modeling_telechat2.py

# coding=utf-8
# Copyright 2022 HuggingFace Inc. team and BigScience workshop.
#
# 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.

# Copyright (c) 2023, NVIDIA CORPORATION.  All rights reserved.

# Copyright (c) 2021 EleutherAI
# This file is based on code by the authors denoted below and has been modified from its original version.
#
# Copyright (c) 2020, NVIDIA CORPORATION.  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 TELECHAT model."""

import warnings
from typing import Optional, Tuple, Union, List, Dict
from threading import Thread

import torch
import math
import copy
from torch import nn
import torch.utils.checkpoint
from torch.nn import functional as F
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, LayerNorm, MSELoss
from transformers.modeling_outputs import (
    BaseModelOutputWithPastAndCrossAttentions,
    CausalLMOutputWithCrossAttentions
)
from transformers.modeling_utils import PreTrainedModel
from transformers.utils import logging
from transformers import GenerationConfig

from .configuration_telechat2 import Telechat2Config


logger = logging.get_logger(__name__)

_CHECKPOINT_FOR_DOC = "telechat"
_CONFIG_FOR_DOC = "Telechat2Config"

TELECHAT_PRETRAINED_MODEL_ARCHIVE_LIST = []

try:
    from einops import rearrange
except ImportError:
    rearrange = None

use_flash_attn = True
try:
    from flash_attn.flash_attn_interface import flash_attn_unpadded_func
except ImportError:
    try:
        from flash_attn.flash_attn_interface import flash_attn_varlen_func as flash_attn_unpadded_func
    except ImportError:
        flash_attn_unpadded_func = None


class RotaryEmbedding(torch.nn.Module):
    # Extracted from: https://github.com/EleutherAI/gpt-neox
    def __init__(self, dim, config):
        super().__init__()
        self.config = config
        self.dim = dim
        self.base = config.rope_theta
        self.inv_freq = 1. / (self.base ** (torch.arange(0, dim, 2).float().half() / dim))
        self.max_seq_len_cached = None
        self.cos_cached = None
        self.sin_cached = None
        self.precision = config.torch_dtype

    def get_mscale(self, scale=1):
        if scale <= 1:
            return 1.0
        return 0.1 * math.log(scale) + 1.0

    def get_ntk_alpha(self, true_seq_len):
        context_value = math.log(true_seq_len / self.config.base_seqlen, 2) + 1
        # ntk_alpha = 2 ** context_value - 1
        ntk_alpha = 2 ** math.ceil(context_value) - 1
        ntk_alpha = max(ntk_alpha, 1)
        return ntk_alpha

    def forward(self, x, seq_dim=0, seq_len=None):
        if seq_len is None:
            seq_len = x.shape[seq_dim]
        seq_len = max(seq_len, self.config.training_seqlen)
        ntk_alpha = self.get_ntk_alpha(seq_len)
        self.mscale = float(self.get_mscale(seq_len / self.config.training_seqlen))
        if True:
            base = self.base * ntk_alpha ** (self.dim / (self.dim - 2))
            self.inv_freq = 1.0 / (base ** (torch.arange(0, self.dim, 2, device=x.device).float() / self.dim))
            self.max_seq_len_cached = seq_len
            t = torch.arange(self.max_seq_len_cached, device=x.device, dtype=self.inv_freq.dtype)
            freqs = torch.einsum('i,j->ij', t, self.inv_freq)
            # Different from paper, but it uses a different permutation in order to obtain the same calculation
            emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
            if self.precision == torch.bfloat16:
                emb = emb.float()
            # [sx, 1 (b * np), hn]
            self.cos_cached = self.mscale * emb.cos()[:, None, :].half()
            self.sin_cached = self.mscale * emb.sin()[:, None, :].half()
            if self.precision == torch.bfloat16:
                self.cos_cached = self.cos_cached.bfloat16()
                self.sin_cached = self.sin_cached.bfloat16()
        return self.cos_cached[:seq_len, ...], self.sin_cached[:seq_len, ...]


# rotary pos emb helpers:
def rotate_half(x):
    x1, x2 = x[..., :x.shape[-1] // 2], x[..., x.shape[-1] // 2:]
    return torch.cat((-x2, x1), dim=x1.ndim - 1)  # dim=-1 triggers a bug in earlier torch versions


def apply_rotary_pos_emb_torch(q, k, cos, sin, offset: int = 0):  # jitting fails with bf16
    cos, sin = cos[offset:q.shape[0] + offset, ...], sin[offset:q.shape[0] + offset, ...]
    return (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)


class MixedFusedRMSNorm(nn.Module):
    # Extracted from https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/modeling_llama.py
    def __init__(self, hidden_size, eps=1e-6):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        return self.weight * hidden_states.to(input_dtype)


class FlashSelfAttention(torch.nn.Module):
    # Extracted from https://github.com/microsoft/Megatron-DeepSpeed/blob/main/megatron/model/transformer.py
    """Implement the scaled dot product attention with softmax.
    Arguments
    ---------
        softmax_scale: The temperature to use for the softmax attention.
                      (default: 1/sqrt(d_keys) where d_keys is computed at
                      runtime)
        attention_dropout: The dropout rate to apply to the attention
                           (default: 0.0)
    """

    def __init__(self, causal=False, softmax_scale=None, attention_dropout=0.0,
                 device=None, dtype=None):
        super().__init__()
        assert flash_attn_unpadded_func is not None, ('Please install FlashAttention first, '
                                                      'e.g., with pip install flash-attn')
        assert rearrange is not None, 'Please install einops first, e.g., with pip install einops'
        self.causal = causal
        self.softmax_scale = softmax_scale
        self.dropout_p = attention_dropout

    def forward(self, q, k, v):
        """Implements the multihead softmax attention.
        Arguments
        ---------
            q, k, v: The tensor containing the query, key, and value. (B, S, H, D)
        """
        assert all((i.dtype in [torch.float16, torch.bfloat16] for i in (q, k, v)))
        assert all((i.is_cuda for i in (q, k, v)))

        batch_size, seqlen_q = q.shape[0], q.shape[1]
        seqlen_k = k.shape[1]

        q, k, v = [rearrange(x, 'b s ... -> (b s) ...') for x in [q, k, v]]
        cu_seqlens_q = torch.arange(0, (batch_size + 1) * seqlen_q, step=seqlen_q, dtype=torch.int32,
                                    device=q.device)
        self.training = False
        if self.training:
            # during training q,k,v always have same seqlen
            assert seqlen_k == seqlen_q

            is_causal = self.causal
            cu_seqlens_k = cu_seqlens_q
            dropout_p = self.dropout_p
        else:
            # turn off FA causal mask after first inference autoregressive iteration
            # only on first autoregressive step q,k,v have same seqlen
            is_causal = seqlen_q == seqlen_k
            cu_seqlens_k = torch.arange(0, (batch_size + 1) * seqlen_k, step=seqlen_k, dtype=torch.int32,
                                        device=q.device)
            dropout_p = 0

        output = flash_attn_unpadded_func(
            q, k, v, cu_seqlens_q, cu_seqlens_k, seqlen_q, seqlen_k,
            dropout_p=dropout_p,
            softmax_scale=self.softmax_scale, causal=is_causal
        )

        output = rearrange(output, '(b s) ... -> b s ...', b=batch_size)
        return output


def _make_causal_mask(
        input_ids_shape: torch.Size, device: torch.device, past_key_values_length: int
) -> torch.BoolTensor:
    """
    Make causal mask used for self-attention.
    """
    batch_size, target_length = input_ids_shape
    mask = torch.empty((target_length, target_length + past_key_values_length), dtype=torch.bool, device=device)
    # ONNX doesn't support `torch.Tensor.triu` properly, thus we use this workaround
    seq_ids = torch.arange(target_length, device=device)
    mask[:, past_key_values_length:] = seq_ids[:, None] < seq_ids[None, :]

    if past_key_values_length > 0:
        mask[:, :past_key_values_length] = False

    expanded_mask = mask[None, None, :, :].expand(batch_size, 1, target_length, target_length + past_key_values_length)
    return expanded_mask


def _expand_mask(mask: torch.Tensor, tgt_length: int) -> torch.BoolTensor:
    """
    Expands attention_mask from `[batch_size, src_length]` to `[batch_size, 1, tgt_length, src_length]`.
    """
    batch_size, src_length = mask.shape
    tgt_length = tgt_length if tgt_length is not None else src_length

    expanded_mask = ~(mask[:, None, None, :].to(torch.bool))
    return expanded_mask.expand(batch_size, 1, tgt_length, src_length)


def dropout_add(x: torch.Tensor, residual: torch.Tensor, prob: float, training: bool) -> torch.Tensor:
    """
    Dropout add function

    Args:
        x (`torch.tensor`, *required*):
            input tensor
        residual (`torch.tensor`, *required*):
            residual tensor
        prob (`float`, *required*):
            dropout probability
        training (`bool`, *required*):
            training mode
    """
    out = F.dropout(x, p=prob, training=training)
    out = residual + out
    return out


def telechat_gelu_forward(x: torch.Tensor) -> torch.Tensor:
    """
    Custom bias GELU function. Adapted from Megatron-DeepSpeed code. Here we use a simple implementation (inference) to
    make the model jitable.

    Args:
        x (`torch.tensor`, *required*):
            input hidden states
    """
    return x * 0.5 * (1.0 + torch.tanh(0.79788456 * x * (1 + 0.044715 * x * x)))


def telechat_gelu_back(g: torch.Tensor, x: torch.Tensor) -> torch.Tensor:
    """
    gradient of tanh approximation of gelu gradient of actual gelu is: 0.5 * (1. + torch.erf(x * 0.70710678)) +
    0.3989423 * x * torch.exp(-0.5 * x * x)

    Args:
        g (`torch.tensor`, *required*):
            gradient output tensor
        x (`torch.tensor`, *required*):
            input tensor
    """
    x = x[0]  # x is a tuple of 1 element, needs to unpack it first
    tanh_out = torch.tanh(0.79788456 * x * (1 + 0.044715 * x * x))
    # sqrt(2/pi) * 3 * 0.044715 -> 0.1070322243
    ff = 0.5 * x * ((1 - tanh_out * tanh_out) * (0.79788456 + 0.1070322243 * x * x)) + 0.5 * (1 + tanh_out)
    return ff * g


class GeLUFunction(torch.autograd.Function):
    @staticmethod
    def forward(ctx, input: torch.Tensor) -> torch.Tensor:
        ctx.save_for_backward(input)
        return telechat_gelu_forward(input)

    @staticmethod
    def backward(ctx, grad_output: torch.Tensor) -> torch.Tensor:
        input = ctx.saved_tensors
        tmp = telechat_gelu_back(grad_output, input)
        return tmp


class TelechatGelu(nn.Module):
    """
    TelechatBiasGelu wrapper function that make use of the simple function on inference mode to make the model
    torchscriptable and use the autograd function in training mode to get the accurate results of the gradients Partly
    copied from Megatron-DeepSpeed code and adapted for our needs

    See here why autograd functions are not torchscriptable: https://github.com/pytorch/pytorch/issues/22329
    """

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

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        if self.training:
            return GeLUFunction.apply(x)
        else:
            return telechat_gelu_forward(x)


class TelechatAttention(nn.Module):
    def __init__(self, config: Telechat2Config, layer_idx):
        super().__init__()
        self.kv_cache = None
        self.layer_idx = layer_idx

        self.hidden_size = config.hidden_size
        self.num_heads = config.n_head
        self.head_dim = self.hidden_size // self.num_heads
        self.split_size = self.hidden_size
        self.hidden_dropout = config.hidden_dropout
        self.config = config

        if self.head_dim * self.num_heads != self.hidden_size:
            raise ValueError(
                f"`hidden_size` must be divisible by num_heads (got `hidden_size`: {self.hidden_size} and `num_heads`:"
                f" {self.num_heads})."
            )

        # Layer-wise attention scaling
        self.inv_norm_factor = 1.0 / math.sqrt(self.head_dim)
        self.beta = 1.0

        self.num_key_value_heads = config.num_key_value_heads if config.num_key_value_heads else self.num_heads
        self.kv_projection_size = self.head_dim * self.num_key_value_heads
        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
        self.query = nn.Linear(self.hidden_size, self.hidden_size, bias=False)
        self.key_value = nn.Linear(self.hidden_size, self.kv_projection_size * 2, bias=False)
        self.dense = nn.Linear(self.hidden_size, self.hidden_size)
        self.attention_dropout = nn.Dropout(config.attention_dropout)
        self.rotary_emb = RotaryEmbedding(self.head_dim, config=config)

        self.core_attention_flash = FlashSelfAttention(
            causal=True, attention_dropout=config.attention_dropout
        )

        self.last_key_layer = None
        # logn_list = [math.log(i, 4096) if i > 4096 else 1 for i in range(1, 32768)]
        # self.logn_tensor = torch.tensor(logn_list)[None, :, None, None].half().cuda()

    def repeat_kv(self, hidden_states, n_rep):
        slen, batch, num_key_value_heads_per_partition, head_dim = hidden_states.shape
        if n_rep == 1:
            return hidden_states
        hidden_states = hidden_states[:, :, :, None, :].expand(slen, batch, num_key_value_heads_per_partition, n_rep,
                                                               head_dim)
        return hidden_states.reshape(slen, batch, num_key_value_heads_per_partition * n_rep, head_dim)

    def split_tensor_along_last_dim(self,
                                    tensor: torch.Tensor,
                                    num_partitions: int,
                                    contiguous_split_chunks: bool = False,
                                    ):

        # Get the size and dimension.
        last_dim = tensor.dim() - 1
        last_dim_size = tensor.size()[last_dim] // num_partitions
        # Split.
        tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
        # Note: torch.split does not create contiguous tensors by default.
        if contiguous_split_chunks:
            return tuple(chunk.contiguous() for chunk in tensor_list)

        return tensor_list

    def _merge_heads(self, x: torch.Tensor) -> torch.Tensor:
        batch_size_and_num_heads, seq_length, _ = x.shape
        batch_size = batch_size_and_num_heads // self.num_heads
        x = x.view(batch_size, self.num_heads, seq_length, self.head_dim)
        x = x.permute(0, 2, 1, 3)
        return x.reshape(batch_size, seq_length, self.num_heads * self.head_dim)

    def forward(
            self,
            hidden_states: torch.Tensor,
            residual: torch.Tensor,
            attention_mask: torch.Tensor,
            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
            use_cache: bool = False,
            output_attentions: bool = False,
    ):
        hidden_states = hidden_states.transpose(1, 0)
        query_layer = self.query(hidden_states)
        new_tensor_shape = query_layer.size()[:-1] + \
                           (self.num_heads,
                            self.head_dim)
        query_layer = query_layer.view(*new_tensor_shape)

        mixed_kv_layer = self.key_value(hidden_states)
        new_tensor_shape = mixed_kv_layer.size()[:-1] + \
                           (self.num_key_value_heads,
                            2 * self.head_dim)
        mixed_kv_layer = mixed_kv_layer.view(*new_tensor_shape)
        (key_layer, value_layer) = self.split_tensor_along_last_dim(mixed_kv_layer, 2)

        output_size = (query_layer.size(1),
                       query_layer.size(2),
                       query_layer.size(0),
                       key_layer.size(0),
                       key_layer.size(2)
                       )

        query_layer = query_layer.view(output_size[2], output_size[0] * output_size[1], -1)
        key_layer = key_layer.view(output_size[3], output_size[0] * output_size[4], -1)

        apply_rotary_fn = apply_rotary_pos_emb_torch

        seq_len = key_layer.shape[0]
        offset = 0

        if use_cache and layer_past != None:
            past_key, past_value = layer_past
            offset = past_key.shape[0]
            seq_len += offset

        cos, sin = self.rotary_emb(value_layer, seq_len=seq_len)

        query_layer, key_layer = apply_rotary_fn(query_layer, key_layer, cos, sin, offset=offset)
        if use_cache:
            if layer_past != None:
                past_key, past_value = layer_past
                key_layer = torch.cat((past_key, key_layer[-1, ...].unsqueeze(0)), dim=0)
                value_layer = torch.cat((past_value, value_layer[-1, ...].unsqueeze(0)), dim=0)
            layer_past = key_layer, value_layer

        s_value, bz, kv_head, dim = value_layer.shape
        s_key = key_layer.shape[0]
        s_query = query_layer.shape[0]
        q_head = output_size[1]

        query_layer = query_layer.reshape((s_query, bz, q_head, dim))
        key_layer = key_layer.reshape((s_key, bz, kv_head, dim))

        key_layer = self.repeat_kv(key_layer, self.num_key_value_groups)
        value_layer = self.repeat_kv(value_layer, self.num_key_value_groups)

        if self.config.flash_attn:
            q, k, v = [rearrange(x, 's b ... -> b s ...').contiguous() for x in
                       (query_layer, key_layer, value_layer)]
            context_layer = self.core_attention_flash(q, k, v)
            context_layer = rearrange(context_layer, 'b s h d -> b s (h d)').contiguous()
        else:
            ##[sq, b, np, hn] -> [sq, b * np, hn]
            query_layer = query_layer.reshape(s_query, bz * self.num_heads, dim)
            # [sk, b, np, hn] -> [sk, b * np, hn]
            key_layer = key_layer.reshape(s_key, bz * self.num_heads, dim)
            matmul_result = self.inv_norm_factor * torch.einsum('bik,bkj->bij', query_layer.transpose(0, 1),
                                                                key_layer.transpose(0, 1).transpose(1, 2))

            attention_scores = matmul_result.view(bz, self.num_heads, s_query, s_key)

            input_dtype = attention_scores.dtype
            if input_dtype == torch.float16:
                attention_scores = attention_scores.to(torch.float)
            attn_weights = torch.masked_fill(attention_scores, attention_mask, torch.finfo(attention_scores.dtype).min)
            attention_probs = F.softmax(attn_weights, dim=-1).to(input_dtype)  ##dtype = torch.float32
            attention_probs = self.attention_dropout(attention_probs)
            attention_probs_reshaped = attention_probs.view(bz * self.num_heads, s_query, s_key)

            value_layer = value_layer.reshape(s_key, bz * self.num_heads, dim)
            context_layer = torch.bmm(attention_probs_reshaped, value_layer.transpose(0, 1))
            context_layer = self._merge_heads(context_layer)
        output_tensor = self.dense(context_layer)

        output_tensor = dropout_add(output_tensor, residual, self.hidden_dropout, self.training)
        present = None
        outputs = (output_tensor, present)
        if output_attentions:
            outputs += (attention_probs,)

        return output_tensor, layer_past


class TelechatMLP(nn.Module):
    def __init__(self, config: Telechat2Config):
        super().__init__()
        hidden_size = config.hidden_size
        self.gate_proj = nn.Linear(hidden_size, config.ffn_hidden_size, bias=False)
        self.up_proj = nn.Linear(hidden_size, config.ffn_hidden_size, bias=False)
        self.down_proj = nn.Linear(config.ffn_hidden_size, hidden_size, bias=True)
        self.hidden_dropout = config.hidden_dropout

    def forward(self, hidden_states: torch.Tensor, residual: torch.Tensor) -> torch.Tensor:
        intermediate_output = self.down_proj(F.silu(self.gate_proj(hidden_states)) * self.up_proj(hidden_states))
        output = dropout_add(intermediate_output, residual, self.hidden_dropout, self.training)
        return output


class TelechatBlock(nn.Module):
    def __init__(self, config: Telechat2Config, layer_idx):
        super().__init__()
        hidden_size = config.hidden_size

        self.input_layernorm = MixedFusedRMSNorm(hidden_size, eps=config.layer_norm_epsilon)
        self.num_heads = config.n_head
        self.layer_idx = layer_idx
        self.self_attention = TelechatAttention(config, layer_idx)
        self.post_attention_layernorm = MixedFusedRMSNorm(hidden_size, eps=config.layer_norm_epsilon)

        self.mlp = TelechatMLP(config)

        self.apply_residual_connection_post_layernorm = config.apply_residual_connection_post_layernorm
        self.hidden_dropout = config.hidden_dropout

    def forward(
            self,
            hidden_states: torch.Tensor,
            attention_mask: torch.Tensor,
            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
            use_cache: bool = False,
            output_attentions: bool = False,
    ):
        layernorm_output = self.input_layernorm(hidden_states)
        if self.apply_residual_connection_post_layernorm:
            residual = layernorm_output
        else:
            residual = hidden_states

        attn_outputs = self.self_attention(
            layernorm_output,
            residual,
            layer_past=layer_past,
            attention_mask=attention_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
        )

        attention_output = attn_outputs[0]
        outputs = attn_outputs[1:]
        layernorm_output = self.post_attention_layernorm(attention_output)

        if self.apply_residual_connection_post_layernorm:
            residual = layernorm_output
        else:
            residual = attention_output
        output = self.mlp(layernorm_output, residual)

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

        return outputs


class TelechatPreTrainedModel(PreTrainedModel):
    config_class = Telechat2Config
    base_model_prefix = "transformer"
    supports_gradient_checkpointing = True
    _no_split_modules = ["TelechatBlock"]
    _skip_keys_device_placement = "past_key_values"

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

    def _init_weights(self, module: nn.Module):
        """Initialize the weights."""
        if isinstance(module, nn.Linear):
            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, LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)

    def _set_gradient_checkpointing(self, module: nn.Module, value: bool = False):
        if isinstance(module, TelechatModel):
            module.gradient_checkpointing = value


class TelechatModel(TelechatPreTrainedModel):
    def __init__(self, config: Telechat2Config):
        super().__init__(config)

        self.embed_dim = config.hidden_size
        self.num_heads = config.n_head
        self.config = config
        self.word_embeddings = nn.Embedding(config.vocab_size, self.embed_dim)
        if self.config.embed_layernorm:
            self.word_embeddings_layernorm = MixedFusedRMSNorm(self.embed_dim, eps=config.layer_norm_epsilon)

        self.h = nn.ModuleList([TelechatBlock(config, _) for _ in range(config.num_hidden_layers)])
        self.ln_f = MixedFusedRMSNorm(self.embed_dim, eps=config.layer_norm_epsilon)
        self.gradient_checkpointing = False
        self.post_init()

    def get_input_embeddings(self):
        return self.word_embeddings

    def _prepare_attn_mask(
            self, attention_mask: torch.Tensor, input_shape: Tuple[int, int], past_key_values_length: int
    ) -> torch.BoolTensor:
        combined_attention_mask = None
        device = attention_mask.device
        _, src_length = input_shape

        if src_length > 1:
            combined_attention_mask = _make_causal_mask(
                input_shape, device=device, past_key_values_length=past_key_values_length
            )
        expanded_attn_mask = _expand_mask(attention_mask, tgt_length=src_length)
        combined_attention_mask = (
            expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask | combined_attention_mask
        )

        return combined_attention_mask

    def set_input_embeddings(self, new_embeddings: torch.Tensor):
        self.word_embeddings = new_embeddings

    def forward(
            self,
            input_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
            attention_mask: Optional[torch.Tensor] = None,
            inputs_embeds: 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,
            **deprecated_arguments,
    ) -> 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:
            batch_size, seq_length = input_ids.shape
        elif inputs_embeds is not None:
            batch_size, seq_length, _ = inputs_embeds.shape

        if past_key_values is None:
            past_key_values = tuple([None] * len(self.h))
        # input_ids = torch.load("Megatron-LM-0624-3B/tensors/input_ids.pt").to(input_ids.device)
        if inputs_embeds is None:
            inputs_embeds = self.word_embeddings(input_ids)
        hidden_states = inputs_embeds
        # print(f"[INFO_Telechat]: inputs_embeds={inputs_embeds}")
        if self.config.embed_layernorm:
            hidden_states = self.word_embeddings_layernorm(inputs_embeds)

        presents = () if use_cache else None
        all_self_attentions = () if output_attentions else None
        all_hidden_states = () if output_hidden_states else None

        if self.gradient_checkpointing and self.training:
            if use_cache:
                use_cache = False

        seq_length_with_past = seq_length
        past_key_values_length = 0
        if past_key_values[0] is not None:
            past_key_values_length = past_key_values[0][0].shape[2]
            seq_length_with_past = seq_length_with_past + past_key_values_length
        if attention_mask is None:
            attention_mask = torch.ones((batch_size, seq_length_with_past), device=hidden_states.device)
        else:
            attention_mask = attention_mask.to(hidden_states.device)
        causal_mask = self._prepare_attn_mask(
            attention_mask,
            input_shape=(batch_size, seq_length),
            past_key_values_length=past_key_values_length,
        )

        # print(f"[INFO_Telechat]: word_embeddings_layernorm={hidden_states}")
        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:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        # None for past_key_value
                        return module(*inputs, use_cache=use_cache, output_attentions=output_attentions)

                    return custom_forward

                outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(block),
                    hidden_states,
                    causal_mask,
                    layer_past,
                )
            else:
                outputs = block(
                    hidden_states,
                    layer_past=layer_past,
                    attention_mask=causal_mask,
                    use_cache=use_cache,
                    output_attentions=output_attentions,
                )

            # print(f"[INFO_Telechat]: outputs{i}={outputs}")
            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)
        # print(f"[INFO_Telechat]: hidden_states={hidden_states}")
        # ref = torch.load("Megatron-LM-0624-3B/tensors/final_layernorm.pt")
        # print(hidden_states.squeeze()[2048:])
        # print(ref.squeeze())
        # print(torch.max(hidden_states.squeeze()[2048:] - ref.squeeze().to(hidden_states.device)))
        # exit()
        # print(ref.shape,hidden_states.shape)
        # print(hidden_states)
        # exit()
        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 BaseModelOutputWithPastAndCrossAttentions(
            last_hidden_state=hidden_states,
            past_key_values=presents,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )


class Telechat2ForCausalLM(TelechatPreTrainedModel):
    # _tied_weights_keys = ["lm_head.weight"]
    _keys_to_ignore_on_load_missing = [r"lm_head.weight"]

    def __init__(self, config: Telechat2Config):
        super().__init__(config)
        self.transformer = TelechatModel(config)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        self.post_init()

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings: torch.Tensor):
        self.lm_head = new_embeddings

    def prepare_inputs_for_generation(
            self,
            input_ids: torch.LongTensor,
            past_key_values: Optional[torch.Tensor] = None,
            attention_mask: Optional[torch.Tensor] = None,
            inputs_embeds: Optional[torch.Tensor] = None,
            **kwargs,
    ) -> dict:
        if past_key_values:
            input_ids = input_ids[:, -1].unsqueeze(-1)
        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"),
                "attention_mask": attention_mask,
            }
        )
        return model_inputs

    def forward(
            self,
            input_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
            attention_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,
            **deprecated_arguments,
    ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:

        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,
            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:
            labels = labels.to(lm_logits.device)
            shift_logits = lm_logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            batch_size, seq_length, vocab_size = shift_logits.shape
            loss_fct = CrossEntropyLoss()
            loss = loss_fct(
                shift_logits.view(batch_size * seq_length, vocab_size), shift_labels.view(batch_size * seq_length)
            )

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

        return CausalLMOutputWithCrossAttentions(
            loss=loss,
            logits=lm_logits,
            past_key_values=transformer_outputs.past_key_values,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )