first commit

README.md

@@ -1,3 +1,7 @@
 ---
 license: apache-2.0
 ---
+# 从零开始构建的大模型
+全过程源码
+[github](https://github.com/fisheeped/creek/tree/main)
+

config.json

@@ -0,0 +1,32 @@
+{
+  "_name_or_path": "/ceph2/yuyang06/note/myllama/output/tf_sft_04061206/model/checkpoint-3315",
+  "architectures": [
+    "creekForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_creek.creekConfig",
+    "AutoModelForCausalLM": "model_creek.creekForCausalLM"
+  },
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 2752,
+  "max_position_embeddings": 2048,
+  "model_type": "creek",
+  "num_attention_heads": 8,
+  "num_hidden_layers": 8,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.39.3",
+  "use_cache": true,
+  "vocab_size": 32000
+}

configuration_creek.py

@@ -0,0 +1,94 @@
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+from transformers import GenerationConfig
+logger = logging.get_logger(__name__)
+
+LLAMA_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+
+
+class creekConfig(PretrainedConfig):
+    model_type = "creek"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=11008,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        hidden_act="silu",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.pretraining_tp = pretraining_tp
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self._rope_scaling_validation()
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    def _rope_scaling_validation(self):
+        """
+        Validate the `rope_scaling` configuration.
+        """
+        if self.rope_scaling is None:
+            return
+
+        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
+            raise ValueError(
+                "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
+                f"got {self.rope_scaling}"
+            )
+        rope_scaling_type = self.rope_scaling.get("type", None)
+        rope_scaling_factor = self.rope_scaling.get("factor", None)
+        if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
+            raise ValueError(
+                f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
+            )
+        if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
+            raise ValueError(f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}")
+
+
+class creekGenerationConfig(GenerationConfig):
+    model = "creek"
+    def __init__(self, **kwargs):
+        kwargs["max_new_tokens"] = kwargs.pop("max_new_tokens", 10)
+        super().__init__(**kwargs)

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "max_length": 1024,
+  "pad_token_id": 2,
+  "transformers_version": "4.39.3"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:753243de8cd80f724d87b240138de310d191646f1514b122c88b4b49b67ddc81
+size 666972384

model_creek.py

@@ -0,0 +1,611 @@
+#!/usr/bin/env python3
+# -*- encoding: utf-8 -*-
+'''
+@File    : model.py
+@Time    : 2024/04/02 10:13:36
+@Author  : YuYang
+@Contact : fisheepman@gmail.com
+@License : Apache License Version 2.0
+@Describe: creek model file
+'''
+
+
+
+from typing import (
+    Optional,
+    Tuple,
+    List,
+    Union
+)
+
+import torch
+from torch import nn
+from torch import Tensor
+import torch.nn.functional as F
+from transformers import PreTrainedModel
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import BaseModelOutputWithPast,CausalLMOutputWithPast
+from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
+# custom 初始化之后需要用相对位置
+from .configuration_creek import creekConfig
+
+import math
+import logging
+import sys
+
+
+logging.basicConfig(
+    format='[%(asctime)s %(pathname)s:%(lineno)s %(levelno)s]\t%(message)s',
+    datefmt='%m/%d/%Y %H:%M:%S',
+    handlers=[logging.StreamHandler(sys.stdout)]
+)
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.INFO)
+
+
+
+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.finfo(dtype).min, 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]`.
+    """
+    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)
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1: # n_rep is num_key_value_groups
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2] # 此次和原始论文推导中不同，正负号不是间隔的，而是分前半块和后半块。但对于结果没有影响
+    x2 = x[..., x.shape[-1] // 2 :] # 后半块
+    return torch.cat((-x2, x1), dim=-1) # 后半块取负，并交换前后块顺序
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
+    # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
+    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] 根据position_ids取出对于的cos值,并扩充维度
+    sin = sin[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim] 根据position_ids取出对于的sin值,并扩充维度
+    q_embed = (q * cos) + (rotate_half(q) * sin) # 对应论文图中RoPE 的简化计算，和原rope论文有区别？, q是 [1,8,5,128] bs,head_num,seq_len,head_dim  这里需要思考，在传入rope前，q行向量的顺序是否有含义
+    k_embed = (k * cos) + (rotate_half(k) * sin) # rotate_half(k) 让后半块取负，且和前板块交换顺序,从而实现 k_0 cos (m * theta) - k_{d/2} sin (m * theta)
+    return q_embed, k_embed
+
+
+
+class RMSNorm(torch.nn.Module):
+    def __init__(self, dim :int, eps: float):
+        super().__init__()
+        self.eps = eps
+        self.weight = nn.Parameter(torch.ones(dim))
+    def forward(self, x: Tensor):
+        x = (x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)).type_as(x) 
+        return x * self.weight
+    
+ALL_LAYERNORM_LAYERS.append(RMSNorm)
+
+class creekRotaryEmbedding(nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()  # refer to https://zhuanlan.zhihu.com/p/632102048 
+        # refer to https://kexue.fm/archives/8265
+        # https://zhuanlan.zhihu.com/p/642884818
+        self.dim = dim # head_dim 128
+        self.max_position_embeddings = max_position_embeddings # 此处 inv_freq 对应公式中的 theta
+        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=device, dtype=self.inv_freq.dtype) # t 表示公式中的 m,表示位置
+        #  freqs [512,64] [max_length,head_dim/2] 行向量表示[m*theta0，m*theta1，m*theta2...] m表示行的值
+        freqs = torch.einsum("i,j->ij", t, self.inv_freq) # 此处 freqs 对应公式中的 m * theta, t 对应公式中的 m，表示位置
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1) # 此处和原始公式不同，theta_0 和 theta_0 不再相邻. 是按块交叉，不是按元素交叉,emd [512,128] [max_length,head_dim] 行向量表示[m*theta0，m*theta1，m*theta2...]再拼接一次
+        self.register_buffer("cos_cached", emb.cos()[None, None, :, :].to(dtype), persistent=False)
+        self.register_buffer("sin_cached", emb.sin()[None, None, :, :].to(dtype), persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)
+
+        return (
+            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype), # 取出序列长度的位置信息
+            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype), 
+        )
+
+class creekPreTrainedModel(PreTrainedModel):
+    config_class = creekConfig
+    base_model_prefix = "transformer"
+    _skip_keys_device_placement = "past_key_values"
+    _no_split_modules = ["creekDecoderLayer"] # for accelerate to auto split limit
+    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_()
+
+class creekMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+class creekAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: creekConfig):
+        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.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.rope_theta = config.rope_theta
+        self.sqrt_head_dim = math.sqrt(self.head_dim)
+        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.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=config.attention_bias)
+        self._init_rope()
+
+    def _init_rope(self):
+        self.rotary_emb = creekRotaryEmbedding(
+            self.head_dim,
+            max_position_embeddings=self.max_position_embeddings,
+            base=self.rope_theta,
+        )
+        
+    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,
+        padding_mask: Optional[torch.LongTensor] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size() # [1,5,1024] 第二轮 [1,1,1024]
+    
+        query_states = self.q_proj(hidden_states) # [1,5,1024] 第二轮 [1,1,1024]
+        key_states = self.k_proj(hidden_states) # [1,5,1024] 第二轮 [1,1,1024]
+        value_states = self.v_proj(hidden_states) # [1,5,1024] 第二轮 [1,1,1024]
+        # 第二轮下面全是[1,8,1,128]
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) # [1,8,5,128]
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) # [1,8,5,128] 
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) # [1,8,5,128]
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2] # 第二轮的seq_len+第一轮的seq_len
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len) # 根据seq_len 取出cos,sin 
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids) #  让后半块取负，且和前板块交换顺序,从而实现前半块 k_i cos (m * theta) - k_{d/2 + i |i - d/2} sin (m * theta) ，其中i 表示行向量的位置 [i - 128] 
+        
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2) # [1,8,6,128]
+            value_states = torch.cat([past_key_value[1], value_states], dim=2) # [1,8,6,128]
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups) # 如果num_key_value_groups是1，则不变，否则扩充至num_heads
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+        # [1,8,5,128] matmul [1,8,128,5] -> [1,8,5,5]  # 第二轮是 [1,8,1,128] matmul [1,8,128,6] -> [1,8,1,6]
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / self.sqrt_head_dim
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None: #[1,1,5,5] 上三角为负无穷，下三角和对角线为0 第二轮 [1,1,1,6] 全0
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+
+        # upcast attention to fp32 防止数据类型溢出很有效(因为softmax分母是指数和)
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2).contiguous() # [1,5,8,128] 第二轮 [1,1,8,128]
+
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size) # [1,5,1024] 第二轮 [1,1,1024]
+
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+        # past_key_value is past key and value
+        return attn_output, attn_weights, past_key_value
+    
+class creekDecoderLayer(nn.Module):
+    def __init__(self, config: creekConfig):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.mlp = creekMLP(config)
+        self.self_attn = creekAttention(config=config)
+        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,
+        padding_mask: Optional[torch.LongTensor] = None,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        
+        hidden_states = self.input_layernorm(hidden_states)
+        residual = hidden_states
+        # Self Attention .这里present_key_value is tuple(key,value)
+        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,
+            padding_mask=padding_mask,
+        )
+        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 creekModel(creekPreTrainedModel):
+    def __init__(self, config: creekConfig):
+        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([creekDecoderLayer(config) for _ in range(config.num_hidden_layers)])
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        # 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
+    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: # seq_len > 1 ，如果是new_tokens_seq，则直接清0
+            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
+            ) # all is 0
+            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,  # torch.Size([1, 5])
+        attention_mask: Optional[torch.Tensor] = None, # torch.Size([1, 5]) 11111
+        position_ids: Optional[torch.LongTensor] = None, # torch.Size([1, 5]) 01234
+        past_key_values: Optional[List[torch.FloatTensor]] = None, # None
+        inputs_embeds: Optional[torch.FloatTensor] = None, # None
+        use_cache: Optional[bool] = None, # true
+        output_attentions: Optional[bool] = None, # false
+        output_hidden_states: Optional[bool] = None, # false
+        return_dict: Optional[bool] = None, # false
+    ) -> Union[Tuple, BaseModelOutputWithPast]: # mini-llama hidden_size 1024
+        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 input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape  # 1,5
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        seq_length_with_past = seq_length # 5
+        past_key_values_length = 0
+
+        if past_key_values is not None: # 第二轮的时候运行
+            past_key_values_length = past_key_values[0][0].shape[2] # 取出之前的seq_len
+            seq_length_with_past = seq_length_with_past + past_key_values_length # new_token_length + old_input_seq_len
+
+        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() # [1,5] from int64 to int64 # 第二轮为[1,1]
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)  # inputs_embeds:torch.Size([1, 5, 1024])  embed_tokens:[vocab_size,hidden_size]
+        # embed positions
+        if attention_mask is None:
+            attention_mask = torch.ones(
+                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
+            )
+            padding_mask = None
+        else:
+            if 0 in attention_mask:
+                padding_mask = attention_mask # tensor([[1, 1, 1, 1, 1]], device='cuda:0')
+            else:
+                padding_mask = None
+
+        attention_mask = self._prepare_decoder_attention_mask(
+            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
+        ) # [1,1,5,5]上三角负无穷，下三角和对角线为0 第二轮 [1,1,1,6] 全是0
+
+        hidden_states = inputs_embeds
+
+        # 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):  # 关键结构
+            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 # 直接取出对应头的kv
+
+            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,
+                padding_mask=padding_mask,
+            ) # attn + mlp
+
+            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 creekForCausalLM(creekPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = creekModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # 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
+    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)
+        logits = logits.float()
+
+        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 = nn.CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        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.to(past_state.device)) for past_state in layer_past),
+            )
+        return reordered_past

special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,47 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "auto_map": {
+    "AutoTokenizer": [
+      null,
+      "tokenizer_creek.CreekTokenizerFast"
+    ]
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "</s>",
+  "padding_side": "right",
+  "reversion": "main",
+  "tokenizer_class": "CreekTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

tokenizer_creek.py

@@ -0,0 +1,152 @@
+from transformers import PreTrainedTokenizerFast
+from transformers.utils.versions import require_version
+
+from tokenizers import processors
+
+from typing import Dict,Any
+
+import logging
+import sys
+
+require_version("tokenizers>=0.13.3")
+
+
+logging.basicConfig(
+    format='[%(asctime)s %(pathname)s:%(lineno)s %(levelno)s]\t%(message)s',
+    datefmt='%m/%d/%Y %H:%M:%S',
+    handlers=[logging.StreamHandler(sys.stdout)]
+)
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.INFO)
+
+
+VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.json"}
+
+DEFAULT_SYSTEM_PROMPT = """你是一个助理。"""
+
+
+class CreekTokenizerFast(PreTrainedTokenizerFast):
+    slow_tokenizer_class = None
+    vocab_files_names = VOCAB_FILES_NAMES
+    padding_side = 'left'
+    model_input_names = ['input_ids',"attention_mask"]
+    def __init__(
+        self,
+        vocab_file=None,
+        tokenizer_file=None,
+        clean_up_tokenization_spaces=False,
+        unk_token="<unk>",
+        bos_token="<s>",
+        eos_token="</s>",
+        add_bos_token=True,
+        add_eos_token=False,
+        use_default_system_prompt=False,
+        **kwargs,
+    ):
+        super().__init__(
+                vocab_file=vocab_file,
+                tokenizer_file=tokenizer_file,
+                clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+                unk_token=unk_token,
+                bos_token=bos_token,
+                eos_token=eos_token,
+                add_bos_token=add_bos_token,
+                add_eos_token=add_eos_token,
+                use_default_system_prompt=use_default_system_prompt,
+                **kwargs,
+            )
+        self._add_bos_token = add_bos_token
+        self._add_eos_token = add_eos_token
+        self.update_post_processor()
+        self.use_default_system_prompt = use_default_system_prompt
+        self.vocab_file = vocab_file
+
+    def update_post_processor(self):
+        """
+        Updates the underlying post processor with the current `bos_token` and `eos_token`.
+        """
+        bos = self.bos_token
+        bos_token_id = self.bos_token_id
+        if bos is None and self.add_bos_token:
+            raise ValueError("add_bos_token = True but bos_token = None")
+
+        eos = self.eos_token
+        eos_token_id = self.eos_token_id
+        if eos is None and self.add_eos_token:
+            raise ValueError("add_eos_token = True but eos_token = None")
+
+        single = f"{(bos+':0 ') if self.add_bos_token else ''}$A:0{(' '+eos+':0') if self.add_eos_token else ''}"
+        pair = f"{single}{(' '+bos+':1') if self.add_bos_token else ''} $B:1{(' '+eos+':1') if self.add_eos_token else ''}"
+
+        special_tokens = []
+        if self.add_bos_token:
+            special_tokens.append((bos, bos_token_id))
+        if self.add_eos_token:
+            special_tokens.append((eos, eos_token_id))
+        self._tokenizer.post_processor = processors.TemplateProcessing(
+            single=single, pair=pair, special_tokens=special_tokens
+        )
+    @property
+    def add_eos_token(self):
+        return self._add_eos_token
+
+    @property
+    def add_bos_token(self):
+        return self._add_bos_token
+
+    @add_eos_token.setter
+    def add_eos_token(self, value):
+        self._add_eos_token = value
+        self.update_post_processor()
+
+    @add_bos_token.setter
+    def add_bos_token(self, value):
+        self._add_bos_token = value
+        self.update_post_processor()
+    @property
+    def default_chat_template(self):
+        template = (
+            "{% if messages[0]['role']  == 'system' %}"
+            "{% set loop_messages = messages[1:] %}"  # Extract system message if it's present
+            "{% set system_message = messages[0]['content'] %}"
+            "{% elif USE_DEFAULT_PROMPT == true and not '<`q>' in messages[0]['content'] %}"
+            "{% set loop_messages = messages %}"
+            "{% set system_message = 'DEFAULT_SYSTEM_MESSAGE' %}"
+            "{% else %}"
+            "{% set loop_messages = messages %}"
+            "{% set system_message = false %}"
+            "{% endif %}"
+            "{% for message in loop_messages %}"  # Loop over all 
+             "{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}"
+            "{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}"
+            "{% endif %}"
+            "{% if loop.index0 == 0 and system_message != false %}"
+            "{% set content = system_message + ' <`q> ' + message['content'] %}"
+            "{% else %}"
+            "{% set content = message['content'] %}"
+            "{% endif %}"
+            "{% if message['role'] == 'user' %}" 
+            "{{ bos_token + ' ' + content + ' <`a>' }}"
+            "{% elif message['role'] == 'system' %}"
+            "{{ content + ' <`q>' }}"
+            "{% elif message['role'] == 'assistant' %}"
+            "{{ ' '  + content + ' ' + eos_token }}"
+            "{% endif %}"
+            "{% endfor %}"
+        )
+        template = template.replace("USE_DEFAULT_PROMPT", "true" if self.use_default_system_prompt else "false")
+        default_message = DEFAULT_SYSTEM_PROMPT.replace("\n", "\\n").replace("'", "\\'")
+        template = template.replace("DEFAULT_SYSTEM_MESSAGE", default_message)
+
+        return template
+    
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = bos_token_id + token_ids_0 + eos_token_id
+
+        if token_ids_1 is not None:
+            output = output + bos_token_id + token_ids_1 + eos_token_id
+
+        return output