第三课：GPT

第三课：GPT

1、学习总结：

GPT出现的原因

未标注的文本数据远多于已标注的文本数据，并且对于不同的下游任务会存在不同的标注方式

GPT的方法原理

半监督学习

• 基于大量未标注的文本数据，训练预训练语言模型
• 使用已标注文本数据，对模型针对某一特定下游任务进行finetune，只更改output layer（线性层）

目前存在的问题

• 自然语言处理的下游任务非常多元，难以有统一的优化目标
• 难以将预训练模型的信息完全传递到finetune的下游任务中

无监督的预训练

优化目标

模型结构

由于训练objective的选择，gpt在模型选择上不应该看见当前token后的信息，故模型应设计为单向网络，即transformer中的decoder结构。

import os
import logging
import numpy as np
import mindspore
from mindspore import nn
from mindspore import ops
from mindspore import Tensor
from mindspore.common.initializer import initializer, Normal
from mindnlp.models.gpt.gpt_config import GPTConfig
from mindnlp._legacy.nn import Dropout
from mindnlp.abc import PreTrainedModel
from mindnlp.models.utils import Conv1D, prune_conv1d_layer, find_pruneable_heads_and_indices
from mindnlp.models.utils import SequenceSummary
from mindnlp.models.activations import ACT2FN
from mindnlp import GPTConfig

# Feed-Forward 实现
class MLP(nn.Cell):
    r"""
    GPT MLP
	"""

    def __init__(self, n_state, config):
        super().__init__()
        n_embd = config.n_embd
        self.c_fc = Conv1D(n_state, n_embd)
        self.c_proj = Conv1D(n_embd, n_state)
        self.act = ACT2FN[config.afn]
        self.dropout = Dropout(p=config.resid_pdrop)

    def construct(self, x):
        h = self.act(self.c_fc(x))
        h2 = self.c_proj(h)
        return self.dropout(h2)

# Multi-head attention 实现

class Attention(nn.Cell):
    r"""
    GPT Attention
    """

    def __init__(self, nx, n_positions, config, scale=False):
        super().__init__()
        n_state = nx
        if n_state % config.n_head != 0:
            raise ValueError(f"Attention n_state shape: {n_state} must be divisible by config.n_head {config.n_head}")

        self.bias = Tensor(np.tril(np.ones((n_positions, n_positions))), mindspore.float32).view(1, 1, n_positions, n_positions)
        self.n_head = config.n_head
        self.split_size = n_state
        self.scale = scale

        self.c_attn = Conv1D(n_state * 3, n_state)
        self.c_attn = Conv1D(n_state * 3, n_state)
        self.c_proj = Conv1D(n_state, n_state)
        self.attn_dropout = Dropout(p=config.attn_pdrop)
        self.resid_dropout = Dropout(p=config.resid_pdrop)
        self.pruned_heads = set()

        self.output_attentions = config.output_attentions

    def prune_heads(self, heads):
        """
        Prunes heads of the model.
        """
        if len(heads) == 0:
            return
        head_size = self.split_size//self.n_head
        heads, index = find_pruneable_heads_and_indices(heads, self.n_head, head_size, self.pruned_heads)
        index_attn = ops.cat([index, index + self.split_size, index + (2 * self.split_size)])
        self.c_attn = prune_conv1d_layer(self.c_attn, index_attn, axis=1)
        self.c_proj = prune_conv1d_layer(self.c_proj, index, axis=0)
        self.split_size = (self.split_size // self.n_head) * (self.n_head - len(heads))
        self.n_head = self.n_head - len(heads)
        self.pruned_heads = self.pruned_heads.union(heads)

    def _attn(self, q, k, v, attention_mask=None, head_mask=None):
        w = ops.matmul(q, k)
        if self.scale:
            w = w / ops.sqrt(ops.scalar_to_tensor(v.shape[-1]))
        b = self.bias[:, :, : w.shape[-2], : w.shape[-1]]
        w = w * b + -1e9 * (1 - b)

        if attention_mask is not None:
            w = w + attention_mask

        w = ops.softmax(w)
        w = self.attn_dropout(w)

        if head_mask is not None:
            w = w * head_mask

        outputs = (ops.matmul(w, v),)
        if self.output_attentions:
            outputs += (w,)
        return outputs


    def merge_heads(self, x):
        """merge heads"""
        x = x.transpose(0, 2, 1, 3)
        new_x_shape = x.shape[:-2] + (x.shape[-2] * x.shape[-1],)
        return x.view(new_x_shape)

    def split_heads(self, x, k=False):
        """split heads"""
        new_x_shape = x.shape[:-1] + (self.n_head, x.shape[-1] // self.n_head)
        x = x.view(new_x_shape)
        if k:
            return x.transpose(0, 2, 3, 1)
        return x.transpose(0, 2, 1, 3)

    def construct(self, x, attention_mask=None, head_mask=None):
        x = self.c_attn(x)
        query, key, value = ops.split(x, self.split_size, axis=2)
        query = self.split_heads(query)
        key = self.split_heads(key, k=True)
        value = self.split_heads(value)

        attn_outputs = self._attn(query, key, value, attention_mask, head_mask)
        a = attn_outputs[0]

        a = self.merge_heads(a)
        a = self.c_proj(a)
        a = self.resid_dropout(a)
        outputs = (a,) + attn_outputs[1:]
        return outputs

class Block(nn.Cell):
    r"""
    GPT Block
    """

    def __init__(self, n_positions, config, scale=False):
        super().__init__()
        nx = config.n_embd
        self.attn = Attention(nx, n_positions, config, scale)
        self.ln_1 = nn.LayerNorm((nx,), epsilon=config.layer_norm_epsilon)
        self.mlp = MLP(4 * nx, config)
        self.ln_2 = nn.LayerNorm((nx,), epsilon=config.layer_norm_epsilon)

    def construct(self, x, attention_mask=None, head_mask=None):
        output_attn = self.attn(
            x,
            attention_mask=attention_mask,
            head_mask=head_mask
        )

        a = output_attn[0]
        n = self.ln_1(x + a)
        m = self.mlp(n)
        h = self.ln_2(n + m)

        outputs = (h,) + output_attn[1:]
        return outputs

class GPTPreTrainedModel(PreTrainedModel):
    """BertPretrainedModel"""
    convert_torch_to_mindspore = torch_to_mindspore
    pretrained_model_archive_map = PRETRAINED_MODEL_ARCHIVE_MAP
    config_class = GPTConfig
    base_model_prefix = 'transformer'

    def _init_weights(self, cell):
        """Initialize the weights"""
        if isinstance(cell, nn.Dense):
            cell.weight.set_data(initializer(Normal(self.config.initializer_range),
                                                    cell.weight.shape, cell.weight.dtype))
            if cell.has_bias:
                cell.bias.set_data(initializer('zeros', cell.bias.shape, cell.bias.dtype))
        elif isinstance(cell, nn.Embedding):
            embedding_table = initializer(Normal(self.config.initializer_range),
                                                 cell.embedding_table.shape,
                                                 cell.embedding_table.dtype)
            if cell.padding_idx is not None:
                embedding_table[cell.padding_idx] = 0
            cell.embedding_table.set_data(embedding_table)
        elif isinstance(cell, nn.LayerNorm):
            cell.gamma.set_data(initializer('ones', cell.gamma.shape, cell.gamma.dtype))
            cell.beta.set_data(initializer('zeros', cell.beta.shape, cell.beta.dtype))

class GPTModel(GPTPreTrainedModel):
    """
    The bare GPT transformer model outputting raw hidden-states without any specific head on top
    """

    def __init__(self, config):
        super().__init__(config)
        self.config = config
        self.tokens_embed = nn.Embedding(config.vocab_size, config.n_embd)
        self.positions_embed = nn.Embedding(config.n_positions, config.n_embd)
        self.drop = nn.Dropout(p=config.embd_pdrop)
        self.h = nn.CellList([Block(config.n_positions, config, scale=True) for _ in range(config.n_layer)])
        self.position_ids = ops.arange(config.n_positions)

        self.n_layer = self.config.n_layer
        self.output_attentions = self.config.output_attentions
        self.output_hidden_states = self.config.output_hidden_states

    def get_input_embeddings(self):
        """
        return the input embeddings layer
        """
        return self.tokens_embed

    def set_input_embeddings(self, value):
        """
        set the input embeddings layer
        """
        self.tokens_embed = 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}
        """
        for layer, heads in heads_to_prune.items():
            self.h[layer].attn.prune_heads(heads)

    def construct(
            self,
            input_ids=None,
            attention_mask=None,
            token_type_ids=None,
            position_ids=None,
            head_mask=None,
            inputs_embeds=None,
    ):
        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])
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.shape[:-1]
        else:
            raise ValueError("You have to specify either input_ids or inputs_embeds")

        if position_ids is None:
            position_ids = self.position_ids[None, : input_shape[-1]]

        if attention_mask is not None:
            attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
            attention_mask = attention_mask.to(dtype=next(self.parameters()).dtype)
            attention_mask = (1.0 - attention_mask) * Tensor(np.finfo(mindspore.dtype_to_nptype(self.dtype)).min,
                                                             self.dtype)

        head_mask = self.get_head_mask(head_mask, self.n_layer)

        if inputs_embeds is None:
            inputs_embeds = self.tokens_embed(input_ids)
        position_embeds = self.positions_embed(position_ids)
        if token_type_ids is not None:
            token_type_ids = token_type_ids.view(-1, token_type_ids.shape[-1])
            token_type_embeds = self.tokens_embed(token_type_ids)
        else:
            token_type_embeds = 0
        hidden_states = inputs_embeds + position_embeds + token_type_embeds
        hidden_states = self.drop(hidden_states)

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

        all_attentions = ()
        all_hidden_states = ()
        for i, block in enumerate(self.h):
            if self.output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            outputs = block(hidden_states, attention_mask, head_mask[i])
            hidden_states = outputs[0]
            if self.output_attentions:
                all_attentions = all_attentions + (outputs[1],)

        hidden_states = hidden_states.view(*output_shape)

        if self.output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        return (hidden_states, all_hidden_states, all_attentions)

监督微调

在已经预训练好的GPT上额外加一层线性层

并通过缩小目标与计算结果的误差进行模型优化

最终为加速模型收敛及提高模型的泛化性，融入pretrain时language modelling的优化目标

# 文本序列分类任务

class GPTForSequenceClassification(GPTPreTrainedModel):
    """
    The Original GPT Model transformer with a sequence classification head on top (linear layer).
    GPTForSequenceClassification uses the last token in order to do the classification, as other causal
    models (e.g. GPT-2) 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).
    """
    def __init__(self, config):
        super().__init__(config)
        self.config = config
        self.num_labels = config.num_labels
        self.transformer = GPTModel(config)
        self.score = nn.Dense(config.n_embd, self.num_labels, has_bias=False)

        self.pad_token_id = self.config.pad_token_id
        problem_type = config.problem_type
        if problem_type is None:
            self.loss = None
        else:
            if self.num_labels == 1:
                self.problem_type = "regression"
                self.loss = nn.MSELoss()
            elif self.num_labels > 1:
                self.problem_type = "single_label_classification"
                self.loss = nn.CrossEntropyLoss()
            else:
                self.problem_type = "multi_label_classification"
                self.loss = nn.BCEWithLogitsLoss()

    def construct(
        self,
        input_ids = None,
        attention_mask = None,
        token_type_ids = None,
        position_ids = None,
        head_mask = None,
        inputs_embeds = None,
        labels = None,
    ):
        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).
        """
        transformer_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,
        )

        hidden_states = transformer_outputs[0]
        logits = self.score(hidden_states)

        if input_ids is not None:
            batch_size, _ = input_ids.shape[:2]
        else:
            batch_size, _ = inputs_embeds.shape[:2]

        if self.pad_token_id is None and batch_size != 1:
            raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")

        if self.pad_token_id is None:
            sequence_lengths = -1
        else:
            if input_ids is not None:
                sequence_lengths = ops.ne(input_ids, self.pad_token_id).sum(-1) - 1
            else:
                sequence_lengths = -1

        pooled_logits = logits[:, sequence_lengths]

        loss = None

        output = (pooled_logits,) + transformer_outputs[1:]

        if labels is not None:
            if self.num_labels == 1:
                loss = self.loss(pooled_logits.squeeze(), labels.squeeze())
            elif self.num_labels > 1:
                loss = self.loss(pooled_logits.view(-1, self.num_labels), labels.view(-1))
            else:
                loss = self.loss(pooled_logits, labels)

        if loss is not None:
            output = (loss,) + output
        return output

课程ppt及代码地址

• github地址（网络不好的可以访问下面我克隆到gitee上的地址）：GPT

• gitee地址：GPT

2、学习心得：

? 通过本次学习，熟悉了Mindspore这个国产深度学习框架，也对GPT的基本技术原理有所了解，同时也学会了如何在IMDb数据集上微调GPT完成一个情感分类的任务，比较有成就感！！！另外就是Selina小姐姐讲课的氛围比较轻松，学起来比较快乐！

3、经验分享：

? 在启智openI上的npu跑时记得使用mindspore1.7的镜像，同时安装对应mindnlp的版本，不然可能会因为版本不兼容而报错。另外就是IMDB情感分类任务的微调练习一定要做，这样能比较get到整个微调的全流程是怎样的，后面再去学习llama等模型的微调时就会更加得心应手。

4、课程反馈：

? 本次课程中的代码串讲我觉得是做的最好的地方，没有照着ppt一直念，而是在jupyter lab上把代码和原理结合到一块进行讲解，让学习者对代码的理解更加深入。我觉得内容的最后可以稍微推荐一下与Mindspore大模型相关的套件，让学习者在相关套件上可以开发出更多好玩和有趣的东西！

5、使用MindSpore昇思的体验和反馈：

MindSpore昇思的优点和喜欢的方面：

1. 灵活性和可扩展性： MindSpore提供了灵活的编程模型，支持静态计算图和动态计算图。这种设计使得它适用于多种类型的机器学习和深度学习任务，并且具有一定的可扩展性。
2. 跨平台支持： MindSpore支持多种硬件平台，包括CPU、GPU和NPU等，这使得它具有在不同设备上运行的能力，并能充分利用各种硬件加速。
3. 自动并行和分布式训练： MindSpore提供了自动并行和分布式训练的功能，使得用户可以更轻松地处理大规模数据和模型，并更高效地进行训练。
4. 生态系统和社区支持： MindSpore致力于建立开放的生态系统，并鼓励社区贡献，这对于一个开源框架来说非常重要，能够帮助用户更好地学习和解决问题。

一些建议和改进方面：

1. 文档和教程的改进： 文档和教程并不是很详细，希望能够提供更多实用的示例、详细的文档和教程，以帮助用户更快速地上手和解决问题。
2. 更多的应用场景示例： 提供更多真实场景的示例代码和应用案例，可以帮助用户更好地了解如何在实际项目中应用MindSpore。

6、未来展望：

? 大模型的内容还是很多的，希望自己能坚持打卡，将后面的内容都学习完，并做出一些有趣好玩的东西来！