pose_modeling_baichuan.py

# Modification Copyright 2023 Dawei Zhu
# Copyright 2023 Baichuan Inc. All Rights Reserved.

# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# 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.


from my_configuration_baichuan import BaichuanConfig
# from .generation_utils import build_chat_input, TextIterStreamer

import math
from typing import List, Optional, Tuple, Union
from threading import Thread

import numpy as np
import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
from torch.nn import functional as F
from transformers import PreTrainedModel, PretrainedConfig
from transformers.activations import ACT2FN
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
from transformers.generation.utils import GenerationConfig
from transformers.utils import logging, ContextManagers

import os
from contextlib import contextmanager
logger = logging.get_logger(__name__)

try:
    from xformers import ops as xops
except ImportError:
    xops = None
    logger.warning(
        "Xformers is not installed correctly. If you want to use memory_efficient_attention to accelerate training use the following command to install Xformers\npip install xformers."
    )


# Copied from transformers.models.bart.modeling_bart._make_causal_mask
def _make_causal_mask(
        input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
):
    """
    Make causal mask used for bi-directional self-attention.
    """
    bsz, tgt_len = input_ids_shape
    mask = torch.full((tgt_len, tgt_len), torch.tensor(torch.finfo(dtype).min, device=device), device=device)
    mask_cond = torch.arange(mask.size(-1), device=device)
    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
    mask = mask.to(dtype)

    if past_key_values_length > 0:
        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)

def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    if len(mask.size()) == 3:
        bsz, src_len, _ = mask.size()
        tgt_len = tgt_len if tgt_len is not None else src_len
        expanded_mask = mask[:,None,:,:].expand(bsz, 1, tgt_len, src_len).to(dtype)
    else:
        bsz, src_len = mask.size()
        tgt_len = tgt_len if tgt_len is not None else src_len
        expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)


class RMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        RMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

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

        # convert into half-precision if necessary
        if self.weight.dtype in [torch.float16, torch.bfloat16]:
            hidden_states = hidden_states.to(self.weight.dtype)

        return self.weight * hidden_states


class RotaryEmbedding(torch.nn.Module):
    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
        super().__init__()

        self.dim = dim
        self.max_position_embeddings = max_position_embeddings
        self.base = base
        inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)

        # Build here to make `torch.jit.trace` work.
        self._set_cos_sin_cache(
            seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype()
        )

    def _set_cos_sin_cache(self, seq_len, device, dtype):
        self.max_seq_len_cached = seq_len
        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        # Different from paper, but it uses a different permutation in order to obtain the same calculation
        emb = torch.cat((freqs, freqs), dim=-1)
        self.cos_cached = emb.cos()[None, None, :, :].to(torch.float32)
        self.sin_cached = emb.sin()[None, None, :, :].to(torch.float32)


    def forward(self, x, seq_len=None):
        # x: [bs, num_attention_heads, seq_len, head_size]
        # This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
        if seq_len > self.max_seq_len_cached:
            self.max_seq_len_cached = seq_len
            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)

        elif self.cos_cached.device != x.device:
            self.cos_cached = self.cos_cached.to(x.device)
            self.sin_cached = self.sin_cached.to(x.device)  
        return (
            self.cos_cached[:, :, :, ...],
            self.sin_cached[:, :, :, ...],
        )

class LinearScalingRotaryEmbedding(RotaryEmbedding):

    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None, scaling_factor=1.0):
        self.scaling_factor = scaling_factor
        super().__init__(dim, max_position_embeddings, base, device)

    def _set_cos_sin_cache(self, seq_len, device, dtype):
        self.max_seq_len_cached = seq_len
        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
        t = t / self.scaling_factor
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        # Different from paper, but it uses a different permutation in order to obtain the same calculation
        emb = torch.cat((freqs, freqs), dim=-1)
        self.cos_cached = emb.cos()[None, None, :, :].to(torch.float32)
        self.sin_cached = emb.sin()[None, None, :, :].to(torch.float32)


class VanillaNTKScalingRotaryEmbedding(RotaryEmbedding):

    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None, scaling_factor=1.0):
        self.scaling_factor = scaling_factor
        super().__init__(dim, max_position_embeddings, base, device)

    def _set_cos_sin_cache(self, seq_len, device, dtype):
        self.max_seq_len_cached = seq_len

        base = self.base * self.scaling_factor ** (self.dim / (self.dim - 2))
        inv_freq = 1.0 / (base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)

        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        # Different from paper, but it uses a different permutation in order to obtain the same calculation
        emb = torch.cat((freqs, freqs), dim=-1)
        self.cos_cached = emb.cos()[None, None, :, :].to(torch.float32)
        self.sin_cached = emb.sin()[None, None, :, :].to(torch.float32)

def _yarn_find_correction_dim(num_rotations, dim, base=10000, max_position_embeddings=2048):
    return (dim * math.log(max_position_embeddings/(num_rotations * 2 * math.pi)))/(2 * math.log(base))

# Find dim range bounds based on rotations
def _yarn_find_correction_range(low_rot, high_rot, dim, base=10000, max_position_embeddings=2048):
    low = math.floor(_yarn_find_correction_dim(
        low_rot, dim, base, max_position_embeddings))
    high = math.ceil(_yarn_find_correction_dim(
        high_rot, dim, base, max_position_embeddings))
    return max(low, 0), min(high, dim-1)  # Clamp values just in case

def _yarn_linear_ramp_mask(min, max, dim):
    if min == max:
        max += 0.001  # Prevent singularity

    linear_func = (torch.arange(dim, dtype=torch.float32) - min) / (max - min)
    ramp_func = torch.clamp(linear_func, 0, 1)
    return ramp_func

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


class YaRNScaledRotaryEmbedding(torch.nn.Module):
    def __init__(self, dim, max_position_embeddings=2048, base=10000, scale=1, original_max_position_embeddings=2048, extrapolation_factor=1, attn_factor=1, beta_fast=32, beta_slow=1, finetuned=False, device=None):
        super().__init__()

        self.dim = dim
        self.max_position_embeddings = max_position_embeddings
        self.base = base
        self.scale = scale
        self.original_max_position_embeddings = original_max_position_embeddings
        self.extrapolation_factor = extrapolation_factor
        self.attn_factor = attn_factor
        self.beta_fast = beta_fast
        self.beta_slow = beta_slow

        # self.yarn(device)
        self.revised_yarn(device)

        # Build here to make `torch.jit.trace` work.
        self.max_seq_len_cached = max_position_embeddings

        t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=torch.float32)
        freqs = torch.outer(t, self.inv_freq.to(device=t.device).to(t.dtype))
        # t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.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)
        dtype = torch.get_default_dtype()

        self.register_buffer("cos_cached", (emb.cos() * self.mscale)[None, None, :, :].to(dtype), persistent=False)
        self.register_buffer("sin_cached", (emb.sin() * self.mscale)[None, None, :, :].to(dtype), persistent=False)

    def forward(self, x, seq_len=None):
        # x: [bs, num_attention_heads, seq_len, head_size]
        # This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
        if seq_len > self.max_seq_len_cached:
            print("*****notice******")
            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)

            self.register_buffer("cos_cached", (emb.cos() * self.mscale)[None, None, :, :].to(x.dtype), persistent=False)
            self.register_buffer("sin_cached", (emb.sin() * self.mscale)[None, None, :, :].to(x.dtype), persistent=False)
        return (
            self.cos_cached[:, :, :, ...].to(dtype=x.dtype),
            self.sin_cached[:, :, :, ...].to(dtype=x.dtype),
        )

    def yarn(self, device):
        pos_freqs = self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim)
        inv_freq_extrapolation = 1.0 / pos_freqs
        inv_freq_interpolation = 1.0 / (self.scale * pos_freqs)

        low, high = _yarn_find_correction_range(self.beta_fast, self.beta_slow, self.dim, self.base, self.original_max_position_embeddings)
        inv_freq_mask = (1 - _yarn_linear_ramp_mask(low, high, self.dim // 2).float().to(device)) * self.extrapolation_factor # Get n-d rotational scaling corrected for extrapolation
        inv_freq = inv_freq_interpolation * (1 - inv_freq_mask) + inv_freq_extrapolation * inv_freq_mask

        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.mscale = float(_yarn_get_mscale(self.scale) * self.attn_factor) # Get n-d magnitude scaling corrected for interpolation

    def revised_yarn(self, device):
        inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))

        low, high = _yarn_find_correction_range(self.beta_fast, self.beta_slow, self.dim, self.base, self.original_max_position_embeddings)
        inv_freq_mask = (1 - _yarn_linear_ramp_mask(low, high, self.dim // 2).float().to(device)) * self.extrapolation_factor

        inv_freq = inv_freq / ((1-inv_freq_mask)*self.scale + inv_freq_mask)
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.mscale = float(_yarn_get_mscale(self.scale) * self.attn_factor)


def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2:]
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(q, k, cos_, sin_, position_ids):
    cos = cos_.squeeze(1).squeeze(0)  # [seq_len, dim]
    sin = sin_.squeeze(1).squeeze(0)  # [seq_len, dim]
    cos = cos[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    sin = sin[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    q_embed = (q.float() * cos) + (rotate_half(q.float()) * sin)
    k_embed = (k.float() * cos) + (rotate_half(k.float()) * sin)
    return q_embed.to(q.dtype), k_embed.to(k.dtype)


class MLP(nn.Module):
    def __init__(
            self,
            hidden_size: int,
            intermediate_size: int,
            hidden_act: str,
    ):
        super().__init__()
        self.gate_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
        self.down_proj = nn.Linear(intermediate_size, hidden_size, bias=False)
        self.up_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
        self.act_fn = ACT2FN[hidden_act]

    def forward(self, x):
        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))


class Attention(nn.Module):
    """Multi-headed attention from 'Attention Is All You Need' paper"""
    def __init__(self, config: BaichuanConfig):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.hidden_size // self.num_heads
        self.max_position_embeddings = config.max_position_embeddings

        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}"
                f" and `num_heads`: {self.num_heads})."
            )
        self.W_pack = nn.Linear(self.hidden_size, 3 * self.hidden_size, bias=False)
        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
        self._init_rope()
        # self.rotary_emb = RotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)
        
    def _init_rope(self):
        if self.config.rope_scaling is None:
            self.rotary_emb = RotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)
        else:
            scaling_type = self.config.rope_scaling["type"]
            scaling_factor = self.config.rope_scaling["factor"]
            if scaling_type == "linear":
                self.rotary_emb = LinearScalingRotaryEmbedding(
                    self.head_dim, max_position_embeddings=self.max_position_embeddings, scaling_factor=scaling_factor
                )
            elif scaling_type == "vanilla_ntk":
                self.rotary_emb = VanillaNTKScalingRotaryEmbedding(
                    self.head_dim, max_position_embeddings=self.max_position_embeddings, scaling_factor=scaling_factor
                )
            elif scaling_type == "yarn":
                original_max_position_embeddings = self.config.rope_scaling["original_max_position_embeddings"]
                self.rotary_emb = YaRNScaledRotaryEmbedding(
                    self.head_dim, max_position_embeddings=self.max_position_embeddings, scale=scaling_factor, original_max_position_embeddings=original_max_position_embeddings
                )
            else:
                raise ValueError(f"Unknown RoPE scaling type {scaling_type}")

    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()

    def forward(
            self,
            hidden_states: torch.Tensor,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_value: Optional[Tuple[torch.Tensor]] = None,
            output_attentions: bool = False,
            use_cache: bool = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
        bsz, q_len, _ = hidden_states.size()

        proj = self.W_pack(hidden_states)
        proj = proj.unflatten(-1, (3, self.hidden_size)).unsqueeze(0).transpose(0, -2).squeeze(-2)
        query_states = proj[0].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = proj[1].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        value_states = proj[2].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)

        kv_seq_len = key_states.shape[-2]
        if past_key_value is not None:
            kv_seq_len += past_key_value[0].shape[-2]
        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
        # [bsz, nh, t, hd]

        if past_key_value is not None:
            # reuse k, v, self_attention
            key_states = torch.cat([past_key_value[0], key_states], dim=2)
            value_states = torch.cat([past_key_value[1], value_states], dim=2)

        past_key_value = (key_states, value_states) if use_cache else None
        if xops is not None and self.training:
            attn_weights = None
            query_states = query_states.transpose(1, 2)
            key_states = key_states.transpose(1, 2)
            value_states = value_states.transpose(1, 2)
            attn_output = xops.memory_efficient_attention(
                query_states, key_states, value_states, attn_bias=xops.LowerTriangularMask()
            )
        else:
            with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=True):
                attn_output = F.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask = attention_mask)
            attn_output = attn_output.transpose(1, 2)
        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
        attn_output = self.o_proj(attn_output)

        if not output_attentions:
            attn_weights = None

        return attn_output, attn_weights, past_key_value


class DecoderLayer(nn.Module):
    def __init__(self, config: BaichuanConfig):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.self_attn = Attention(config=config)
        self.mlp = MLP(
            hidden_size=self.hidden_size,
            intermediate_size=config.intermediate_size,
            hidden_act=config.hidden_act,
        )
        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(
            self,
            hidden_states: torch.Tensor,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_value: Optional[Tuple[torch.Tensor]] = None,
            output_attentions: Optional[bool] = False,
            use_cache: Optional[bool] = False,
    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:

        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states, self_attn_weights, present_key_value = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs


class BaichuanPreTrainedModel(PreTrainedModel):
    config_class = BaichuanConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["DecoderLayer"]
    _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]

    def _init_weights(self, module):
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()

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


class BaichuanModel(BaichuanPreTrainedModel):
    def __init__(self, config: BaichuanConfig):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
        self.layers = nn.ModuleList([DecoderLayer(config) for _ in range(config.num_hidden_layers)])
        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

        self.gradient_checkpointing = False
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        self.embed_tokens = value

    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
        # create causal mask
        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
        combined_attention_mask = None
        if input_shape[-1] > 1:
            combined_attention_mask = _make_causal_mask(
                input_shape,
                inputs_embeds.dtype,
                device=inputs_embeds.device,
                past_key_values_length=past_key_values_length,
            )

        if attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
                inputs_embeds.device
            )
            combined_attention_mask = (
                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
            )

        return combined_attention_mask

    def forward(
            self,
            input_ids: torch.LongTensor = None,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[List[torch.FloatTensor]] = None,
            inputs_embeds: Optional[torch.FloatTensor] = None,
            use_cache: Optional[bool] = None,
            output_attentions: Optional[bool] = None,
            output_hidden_states: Optional[bool] = None,
            return_dict: Optional[bool] = None,
    ) -> Union[Tuple, BaseModelOutputWithPast]:
        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

        # retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
        elif 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
        else:
            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")

        seq_length_with_past = seq_length
        past_key_values_length = 0

        if past_key_values 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 position_ids is None:
            device = input_ids.device if input_ids is not None else inputs_embeds.device
            position_ids = torch.arange(
                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
            )
            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
        else:
            position_ids = position_ids.view(-1, seq_length).long()

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)
        # embed positions
        if attention_mask is None:
            attention_mask = torch.ones(
                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
            )
        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
        )

        hidden_states = inputs_embeds

        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

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None
        next_decoder_cache = () if use_cache else None

        for idx, decoder_layer in enumerate(self.layers):
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            past_key_value = past_key_values[idx] if past_key_values is not None else None

            if self.gradient_checkpointing and self.training:

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

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),
                    hidden_states,
                    attention_mask,
                    position_ids,
                    None,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    position_ids=position_ids,
                    past_key_value=past_key_value,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                )

            hidden_states = layer_outputs[0]

            if use_cache:
                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)

            if output_attentions:
                all_self_attns += (layer_outputs[1],)

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        next_cache = next_decoder_cache if use_cache else None
        if not return_dict:
            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=next_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
        )


class NormHead(nn.Module):
    def __init__(self, hidden_size, vocab_size, bias=False):
        super().__init__()
        self.weight = nn.Parameter(torch.empty((vocab_size, hidden_size)))
        nn.init.kaiming_uniform_(self.weight, a=math.sqrt(5))
        self.first_flag = True

    def forward(self, hidden_states):
        if self.training:
            norm_weight = nn.functional.normalize(self.weight)
            self.first_flag = True
        elif self.first_flag:
            self.first_flag = False
            self.weight = nn.Parameter(nn.functional.normalize(self.weight))
            norm_weight = self.weight
        else:
            norm_weight = self.weight
        return nn.functional.linear(hidden_states, norm_weight)

_init_weights = True
@contextmanager
def no_init_weights(_enable=True):
    global _init_weights
    old_init_weights = _init_weights
    if _enable:
        _init_weights = False
    try:
        yield
    finally:
        _init_weights = old_init_weights

class BaichuanForCausalLM(BaichuanPreTrainedModel):
    def __init__(self, config, *model_args, **model_kwargs):
        super().__init__(config, *model_args, **model_kwargs)
        self.model = BaichuanModel(config)

        self.lm_head = NormHead(config.hidden_size, config.vocab_size, bias=False)
        if hasattr(config, "quantization_config") and isinstance(config.quantization_config, dict) and config.quantization_config.get('load_in_4bit', False):
            try:
                from .quantizer import quantize_offline, init_model_weight_int4
            except ImportError:
                raise ImportError(f"Needs QLinear to run quantize.")
            quantize_offline(self, 4)
        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        self.model.embed_tokens = value

    def get_output_embeddings(self):
        return self.lm_head

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

    def set_decoder(self, decoder):
        self.model = decoder

    def get_decoder(self):
        return self.model
    
    @classmethod
    def from_pretrained(
        cls,
        pretrained_model_name_or_path: Optional[Union[str, os.PathLike]],
        *model_args,
        config: Optional[Union[PretrainedConfig, str, os.PathLike]] = None,
        cache_dir: Optional[Union[str, os.PathLike]] = None,
        ignore_mismatched_sizes: bool = False,
        force_download: bool = False,
        local_files_only: bool = False,
        token: Optional[Union[str, bool]] = None,
        revision: str = "main",
        use_safetensors: bool = None,
        **kwargs,
    ):
        # Load config if we don't provide a configuration
        if not isinstance(config, PretrainedConfig):
            config_path = config if config is not None else pretrained_model_name_or_path
            config, model_kwargs = cls.config_class.from_pretrained(
                config_path,
                cache_dir=cache_dir,
                return_unused_kwargs=True,
                force_download=force_download,
                resume_download=False,
                proxies=None,
                local_files_only=local_files_only,
                token=token,
                revision=revision,
                subfolder="",
                _from_auto=False,
                _from_pipeline=None,
                **kwargs,
            )
        else:
            model_kwargs = kwargs
        
        if hasattr(config, "quantization_config") and config.quantization_config['load_in_4bit']:
            try:
                from .quantizer import init_model_weight_int4
                from accelerate import init_empty_weights, dispatch_model, infer_auto_device_map
                from accelerate.utils import CustomDtype
                from accelerate.utils import get_balanced_memory
            except ImportError:
                raise ImportError(f"Needs import model weight init func to run quantize.") 
            # Instantiate model.
            init_contexts = [no_init_weights(_enable=True)]
            init_contexts.append(init_empty_weights())
            with ContextManagers(init_contexts):
                model = cls(config)
            
            model_file = os.path.join(pretrained_model_name_or_path, 'pytorch_model.bin')
            state_dict = torch.load(model_file, map_location="cpu") 
            model.is_quantized = True
            
            device_map = kwargs.pop("device_map", None)
            torch_dtype = kwargs.pop("torch_dtype", None)
            
            if device_map is not None:
                kwargs = {"no_split_module_classes": model._no_split_modules}
                target_dtype = CustomDtype.INT4
                max_memory = get_balanced_memory(
                    model,
                    dtype=target_dtype,
                    low_zero=(device_map == "balanced_low_0"),
                    max_memory=None,
                    **kwargs,
                )
                kwargs["max_memory"] = max_memory
                device_map = infer_auto_device_map(model, dtype=target_dtype, **kwargs)
                
            model = init_model_weight_int4(config, model, state_dict)
            
            # Set model in evaluation mode to deactivate DropOut modules by default
            model.eval()
            # If it is a model with generation capabilities, attempt to load the generation config
            if model.can_generate():
                try:
                    model.generation_config = GenerationConfig.from_pretrained(
                        pretrained_model_name_or_path,
                        cache_dir=cache_dir,
                        force_download=force_download,
                        resume_download=False,
                        proxies=None,
                        local_files_only=local_files_only,
                        token=token,
                        revision=revision,
                        subfolder="",
                        _from_auto=False,
                        _from_pipeline=None,
                        **kwargs,
                    )
                except (OSError, TypeError):
                    logger.info(
                        "Generation config file not found, using a generation config created from the model config."
                    )
                    pass
            
            if device_map is not None:
                dispatch_model(model, device_map=device_map)
            
            return model
        return super(BaichuanForCausalLM, cls).from_pretrained(pretrained_model_name_or_path, *model_args, 
                config=config, cache_dir=cache_dir, ignore_mismatched_sizes=ignore_mismatched_sizes, 
                force_download=force_download, local_files_only=local_files_only, token=token, revision=revision, 
                use_safetensors=use_safetensors, **kwargs)   

    def forward(
            self,
            input_ids: torch.LongTensor = None,
            attention_mask: Optional[torch.Tensor] = None,
            position_ids: Optional[torch.LongTensor] = None,
            past_key_values: Optional[List[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, CausalLMOutputWithPast]:

        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        hidden_states = outputs[0]
        logits = self.lm_head(hidden_states)
        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = CrossEntropyLoss()
            shift_logits = shift_logits.view(-1, self.config.vocab_size)
            shift_labels = shift_labels.view(-1)
            softmax_normalizer = shift_logits.max(-1).values ** 2
            z_loss = self.config.z_loss_weight * softmax_normalizer.mean()
            # Enable model parallelism
            shift_labels = shift_labels.to(shift_logits.device)
            loss = loss_fct(shift_logits, shift_labels) + z_loss

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

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

    def prepare_inputs_for_generation(
            self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
    ):
        if past_key_values:
            input_ids = input_ids[:, -1:]

        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)

        # 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(
            {
                "position_ids": position_ids,
                "past_key_values": past_key_values,
                "use_cache": kwargs.get("use_cache"),
                "attention_mask": attention_mask,
            }
        )
        return model_inputs

    @staticmethod
    def _reorder_cache(past_key_values, beam_idx):
        reordered_past = ()
        for layer_past in past_key_values:
            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
        return reordered_past

    def quantize(self, bits: int):
        try:
            from .quantizer import quantize_online
        except ImportError:
            raise ImportError(f"Needs QLinear to run quantize.")
        return quantize_online(self, bits)

    def chat(self, tokenizer, messages: List[dict], stream=False,
             generation_config: Optional[GenerationConfig]=None):
        generation_config = generation_config or self.generation_config
        input_ids = build_chat_input(self, tokenizer, messages, generation_config.max_new_tokens)
        if stream:
            streamer = TextIterStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
            Thread(target=self.generate, kwargs=dict(
                inputs=input_ids, streamer=streamer,
                generation_config=generation_config,
            )).start()
            return streamer
        else:
            outputs = self.generate(input_ids, generation_config=generation_config)
            response = tokenizer.decode(outputs[0][len(input_ids[0]):], skip_special_tokens=True)
            return response