." ) query = query.to(target_dtype) key = key.to(target_dtype) value = value.to(target_dtype) attn_output = self._flash_attention_forward( query, key, value, attention_mask, query_length, dropout=attn_dropout, softmax_scale=1.0 ) attn_weights_reshaped = attn_output.reshape(bsz, query_length, self.num_heads * self.head_dim) attn_output = self.out_proj(attn_weights_reshaped) attn_output = self.resid_dropout(attn_output) outputs = (attn_output, present) if output_attentions: outputs += (attn_weights_reshaped,) return outputs # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._flash_attention_forward def _flash_attention_forward( self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None ): """ Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token first unpad the input, then computes the attention scores and pad the final attention scores. Args: query_states (`torch.Tensor`): Input query states to be passed to Flash Attention API key_states (`torch.Tensor`): Input key states to be passed to Flash Attention API value_states (`torch.Tensor`): Input value states to be passed to Flash Attention API attention_mask (`torch.Tensor`): The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the position of padding tokens and 1 for the position of non-padding tokens. dropout (`int`, *optional*): Attention dropout softmax_scale (`float`, *optional*): The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim) """ if not self._flash_attn_uses_top_left_mask: causal = self.is_causal else: # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__. causal = self.is_causal and query_length != 1 # Contains at least one padding token in the sequence if attention_mask is not None: batch_size = query_states.shape[0] query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input( query_states, key_states, value_states, attention_mask, query_length ) cu_seqlens_q, cu_seqlens_k = cu_seq_lens max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens attn_output_unpad = flash_attn_varlen_func( query_states, key_states, value_states, cu_seqlens_q=cu_seqlens_q, cu_seqlens_k=cu_seqlens_k, max_seqlen_q=max_seqlen_in_batch_q, max_seqlen_k=max_seqlen_in_batch_k, dropout_p=dropout, softmax_scale=softmax_scale, causal=causal, ) attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length) else: attn_output = flash_attn_func( query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal ) return attn_output # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2._upad_input def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length): indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask) batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape key_layer = index_first_axis( key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k ) value_layer = index_first_axis( value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k ) if query_length == kv_seq_len: query_layer = index_first_axis( query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k ) cu_seqlens_q = cu_seqlens_k max_seqlen_in_batch_q = max_seqlen_in_batch_k indices_q = indices_k elif query_length == 1: max_seqlen_in_batch_q = 1 cu_seqlens_q = torch.arange( batch_size + 1, dtype=torch.int32, device=query_layer.device ) # There is a memcpy here, that is very bad. indices_q = cu_seqlens_q[:-1] query_layer = query_layer.squeeze(1) else: # The -q_len: slice assumes left padding. attention_mask = attention_mask[:, -query_length:] query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask) return ( query_layer, key_layer, value_layer, indices_q, (cu_seqlens_q, cu_seqlens_k), (max_seqlen_in_batch_q, max_seqlen_in_batch_k), ) GPT_NEO_ATTENTION_CLASSES = { "eager": GPTNeoSelfAttention, "flash_attention_2": GPTNeoFlashAttention2, } class GPTNeoAttention(nn.Module): def __init__(self, config, layer_id=0): super().__init__() self.layer_id = layer_id self.attention_layers = config.attention_layers self.attention_type = self.attention_layers[layer_id] if self.attention_type in ["global", "local"]: self.attention = GPT_NEO_ATTENTION_CLASSES[config._attn_implementation](config, self.attention_type) else: raise NotImplementedError( "Only attn layer types 'global' and 'local' exist, but got `config.attention_layers`: " f"{config.attention_layers}. Select attn layer types from ['global', 'local'] only." ) def forward( self, hidden_states, layer_past=None, attention_mask=None, head_mask=None, use_cache=False, output_attentions=False, ): return self.attention( hidden_states, attention_mask=attention_mask, layer_past=layer_past, head_mask=head_mask, use_cache=use_cache, output_attentions=output_attentions, ) class GPTNeoMLP(nn.Module): def __init__(self, intermediate_size, config): # in MLP: intermediate_size= 4 * hidden_size super().__init__() embed_dim = config.hidden_size self.c_fc = nn.Linear(embed_dim, intermediate_size) self.c_proj = nn.Linear(intermediate_size, embed_dim) self.act = ACT2FN[config.activation_function] self.dropout = nn.Dropout(float(config.resid_dropout)) def forward(self, hidden_states): hidden_states = self.c_fc(hidden_states) hidden_states = self.act(hidden_states) hidden_states = self.c_proj(hidden_states) hidden_states = self.dropout(hidden_states) return hidden_states class GPTNeoBlock(nn.Module): def __init__(self, config, layer_id): super().__init__() hidden_size = config.hidden_size inner_dim = config.intermediate_size if config.intermediate_size is not None else 4 * hidden_size self.ln_1 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon) self.attn = GPTNeoAttention(config, layer_id) self.ln_2 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon) self.mlp = GPTNeoMLP(inner_dim, config) def forward( self, hidden_states, layer_past=None, attention_mask=None, head_mask=None, use_cache=False, output_attentions=False, ): residual = hidden_states hidden_states = self.ln_1(hidden_states) attn_outputs = self.attn( hidden_states, layer_past=layer_past, attention_mask=attention_mask, head_mask=head_mask, use_cache=use_cache, output_attentions=output_attentions, ) attn_output = attn_outputs[0] # output_attn: a, present, (attentions) outputs = attn_outputs[1:] # residual connection hidden_states = attn_output + residual residual = hidden_states hidden_states = self.ln_2(hidden_states) feed_forward_hidden_states = self.mlp(hidden_states) # residual connection hidden_states = residual + feed_forward_hidden_states if use_cache: outputs = (hidden_states,) + outputs else: outputs = (hidden_states,) + outputs[1:] return outputs # hidden_states, present, (attentions, cross_attentions) class GPTNeoPreTrainedModel(PreTrainedModel): """ An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. """ config_class = GPTNeoConfig load_tf_weights = load_tf_weights_in_gpt_neo base_model_prefix = "transformer" supports_gradient_checkpointing = True _no_split_modules = ["GPTNeoBlock"] _skip_keys_device_placement = "past_key_values" _supports_flash_attn_2 = True def __init__(self, *inputs, **kwargs): super().__init__(*inputs, **kwargs) 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, 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) GPT_NEO_START_DOCSTRING = r""" This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`GPTNeoConfig`]): 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. """ GPT_NEO_INPUTS_DOCSTRING = r""" Args: input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`): `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary. If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as `input_ids`. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) past_key_values (`Tuple[Tuple[torch.Tensor]]` of length `config.num_layers`): Contains precomputed hidden-states (key and values in the attention blocks) as computed by the model (see `past_key_values` output below). Can be used to speed up sequential decoding. The `input_ids` which have their past given to this model should not be passed as `input_ids` as they have already been computed. attention_mask (`torch.FloatTensor` 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) token_type_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`, *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) position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. [What are position IDs?](../glossary#position-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 `(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. If `past_key_values` is used, optionally only the last `inputs_embeds` have to be input (see `past_key_values`). 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 (`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 GPT Neo Model transformer outputting raw hidden-states without any specific head on top.", GPT_NEO_START_DOCSTRING, ) class GPTNeoModel(GPTNeoPreTrainedModel): def __init__(self, config): super().__init__(config) self.embed_dim = config.hidden_size self.wte = nn.Embedding(config.vocab_size, self.embed_dim) self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim) self.drop = nn.Dropout(float(config.embed_dropout)) self.h = nn.ModuleList([GPTNeoBlock(config, layer_id=i) for i in range(config.num_layers)]) self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2" self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon) self.gradient_checkpointing = False # Initialize weights and apply final processing self.post_init() def get_input_embeddings(self): return self.wte def set_input_embeddings(self, new_embeddings): self.wte = new_embeddings @add_start_docstrings_to_model_forward(GPT_NEO_INPUTS_DOCSTRING) @add_code_sample_docstrings( checkpoint=_CHECKPOINT_FOR_DOC, output_type=BaseModelOutputWithPastAndCrossAttentions, config_class=_CONFIG_FOR_DOC, ) def forward( self, input_ids: Optional[torch.Tensor] = None, past_key_values: Optional[Tuple[torch.FloatTensor]] = None, attention_mask: Optional[torch.Tensor] = None, token_type_ids: Optional[torch.Tensor] = None, position_ids: Optional[torch.Tensor] = None, head_mask: Optional[torch.Tensor] = None, inputs_embeds: Optional[torch.Tensor] = None, use_cache: Optional[bool] = None, output_attentions: Optional[bool] = None, output_hidden_states: Optional[bool] = None, return_dict: Optional[bool] = None, ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]: 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 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() input_ids = input_ids.view(-1, input_shape[-1]) 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") device = input_ids.device if input_ids is not None else inputs_embeds.device if token_type_ids is not None: token_type_ids = token_type_ids.view(-1, input_shape[-1]) if past_key_values is None: past_length = 0 past_key_values = tuple([None] * len(self.h)) else: past_length = past_key_values[0][0].size(-2) if position_ids is None: position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device) position_ids = position_ids.unsqueeze(0) # Prepare head mask if needed # 1.0 in head_mask indicate we keep the head # attention_probs has shape bsz x num_heads x N x N # head_mask has shape n_layer x batch x num_heads x N x N head_mask = self.get_head_mask(head_mask, self.config.num_layers) if inputs_embeds is None: inputs_embeds = self.wte(input_ids) position_embeds = self.wpe(position_ids) hidden_states = inputs_embeds + position_embeds # Attention mask. if self._use_flash_attention_2: # 2d mask is passed through the layers attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None else: # 4d mask is passed through the layers attention_mask = _prepare_4d_causal_attention_mask(attention_mask, input_shape, inputs_embeds, past_length) if token_type_ids is not None: token_type_embeds = self.wte(token_type_ids) hidden_states = hidden_states + token_type_embeds hidden_states = self.drop(hidden_states) output_shape = (-1,) + input_shape[1:] + (hidden_states.size(-1),) 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 presents = () if use_cache else None all_self_attentions = () if output_attentions else None all_hidden_states = () if output_hidden_states else None for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)): if output_hidden_states: all_hidden_states = all_hidden_states + (hidden_states,) if self.gradient_checkpointing and self.training: outputs = self._gradient_checkpointing_func( block.__call__, hidden_states, None, attention_mask, head_mask[i], use_cache, output_attentions, ) else: outputs = block( hidden_states, layer_past=layer_past, attention_mask=attention_mask, head_mask=head_mask[i], use_cache=use_cache, output_attentions=output_attentions, ) hidden_states = outputs[0] if use_cache is True: presents = presents + (outputs[1],) if output_attentions: all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],) hidden_states = self.ln_f(hidden_states) hidden_states = hidden_states.view(output_shape) # Add last hidden state if output_hidden_states: all_hidden_states = all_hidden_states + (hidden_states,) if not return_dict: return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None) return BaseModelOutputWithPast( last_hidden_state=hidden_states, past_key_values=presents, hidden_states=all_hidden_states, attentions=all_self_attentions, ) @add_start_docstrings( """ The GPT Neo Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings). """, GPT_NEO_START_DOCSTRING, ) class GPTNeoForCausalLM(GPTNeoPreTrainedModel): _tied_weights_keys = ["lm_head.weight"] def __init__(self, config): super().__init__(config) self.transformer = GPTNeoModel(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_output_embeddings(self): return self.lm_head def set_output_embeddings(self, new_embeddings): self.lm_head = new_embeddings def prepare_inputs_for_generation(self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs): token_type_ids = kwargs.get("token_type_ids", None) # Omit tokens covered by past_key_values if past_key_values: past_length = past_key_values[0][0].shape[2] # Some generation methods already pass only the last input ID if input_ids.shape[1] > past_length: remove_prefix_length = past_length else: # Default to old behavior: keep only final ID remove_prefix_length = input_ids.shape[1] - 1 input_ids = input_ids[:, remove_prefix_length:] if token_type_ids is not None: token_type_ids = token_type_ids[:, -input_ids.shape[1] :] attention_mask = kwargs.get("attention_mask", None) position_ids = kwargs.get("position_ids", None) if attention_mask is not None and position_ids is None: # create position_ids on the fly for batch generation position_ids = attention_mask.long().cumsum(-1) - 1 position_ids.masked_fill_(attention_mask == 0, 1) if past_key_values: position_ids = position_ids[:, -input_ids.shape[1] :] # if `inputs_embeds` are passed, we only want to use them in the 1st generation step if inputs_embeds is not None and past_key_values is None: model_inputs = {"inputs_embeds": inputs_embeds} else: model_inputs = {"input_ids": input_ids} model_inputs.update( { "past_key_values": past_key_values, "use_cache": kwargs.get("use_cache"), "position_ids": position_ids, "attention_mask": attention_mask, "token_type_ids": token_type_ids, } ) return model_inputs @add_start_docstrings_to_model_forward(GPT_NEO_INPUTS_DOCSTRING) @add_code_sample_docstrings( checkpoint=_CHECKPOINT_FOR_DOC, output_type=CausalLMOutputWithCrossAttentions, config_class=_CONFIG_FOR_DOC, ) def forward( self, input_ids: Optional[torch.Tensor] = None, past_key_values: Optional[Tuple[torch.FloatTensor]] = None, attention_mask: Optional[torch.Tensor] = None, token_type_ids: Optional[torch.Tensor] = None, position_ids: Optional[torch.Tensor] = None, head_mask: Optional[torch.Tensor] = None, inputs_embeds: Optional[torch.Tensor] = None, labels: Optional[torch.Tensor] = None, use_cache: Optional[bool] = None, output_attentions: Optional[bool] = None, output_hidden_states: Optional[bool] = None, return_dict: Optional[bool] = None, ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]: r""" labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100` are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]` """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict transformer_outputs = self.transformer( input_ids, past_key_values=past_key_values, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) hidden_states = transformer_outputs[0] lm_logits = self.lm_head(hidden_states) loss = None if labels is not None: # move labels to correct device to enable model parallelism labels = labels.to(lm_logits.device) # Compute loss in fp32 to match with mesh-tf version # https://github.com/EleutherAI/gpt-neo/blob/89ce74164da2fb16179106f54e2269b5da8db333/models/gpt2/gpt2.py#L179 lm_logits = lm_logits.to(torch.float32) # Shift so that tokens < n predict n shift_logits = lm_logits[..., :-1, :].contiguous() shift_labels = labels[..., 1:].contiguous() # Flatten the tokens loss_fct = CrossEntropyLoss() loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1)) lm_logits = lm_logits.to(hidden_states.dtype) loss = loss.to(hidden_states.dtype) if not return_dict: output = (lm_logits,) + transformer_outputs[1:] return ((loss,) + output) if loss is not None else output return CausalLMOutputWithPast( loss=loss, logits=lm_logits, past_key_values=transformer_outputs.past_key_values, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, ) @staticmethod def _reorder_cache( past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor ) -> Tuple[Tuple[torch.Tensor]]: """ This function is used to re-order the `past_key_values` cache if [`~PretrainedModel.beam_search`] or [`~PretrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct beam_idx at every generation step. """ return tuple( tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past) for layer_past in past_key_values ) @add_start_docstrings( """ The GPTNeo Model transformer with a sequence classification head on top (linear layer). [`GPTNeoForSequenceClassification`] uses the last token in order to do the classification, as other causal models (e.g. GPT-1) do. Since it does classification on the last token, it requires to know the position of the last token. If a `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in each row of the batch). """, GPT_NEO_START_DOCSTRING, ) class GPTNeoForSequenceClassification(GPTNeoPreTrainedModel): def __init__(self, config): super().__init__(config) self.num_labels = config.num_labels self.transformer = GPTNeoModel(config) self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False) # Initialize weights and apply final processing self.post_init() @add_start_docstrings_to_model_forward(GPT_NEO_INPUTS_DOCSTRING) @add_code_sample_docstrings( checkpoint=_CHECKPOINT_FOR_DOC, output_type=SequenceClassifierOutputWithPast, config_class=_CONFIG_FOR_DOC, ) def forward( self, input_ids: Optional[torch.Tensor] = None, past_key_values: Optional[Tuple[torch.FloatTensor]] = None, attention_mask: Optional[torch.Tensor] = None, token_type_ids: Optional[torch.Tensor] = None, position_ids: Optional[torch.Tensor] = None, head_mask: Optional[torch.Tensor] = None, inputs_embeds: Optional[torch.Tensor] = None, labels: Optional[torch.Tensor] = None, use_cache: Optional[bool] = None, output_attentions: Optional[bool] = None, output_hidden_states: Optional[bool] = None, return_dict: Optional[bool] = None, ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutputWithPast]: r""" 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 regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict transformer_outputs = self.transformer( input_ids, past_key_values=past_key_values, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) hidden_states = transformer_outputs[0] logits = self.score(hidden_states) if input_ids is not None: batch_size, sequence_length = input_ids.shape[:2] else: batch_size, sequence_length = inputs_embeds.shape[:2] if self.config.pad_token_id is None and batch_size != 1: raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.") if self.config.pad_token_id is None: sequence_lengths = -1 else: if input_ids is not None: sequence_lengths = (torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1).to( logits.device ) else: sequence_lengths = -1 logger.warning( f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be " "unexpected if using padding tokens in conjunction with `inputs_embeds.`" ) pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths] loss = None if labels is not None: if self.config.problem_type is None: if self.num_labels == 1: self.config.problem_type = "regression" elif self.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.num_labels == 1: loss = loss_fct(pooled_logits.squeeze(), labels.squeeze()) else: loss = loss_fct(pooled_logits, labels) elif self.config.problem_type == "single_label_classification": loss_fct = CrossEntropyLoss() loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1)) elif self.config.problem_type == "multi_label_classification": loss_fct = BCEWithLogitsLoss() loss = loss_fct(pooled_logits, labels) if not return_dict: output = (pooled_logits,) + transformer_outputs[1:] return ((loss,) + output) if loss is not None else output return SequenceClassifierOutputWithPast( loss=loss, logits=pooled_logits, past_key_values=transformer_outputs.past_key_values, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, ) @add_start_docstrings( """ GPT Neo model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. """, GPT_NEO_START_DOCSTRING, ) class GPTNeoForTokenClassification(GPTNeoPreTrainedModel): def __init__(self, config): super().__init__(config) self.num_labels = config.num_labels self.transformer = GPTNeoModel(config) self.dropout = nn.Dropout(config.classifier_dropout) self.classifier = nn.Linear(config.hidden_size, config.num_labels) # Initialize weights and apply final processing self.post_init() @add_start_docstrings_to_model_forward(GPT_NEO_INPUTS_DOCSTRING) @add_code_sample_docstrings( checkpoint="EleutherAI/gpt-neo-125m", output_type=TokenClassifierOutput, config_class=_CONFIG_FOR_DOC, expected_loss=0.25, ) def forward( self, input_ids: Optional[torch.LongTensor] = None, past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None, attention_mask: Optional[torch.FloatTensor] = None, token_type_ids: Optional[torch.LongTensor] = None, position_ids: Optional[torch.LongTensor] = None, head_mask: Optional[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, ) -> Union[Tuple, TokenClassifierOutput]: r""" labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict transformer_outputs = self.transformer( input_ids, past_key_values=past_key_values, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, use_cache=use_cache, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, ) hidden_states = transformer_outputs[0] hidden_states = self.dropout(hidden_states) logits = self.classifier(hidden_states) loss = None if labels is not None: labels = labels.to(logits.device) loss_fct = CrossEntropyLoss() loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1)) if not return_dict: output = (logits,) + transformer_outputs[2:] return ((loss,) + output) if loss is not None else output return TokenClassifierOutput( loss=loss, logits=logits, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, ) @add_start_docstrings( """ The GPT-Neo Model transformer with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layer on top of the hidden-states output to compute `span start logits` and `span end logits`). """, GPT_NEO_START_DOCSTRING, ) class GPTNeoForQuestionAnswering(GPTNeoPreTrainedModel): def __init__(self, config): super().__init__(config) self.num_labels = config.num_labels self.transformer = GPTNeoModel(config) self.qa_outputs = nn.Linear(config.hidden_size, 2) # Initialize weights and apply final processing self.post_init() @add_start_docstrings_to_model_forward(GPT_NEO_INPUTS_DOCSTRING.format("batch_size, sequence_length")) @add_code_sample_docstrings( checkpoint=_CHECKPOINT_FOR_DOC, output_type=QuestionAnsweringModelOutput, config_class=_CONFIG_FOR_DOC, real_checkpoint=_CHECKPOINT_FOR_DOC, ) def forward( self, input_ids: Optional[torch.LongTensor] = None, attention_mask: Optional[torch.FloatTensor] = None, token_type_ids: Optional[torch.LongTensor] = None, position_ids: Optional[torch.LongTensor] = None, head_mask: Optional[torch.FloatTensor] = None, inputs_embeds: Optional[torch.FloatTensor] = None, start_positions: Optional[torch.LongTensor] = None, end_positions: Optional[torch.LongTensor] = None, output_attentions: Optional[bool] = None, output_hidden_states: Optional[bool] = None, return_dict: Optional[bool] = None, ) -> Union[Tuple, QuestionAnsweringModelOutput]: r""" start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. """ return_dict = return_dict if return_dict is not None else self.config.use_return_dict outputs = self.transformer( input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, 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[2:] return ((total_loss,) + output) if total_loss is not None else output return QuestionAnsweringModelOutput( loss=total_loss, start_logits=start_logits, end_logits=end_logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions, )