codegen.py

# coding=utf-8
# Copyright 2022 Salesforce authors, The EleutherAI, and HuggingFace Teamindspore. 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.
# pylint: disable=C0103

""" MindSpore CodeGen model."""
import os
from typing import Optional, Tuple, Union

import mindspore
import numpy as np
from mindspore import nn, Tensor, Parameter
from mindspore import ops

from mindnlp.models.utils import logging
from mindnlp.models.utils.activations import ACT2FN
from mindnlp.models.codegen.codegen_config import CodeGenConfig
from mindnlp.abc.backbones.pretrained import PreTrainedModel

logger = logging.get_logger(__name__)

_CHECKPOINT_FOR_DOC = "Salesforce/codegen-2B-mono"
_CONFIG_FOR_DOC = "CodeGenConfig"


#
def fixed_pos_embedding(tensor, seq_dim=1, seq_len=None):
    """
    fixed_pos_embedding
    """
    axis = tensor.shape[-1]
    if seq_len is None:
        seq_len = tensor.shape[seq_dim]
    inv_freq = 1.0 / (10000 ** (np.arange(0, axis, 2) / axis))
    sinusoid_inp = (
        np.einsum("i,j->ij", np.arange(seq_len, dtype=np.float32), inv_freq).astype(np.float32)
    )
    return Tensor(np.sin(sinusoid_inp)), Tensor(np.cos(sinusoid_inp))


def rotate_every_two(tensor):
    """
    rotate_every_two
    """
    tensor1 = tensor[:, :, :, ::2]
    tensor2 = tensor[:, :, :, 1::2]
    tensor = ops.stack((-tensor2, tensor1), axis=-1)
    return tensor.flatten(order='C', start_dim=-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')


def duplicate_interleave(tensor):
    """
    A simple version of `torch.repeat_interleave` for duplicating a matrix while interleaving the copy.
    """
    dim0 = tensor.shape[0]
    tensor = tensor.view(-1, 1)  # flatten the matrix
    tensor = ops.tile(tensor, (1, 2))  # repeat all elements into the 2nd dimension
    tensor = tensor.view(dim0, -1)  # reshape into a matrix, interleaving the copy
    return tensor


def apply_rotary_pos_emb(tensor, sincos, offset=0):
    """
    apply_rotary_pos_emb
    """
    sin, cos = (duplicate_interleave(t)[None, offset: tensor.shape[1] + offset, None, :] for t in sincos)
    # einsum notation for lambda t: repeat(t[offset:x.shape[1]+offset,:], "n d -> () n () (d j)", j=2)
    return (tensor * cos) + (rotate_every_two(tensor) * sin)


class CodeGenAttention(nn.Cell):
    """
    CodeGenAttention
    """

    def __init__(self, config):
        super().__init__()

        max_positions = config.n_positions
        self.causal_mask = Parameter(ops.tril(ops.ones((max_positions, max_positions), dtype=mindspore.uint8)).view(
            1, 1, max_positions, max_positions
        ), requires_grad=False)

        self.attn_dropout = nn.Dropout(p=config.attn_pdrop)
        self.resid_dropout = nn.Dropout(p=config.resid_pdrop)

        self.embed_dim = config.n_embd
        self.num_attention_heads = config.n_head
        self.head_dim = self.embed_dim // self.num_attention_heads
        if self.head_dim * self.num_attention_heads != self.embed_dim:
            raise ValueError(
                f"embed_dim must be divisible by num_attention_heads (got `embed_dim`: {self.embed_dim} and"
                f" `num_attention_heads`: {self.num_attention_heads})."
            )
        self.scale_attn = ops.sqrt(mindspore.Tensor(self.head_dim, dtype=mindspore.float32)).astype(mindspore.float32)
        self.qkv_proj = nn.Dense(self.embed_dim, self.embed_dim * 3, has_bias=False)

        self.out_proj = nn.Dense(self.embed_dim, self.embed_dim, has_bias=False)
        self.rotary_dim = None
        if config.rotary_dim is not None:
            self.rotary_dim = config.rotary_dim

    def _split_heads(self, tensor, n_head, dim_head, mp_num):
        reshaped = tensor.reshape(tensor.shape[:-1] + (n_head // mp_num, dim_head))
        reshaped = reshaped.reshape(tensor.shape[:-2] + (-1,) + reshaped.shape[-1:])
        return reshaped

    def _merge_heads(self, tensor, num_attention_heads, attn_head_size):
        """
        Merges attn_head_size dim and num_attn_heads dim into n_ctx
        """
        if len(tensor.shape) == 5:
            tensor = tensor.transpose(0, 1, 3, 2, 4)
        elif len(tensor.shape) == 4:
            tensor = tensor.transpose(0, 2, 1, 3)
        else:
            raise ValueError(f"Input tensor rank should be one of [4, 5], but is: {len(tensor.shape)}")
        new_shape = tensor.shape[:-2] + (num_attention_heads * attn_head_size,)
        return tensor.view(new_shape)

    def _attn(
            self,
            query,
            key,
            value,
            attention_mask=None,
            head_mask=None,
    ):

        # compute causal mask from causal mask buffer
        # query_length, key_length = query.shape(-2), key.shape(-2)
        query_length, key_length = query.shape[-2], key.shape[-2]
        causal_mask = self.causal_mask[:, :, key_length - query_length: key_length, :key_length]

        # Keep the attention weights computation in fp32 to avoid overflow issues
        query = query.astype(mindspore.float32)
        key = key.astype(mindspore.float32)

        attn_weights = ops.matmul(query, key.swapaxes(-1, -2))

        attn_weights = attn_weights / self.scale_attn
        # mask_value = torch.finfo(attn_weights.dtype).min
        dtype_info = np.finfo(np.float32 if attn_weights.dtype == mindspore.float32 else np.float64)
        mask_value = Tensor(np.array([dtype_info.min]), dtype=attn_weights.dtype)
        # 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 = Tensor(mask_value, dtype=attn_weights.dtype).astype(attn_weights.device)
        mask_value = Tensor(mask_value, dtype=attn_weights.dtype)
        attn_weights = mindspore.numpy.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.Softmax(axis=-1)(attn_weights)
        attn_weights = attn_weights.astype(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 = ops.matmul(attn_weights, value)

        return attn_output, attn_weights

    def construct(
            self,
            hidden_states: Optional[mindspore.Tensor],
            attention_mask: Optional[mindspore.Tensor] = None,
            layer_past: Optional[Tuple[mindspore.Tensor]] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            use_cache: Optional[bool] = False,
            output_attentions: Optional[bool] = False,
    ):

        qkv = self.qkv_proj(hidden_states)

        mp_num = 4
        qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))

        local_dim = self.head_dim * self.num_attention_heads // mp_num
        query, value, key = mindspore.ops.split(qkv_split, local_dim, axis=-1)
        query = self._split_heads(query, self.num_attention_heads, self.head_dim, mp_num=mp_num)
        key = self._split_heads(key, self.num_attention_heads, self.head_dim, mp_num=mp_num)

        value = self._split_heads(value, self.num_attention_heads, self.head_dim, mp_num=mp_num)
        value = value.transpose(0, 2, 1, 3)

        seq_len = key.shape[1]
        offset = 0

        if layer_past is not None:
            offset = layer_past[0].shape[-2]
            seq_len += offset

        if self.rotary_dim is not None:
            k_rot = key[:, :, :, : self.rotary_dim]
            k_pass = key[:, :, :, self.rotary_dim:]

            q_rot = query[:, :, :, : self.rotary_dim]
            q_pass = query[:, :, :, self.rotary_dim:]

            sincos = fixed_pos_embedding(k_rot, 1, seq_len=seq_len)
            k_rot = apply_rotary_pos_emb(k_rot, sincos, offset=offset)
            q_rot = apply_rotary_pos_emb(q_rot, sincos, offset=offset)

            key = mindspore.ops.cat([k_rot, k_pass], axis=-1)
            query = mindspore.ops.cat([q_rot, q_pass], axis=-1)
        else:
            sincos = fixed_pos_embedding(key, 1, seq_len=seq_len)
            key = apply_rotary_pos_emb(key, sincos, offset=offset)
            query = apply_rotary_pos_emb(query, sincos, offset=offset)

        key = key.transpose(0, 2, 1, 3)
        query = query.transpose(0, 2, 1, 3)

        if layer_past is not None:
            past_key = layer_past[0]
            past_value = layer_past[1]
            key = mindspore.ops.cat((past_key, key), axis=-2)
            value = mindspore.ops.cat((past_value, value), axis=-2)

        if use_cache is True:
            present = (key, value)
        else:
            present = None

        # compute self-attention: V x Softmax(QK^T)
        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)

        attn_output = self._merge_heads(attn_output, self.num_attention_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)


# Copied from transformers.models.gptj.modeling_gptj.GPTJMLP with GPTJ->CodeGen
class CodeGenMLP(nn.Cell):
    """
    CodeGenMLP
    """

    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * embed_dim
        super().__init__()
        embed_dim = config.n_embd

        self.fc_in = nn.Dense(embed_dim, intermediate_size)
        self.fc_out = nn.Dense(intermediate_size, embed_dim)

        self.act = ACT2FN[config.activation_function]
        self.dropout = nn.Dropout(p=config.resid_pdrop)

    def construct(self, hidden_states: Optional[mindspore.Tensor]) -> mindspore.Tensor:
        hidden_states = self.fc_in(hidden_states)
        hidden_states = self.act(hidden_states)
        hidden_states = self.fc_out(hidden_states)
        hidden_states = self.dropout(hidden_states)
        return hidden_states


# Copied from transformers.models.gptj.modeling_gptj.GPTJBlock with GPTJ->CodeGen
class CodeGenBlock(nn.Cell):
    """CodeGenBlock"""

    def __init__(self, config):
        super().__init__()
        inner_dim = config.n_inner if config.n_inner is not None else 4 * config.n_embd
        self.ln_1 = nn.LayerNorm((config.n_embd,), epsilon=config.layer_norm_epsilon)
        self.attn = CodeGenAttention(config)
        self.mlp = CodeGenMLP(inner_dim, config)

    def construct(
            self,
            hidden_states: Optional[mindspore.Tensor],
            layer_past: Optional[Tuple[mindspore.Tensor]] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            use_cache: Optional[bool] = False,
            output_attentions: Optional[bool] = False,
    ) -> Union[Tuple[mindspore.Tensor], Optional[Tuple[mindspore.Tensor, Tuple[mindspore.Tensor, ...]]]]:
        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:]

        feed_forward_hidden_states = self.mlp(hidden_states)
        hidden_states = attn_output + feed_forward_hidden_states + residual

        if use_cache:
            outputs = (hidden_states,) + outputs
        else:
            outputs = (hidden_states,) + outputs[1:]

        return outputs  # hidden_states, present, (attentions)


class CodeGenPreTrainedModel(PreTrainedModel):
    r"""
        CodeGenPreTrainedModel
    """

    config_class = CodeGenConfig
    base_model_prefix = "transformer"
    supports_gradient_checkpointing = True
    _no_split_modules = ["CodeGenBlock"]

    # def __init__(self, *inputs, **kwargs):
    #     super().__init__(*inputs, **kwargs)

    def get_position_embeddings(self):
        pass

    def resize_position_embeddings(self, new_num_position_embeddings: int):
        pass

    def save(self, save_dir: Union[str, os.PathLike]):
        pass

    def init_model_weights(self):
        pass

    def post_init(self):
        pass

    def _init_weights(self, module):
        if isinstance(module, (nn.Dense,)):
            # Slightly different from Mesh Transformer JAX 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)

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, CodeGenModel):
            module.gradient_checkpointing = value

    def get_head_mask(self, head_mask, num_hidden_layers, is_attention_chunked=False):
        """
        Prepare the head mask if needed.
        """
        if head_mask is not None:
            head_mask = self._convert_head_mask_to_5d(head_mask, num_hidden_layers)
            if is_attention_chunked is True:
                head_mask = head_mask.expand_dims(-1)
        else:
            head_mask = [None] * num_hidden_layers

        return head_mask


class CodeGenModel(CodeGenPreTrainedModel):
    r"""
        CodeGenModel
    """

    def __init__(self, config):
        super().__init__(config)

        self.embed_dim = config.n_embd
        self.vocab_size = config.vocab_size
        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
        self.drop = nn.Dropout(p=config.embd_pdrop)
        self.h = nn.CellList([CodeGenBlock(config) for _ in range(config.n_layer)])
        self.ln_f = nn.LayerNorm([self.embed_dim], epsilon=config.layer_norm_epsilon)
        self.rotary_dim = min(config.rotary_dim, config.n_ctx // config.n_head)

        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, value):
        self.wte = value

    def init_model_weights(self):
        pass

    def post_init(self):
        pass

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            past_key_values: Optional[Tuple[Tuple[mindspore.Tensor]]] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            position_ids: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            use_cache: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
    ):
        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

        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")
        if input_ids is not None:
            input_shape = input_ids.shape
            input_ids = input_ids.view(-1, input_shape[-1])
            batch_size = input_ids.shape[0]
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.shape[:-1]
            batch_size = inputs_embeds.shape[0]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        if token_type_ids is not None:
            token_type_ids = token_type_ids.view(-1, input_shape[-1])

        if position_ids is not None:
            position_ids = position_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].shape[-2]

        if position_ids is None:
            position_ids = ops.arange(past_length, input_shape[-1] + past_length, dtype=mindspore.int64)
            position_ids = position_ids.expand_dims(0).view(-1, input_shape[-1])

        # Attention mask.
        if attention_mask is not None:
            if batch_size <= 0:
                raise ValueError("batch_size has to be defined and > 0")
            attention_mask = attention_mask.view(batch_size, -1)
            # We create a 3D attention mask from a 2D tensor mask.
            # Sizes are [batch_size, 1, 1, to_seq_length]
            # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
            # this attention mask is more simple than the triangular masking of causal attention
            # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
            attention_mask = attention_mask[:, None, None, :]

            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
            # masked positions, this operation will create a tensor which is 0.0 for
            # positions we want to attend and the dtype's smallest value for masked positions.
            # Since we are adding it to the raw scores before the softmax, this is
            # effectively the same as removing these entirely.
            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
            attention_mask = (1.0 - attention_mask) * np.finfo(self.dtype).min

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x num_attention_heads x N x N
        # head_mask has shape n_layer x batch x num_attention_heads x N x N
        head_mask = self.get_head_mask(head_mask, self.config.n_layer)

        if inputs_embeds is None:
            inputs_embeds = self.wte(input_ids)

        hidden_states = inputs_embeds

        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 = input_shape + (hidden_states.shape[-1],)

        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,)

            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,)

        return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)


class CodeGenForCausalLM(CodeGenPreTrainedModel):
    """
        CodeGenForCausalLM
    """

    def __init__(self, config):
        super().__init__(config)
        self.transformer = CodeGenModel(config)
        self.lm_head = nn.Dense(config.n_embd, config.vocab_size)

    def get_output_embeddings(self):
        """
        get_output_embeddings
        """
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        """
        set_output_embeddings
        """
        self.lm_head = new_embeddings

    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, **kwargs):
        """
        prepare_inputs_for_generation
        """
        token_type_ids = kwargs.get("token_type_ids", None)
        # only last token for inputs_ids if past is defined in kwargs
        if past_key_values:
            input_ids = input_ids[:, -1].unsqueeze(-1)
            if token_type_ids is not None:
                token_type_ids = token_type_ids[:, -1].unsqueeze(-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[:, -1].unsqueeze(-1)
        else:
            position_ids = None
        return {
            "input_ids": input_ids,
            "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,
        }

    def construct(
            self,
            input_ids: Optional[mindspore.Tensor] = None,
            past_key_values: Optional[Tuple[Tuple[mindspore.Tensor]]] = None,
            attention_mask: Optional[mindspore.Tensor] = None,
            token_type_ids: Optional[mindspore.Tensor] = None,
            position_ids: Optional[mindspore.Tensor] = None,
            head_mask: Optional[mindspore.Tensor] = None,
            inputs_embeds: Optional[mindspore.Tensor] = None,
            labels: Optional[mindspore.Tensor] = None,
            use_cache: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
    ):
        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,
        )
        hidden_states = transformer_outputs[0]

        # make sure sampling in fp16 works correctly and
        # 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 = self.lm_head(hidden_states).to(mindspore.float32)

        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = lm_logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = nn.CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, shift_logits.shape[-1]), shift_labels.view(-1))

            loss = loss.astype(hidden_states.dtype)

        output = (lm_logits,) + transformer_outputs[1:]
        return ((loss,) + output) if loss is not None else output

    @staticmethod
    def _reorder_cache(
            past_key_values: Tuple[Tuple[mindspore.Tensor]], beam_idx: mindspore.Tensor
    ):
        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
        )