BatGPT-15B-sirius / modeling_batgpt.py
Charlie
Update Readme
af2f908
# This code serves as a port of the models described in BatGPT.
# It is based on the bloom codebase, which provides the initial framework for our model implementation.
# To understand how to use these models, please refer to the documentation and usage instructions provided in the bloom models repository.
# Additionally, we draw inspiration from the ChatGLM and Baichuan codebase, which includes implementations for prefix encoder, chat, and stream_chat functionalities. These components are utilized in our ported models.
# Feel free to explore the ChatGLM and Baichuan codebase for further insights on how these components can be utilized effectively.
import math
import warnings
from typing import Optional, Tuple, Union, List, Callable, Dict, Any
import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, LayerNorm, MSELoss
from torch.nn import functional as F
from torch.nn.utils import skip_init
import copy
import re
import sys
from transformers.modeling_outputs import (
BaseModelOutputWithPast,
CausalLMOutputWithPast,
)
from transformers.modeling_utils import PreTrainedModel
from transformers.utils import logging
from transformers.generation.logits_process import LogitsProcessor
from transformers.generation.utils import LogitsProcessorList, StoppingCriteriaList, GenerationConfig, ModelOutput
from .configuration_batgpt import BatGPTConfig
logger = logging.get_logger(__name__)
# flags required to enable jit fusion kernels
if sys.platform != 'darwin':
torch._C._jit_set_profiling_mode(False)
torch._C._jit_set_profiling_executor(False)
torch._C._jit_override_can_fuse_on_cpu(True)
torch._C._jit_override_can_fuse_on_gpu(True)
# For faster llm model initilization
def module_init(cls, empty_init, *args, **kwargs):
if empty_init:
return skip_init(cls, *args, **kwargs)
else:
return cls(*args, **kwargs)
class InvalidScoreLogitsProcessor(LogitsProcessor):
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
if torch.isnan(scores).any() or torch.isinf(scores).any():
scores.zero_()
scores[..., 5] = 5e4
return scores
class PrefixEncoder(torch.nn.Module):
"""
The torch.nn model to encode the prefix
Input shape: (batch-size, prefix-length)
Output shape: (batch-size, prefix-length, 2*layers*hidden)
"""
def __init__(self, config: BatGPTConfig):
super().__init__()
self.prefix_proj = config.prefix_proj
self.head_dim = config.hidden_size // config.n_head
if self.prefix_proj:
# Use a two-layer MLP to encode the prefix
kv_size = config.n_layer * self.head_dim * config.num_heads_per_kv * 2
self.embedding = torch.nn.Embedding(config.prefix_size, kv_size)
self.trans = torch.nn.Sequential(
torch.nn.Linear(kv_size, config.hidden_size),
torch.nn.Tanh(),
torch.nn.Linear(config.hidden_size, kv_size)
)
else:
self.embedding = torch.nn.Embedding(config.prefix_size,
config.n_layer * self.head_dim * config.num_heads_per_kv * 2)
def forward(self, prefix: torch.Tensor):
if self.prefix_proj:
prefix_tokens = self.embedding(prefix)
past_key_values = self.trans(prefix_tokens)
else:
past_key_values = self.embedding(prefix)
return past_key_values
def _get_interleave(n):
def _get_interleave_power_of_2(n):
start = (2 ** (-2 ** -(math.log2(n) - 3)))
ratio = start
return [start * ratio ** i for i in range(n)]
if math.log2(n).is_integer():
return _get_interleave_power_of_2(n)
else:
closest_power_of_2 = 2 ** math.floor(math.log2(n))
return _get_interleave_power_of_2(closest_power_of_2) + \
_get_interleave(2 * closest_power_of_2)[0::2][:n - closest_power_of_2]
def _fill_with_neg_inf(t):
"""FP16-compatible function that fills a tensor with -inf."""
return t.float().fill_(float("-inf")).type_as(t)
def _gen_alibi_mask(n_head, max_pos):
"""used in inference only"""
slopes = torch.Tensor(_get_interleave(n_head))
alibi = slopes.unsqueeze(1).unsqueeze(1) * torch.arange(max_pos).unsqueeze(0).unsqueeze(0).expand(
n_head, -1, -1)
alibi = alibi.view(n_head, 1, max_pos)
alibi_mask = torch.triu(
_fill_with_neg_inf(torch.zeros([max_pos, max_pos])), 1
)
alibi_mask = alibi_mask.unsqueeze(0) + alibi
return alibi_mask
def _build_position_ids(input_ids, device):
batch_size, seq_length = input_ids.shape
position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
return position_ids
def _buffered_future_mask(tensor, maxpos, alibi, attn_heads):
"""used in training only"""
dim = tensor.size(0)
_future_mask = torch.triu(
_fill_with_neg_inf(torch.zeros([maxpos, maxpos])), 1
)
_future_mask = _future_mask.unsqueeze(0) + alibi
_future_mask = _future_mask.to(tensor)
return _future_mask[:tensor.shape[1] * attn_heads, :maxpos, :maxpos]
@torch.jit.script
def apply_rotary_pos_emb(x: torch.Tensor, rope_cache: torch.Tensor) -> torch.Tensor:
# x: [sq, b, np, hn]
sq, b, np, hn = x.size(0), x.size(1), x.size(2), x.size(3)
rot_dim = rope_cache.shape[-2] * 2
x, x_pass = x[..., :rot_dim], x[..., rot_dim:]
# truncate to support variable sizes
rope_cache = rope_cache[:sq]
xshaped = x.reshape(sq, -1, np, rot_dim // 2, 2)
rope_cache = rope_cache.view(sq, -1, 1, xshaped.size(3), 2)
x_out2 = torch.stack(
[
xshaped[..., 0] * rope_cache[..., 0] - xshaped[..., 1] * rope_cache[..., 1],
xshaped[..., 1] * rope_cache[..., 0] + xshaped[..., 0] * rope_cache[..., 1],
],
-1,
)
x_out2 = x_out2.flatten(3)
return torch.cat((x_out2, x_pass), dim=-1)
class RMSNorm(torch.nn.Module):
def __init__(self, normalized_shape, eps=1e-5, device=None, dtype=None, **kwargs):
super().__init__()
self.weight = torch.nn.Parameter(torch.empty(normalized_shape, device=device, dtype=dtype))
self.eps = eps
def forward(self, hidden_states: torch.Tensor):
input_dtype = hidden_states.dtype
variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
hidden_states = hidden_states * torch.rsqrt(variance + self.eps)
return (self.weight * hidden_states).to(input_dtype)
class SelfAttention(torch.nn.Module):
def __init__(self, config: BatGPTConfig, device=None):
super(SelfAttention, self).__init__()
self.num_heads = config.n_head
self.use_multi_query_attn = config.use_multi_query_attn
self.num_heads_per_kv = config.num_heads_per_kv
self.qkv_bias = config.qkv_bias
self.use_native_attn_impl = config.use_native_attn_impl
if not self.use_multi_query_attn:
assert self.num_heads_per_kv == self.num_heads, "num_heads_per_kv must equal to num_heads when not use_multi_query_attn"
self.head_dim = config.hidden_size // config.n_head
self.query_proj = nn.Linear(
config.hidden_size, config.hidden_size, bias=self.qkv_bias,
device=device, **_config_to_kwargs(config)
)
self.key_proj = nn.Linear(
config.hidden_size, self.head_dim * self.num_heads_per_kv, bias=self.qkv_bias,
device=device, **_config_to_kwargs(config)
)
self.value_proj = nn.Linear(
config.hidden_size, self.head_dim * self.num_heads_per_kv, bias=self.qkv_bias,
device=device, **_config_to_kwargs(config)
)
# Output.
self.dense = nn.Linear(
config.hidden_size, config.hidden_size, bias=False,
device=device, **_config_to_kwargs(config)
)
def forward(
self,
hidden_states,
attention_mask,
rotary_pos_emb,
kv_cache=None,
use_cache=True
):
# 1. query/key/value mapping
# hidden_states: [seq_len, batch_size, hidden_size]
seq_len, batch_size, hidden_size = hidden_states.shape
query_layer = self.query_proj(hidden_states)
key_layer = self.key_proj(hidden_states)
value_layer = self.value_proj(hidden_states)
query_layer = query_layer.view(seq_len, batch_size, self.num_heads, self.head_dim)
key_layer = key_layer.view(seq_len, batch_size, self.num_heads_per_kv, self.head_dim)
value_layer = value_layer.view(seq_len, batch_size, self.num_heads_per_kv, self.head_dim)
# 2. apply the rotary position embedding
if rotary_pos_emb is not None:
query_layer = apply_rotary_pos_emb(query_layer, rotary_pos_emb)
key_layer = apply_rotary_pos_emb(key_layer, rotary_pos_emb)
# 3. adjust key and value for inference
if kv_cache is not None:
cache_k, cache_v = kv_cache
key_layer = torch.cat((cache_k, key_layer), dim=0)
value_layer = torch.cat((cache_v, value_layer), dim=0)
if use_cache:
kv_cache = (key_layer, value_layer)
else:
kv_cache = None
# 4. repeat the key and value for attention
if self.num_heads_per_kv != self.num_heads:
key_layer = key_layer.unsqueeze(-2)
key_layer = key_layer.expand(
-1, -1, -1, self.num_heads // self.num_heads_per_kv, -1
)
key_layer = key_layer.contiguous().view(
key_layer.size()[:2] + (self.num_heads, self.head_dim)
)
value_layer = value_layer.unsqueeze(-2)
value_layer = value_layer.expand(
-1, -1, -1, self.num_heads // self.num_heads_per_kv, -1
)
value_layer = value_layer.contiguous().view(
value_layer.size()[:2] + (self.num_heads, self.head_dim)
)
# 5. attention [seq_len, batch_size, num_heads, head_dim] -> [batch_size, num_heads, seq_len, head_dim]
query_layer, key_layer, value_layer = [k.permute(1, 2, 0, 3) for k in [query_layer, key_layer, value_layer]]
pytorch_version = int(torch.__version__.split('.')[0])
if self.use_native_attn_impl and pytorch_version >= 2:
if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
is_causal=True)
else:
if attention_mask is not None:
attention_mask = ~attention_mask
context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
attention_mask)
else:
attention_scores = torch.matmul(query_layer, key_layer.transpose(2, 3)) / math.sqrt(self.head_dim)
if attention_mask is not None:
if seq_len == 1: # inference with cache
if len(attention_mask.size()) == 4:
attention_mask = attention_mask[:, :, -1:, :]
else:
attention_mask = attention_mask[:, -1:, :]
attention_scores = attention_scores + attention_mask
attention_scores = torch.max(attention_scores, torch.tensor(torch.finfo(attention_scores.dtype).min))
attention_probs = torch.nn.functional.softmax(attention_scores, dim=-1)
context_layer = torch.matmul(attention_probs, value_layer)
# [batch_size, num_heads, seq_len, head_dim] -> [seq_len, batch_size, num_heads, head_dim]
context_layer = context_layer.permute(2, 0, 1, 3)
# [seq_len, batch_size, hidden_size]
context_layer = context_layer.reshape(seq_len, batch_size, hidden_size)
#
output = self.dense(context_layer)
return output, kv_cache
def _config_to_kwargs(args):
common_kwargs = {
"dtype": args.torch_dtype,
}
return common_kwargs
class MLP(torch.nn.Module):
def __init__(self, config: BatGPTConfig, device=None):
super(MLP, self).__init__()
self.mlp_activation = config.mlp_activation
def swiglu(x):
x = torch.chunk(x, 2, dim=-1)
return F.silu(x[0]) * x[1]
def silu(x):
return F.silu(x)
# Project to 4h. If using swiglu double the output width, see https://arxiv.org/pdf/2002.05202.pdf
if self.mlp_activation == "swiglu":
self.activation_func = swiglu
self.gate_proj = None
self.dense_h_to_4h = nn.Linear(
config.hidden_size,
config.ffn_hidden_size * 2,
bias=False,
device=device,
**_config_to_kwargs(config)
)
elif self.mlp_activation == "silu":
self.activation_func = silu
self.gate_proj = nn.Linear(
config.hidden_size,
config.ffn_hidden_size,
bias=False,
device=device,
**_config_to_kwargs(config)
)
self.dense_h_to_4h = nn.Linear(
config.hidden_size,
config.ffn_hidden_size,
bias=False,
device=device,
**_config_to_kwargs(config)
)
else:
raise NotImplementedError("mlp_activation {} not supported".format(self.mlp_activation))
# Project back to h.
self.dense_4h_to_h = nn.Linear(
config.ffn_hidden_size,
config.hidden_size,
bias=False,
device=device,
**_config_to_kwargs(config)
)
def forward(self, hidden_states):
# [s, b, 4hp]
intermediate_parallel = self.dense_h_to_4h(hidden_states)
if self.mlp_activation == "swiglu":
intermediate_parallel = self.activation_func(intermediate_parallel)
elif self.mlp_activation == "silu":
gated_weight = self.activation_func(self.gate_proj(hidden_states))
intermediate_parallel = gated_weight * intermediate_parallel
else:
raise NotImplementedError("mlp_activation {} not supported".format(self.mlp_activation))
# [s, b, h]
output = self.dense_4h_to_h(intermediate_parallel)
return output
class BatGPTLayer(torch.nn.Module):
"""A single transformer layer.
Transformer layer takes input with size [s, b, h] and returns an
output of the same size.
"""
def __init__(self, config: BatGPTConfig, device=None):
super(BatGPTLayer, self).__init__()
# Layernorm on the input data.
self.input_layernorm = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon, device=device,
dtype=config.torch_dtype)
# Self attention.
self.self_attention = SelfAttention(config, device=device)
self.hidden_dropout = config.hidden_dropout
# Layernorm on the attention output
self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon, device=device,
dtype=config.torch_dtype)
# MLP
self.mlp = MLP(config, device=device)
def forward(
self,
hidden_states,
attention_mask,
rotary_pos_emb,
kv_cache=None,
use_cache=True,
):
# hidden_states: [s, b, h]
residual = hidden_states
# Layer norm at the beginning of the transformer layer.
layernorm_output = self.input_layernorm(hidden_states)
# Self attention.
attention_output, kv_cache = self.self_attention(
layernorm_output,
attention_mask,
rotary_pos_emb,
kv_cache=kv_cache,
use_cache=use_cache
)
# Residual connection.
layernorm_input = torch.nn.functional.dropout(attention_output, p=self.hidden_dropout, training=self.training)
layernorm_input = residual + layernorm_input
# Layer norm post the self attention.
layernorm_output = self.post_attention_layernorm(layernorm_input)
# MLP.
mlp_output = self.mlp(layernorm_output)
# Second residual connection.
residual = layernorm_input
output = torch.nn.functional.dropout(mlp_output, p=self.hidden_dropout, training=self.training)
output = residual + output
return output, kv_cache
class BatGPTTransformer(torch.nn.Module):
"""Transformer class."""
def __init__(self, config: BatGPTConfig, device=None):
super(BatGPTTransformer, self).__init__()
# Number of layers.
self.num_layers = config.n_layer
# Transformer layers.
def build_layer():
return BatGPTLayer(config, device=device)
self.layers = torch.nn.ModuleList([build_layer() for i in range(self.num_layers)])
# final layer norm before output.
self.ln_f = RMSNorm(config.hidden_size, eps=config.layer_norm_epsilon, device=device,
dtype=config.torch_dtype)
self.gradient_checkpointing = False
def _get_layer(self, layer_number):
return self.layers[layer_number]
def forward(
self,
hidden_states,
attention_mask,
rotary_pos_emb,
kv_caches=None,
use_cache: Optional[bool] = True,
output_hidden_states: Optional[bool] = False,
):
if not kv_caches:
kv_caches = [None for _ in range(self.num_layers)]
presents = () if use_cache else None
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
all_self_attentions = None
all_hidden_states = () if output_hidden_states else None
for index in range(self.num_layers):
if output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states,)
layer = self._get_layer(index)
if self.gradient_checkpointing and self.training:
layer_ret = torch.utils.checkpoint.checkpoint(
layer,
hidden_states,
attention_mask,
rotary_pos_emb,
kv_caches[index],
use_cache
)
else:
layer_ret = layer(
hidden_states,
attention_mask,
rotary_pos_emb,
kv_cache=kv_caches[index],
use_cache=use_cache
)
hidden_states, kv_cache = layer_ret
if use_cache:
presents = presents + (kv_cache,)
if output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states,)
hidden_states = self.ln_f(hidden_states)
return hidden_states, presents, all_hidden_states, all_self_attentions
class BatGPTPreTrainedModel(PreTrainedModel):
"""
An abstract class to handle weights initialization and
a simple interface for downloading and loading pretrained models.
"""
is_parallelizable = False
supports_gradient_checkpointing = True
config_class = BatGPTConfig
base_model_prefix = "transformer"
_no_split_modules = ["BatGPTLayer"]
def _init_weights(self, module: nn.Module):
"""Initialize the weights."""
return
def _set_gradient_checkpointing(self, module, value=False):
if isinstance(module, BatGPTTransformer):
module.gradient_checkpointing = value
class BatGPTModel(BatGPTPreTrainedModel):
def __init__(self, config: BatGPTConfig, device=None):
super().__init__(config)
self.num_layers = config.n_layer
self.num_heads = config.n_head
self.head_dim = config.hidden_size // config.n_head
self.max_seq_len = config.max_seq_len
self.pos_emb_impl = config.pos_emb_impl
self.model_cache_seq_len = 1024
# word embedding
self.word_embeddings = module_init(nn.Embedding,
config.empty_init,
config.vocab_size,
config.emb_dim,
dtype=config.torch_dtype,
device=device
)
self.emb_fact = None
if config.use_emb_factorization or config.emb_dim != config.hidden_size:
self.emb_fact = nn.Linear(config.emb_dim, config.hidden_size, bias=False,
dtype=config.torch_dtype, device=device)
init_kwargs = {}
if device is not None:
init_kwargs["device"] = device
self.encoder = module_init(BatGPTTransformer, config.empty_init, config, **init_kwargs)
self.first_run = True
self.alibi_mask = None
self.prefix_size = config.prefix_size
self.prefix_proj = config.prefix_proj
if self.prefix_size is not None:
for param in self.parameters():
param.requires_grad = False
self.prefix_tokens = torch.arange(self.prefix_size).long()
self.prefix_encoder = PrefixEncoder(config)
self.dropout = torch.nn.Dropout(0.1)
def get_input_embeddings(self):
return self.word_embeddings
def get_prompt(self, batch_size, device, dtype=torch.half):
prefix_tokens = self.prefix_tokens.unsqueeze(0).expand(batch_size, -1).to(device)
past_key_values = self.prefix_encoder(prefix_tokens).type(dtype)
past_key_values = past_key_values.view(
batch_size,
self.prefix_size,
self.num_layers * 2,
self.multi_query_group_num,
self.kv_channels
)
# seq_len, b, nh, hidden_size
past_key_values = self.dropout(past_key_values)
past_key_values = past_key_values.permute([2, 1, 0, 3, 4]).split(2)
return past_key_values
def get_rotary_tensor(self, seq_len: int, head_dim: int, dtype: torch.dtype, device: torch.device, base: int = 10000):
n_elem = head_dim // 2
# $\Theta = {\theta_i = 10000^{\frac{2(i-1)}{d}}, i \in [1, 2, ..., \frac{d}{2}]}$
theta = 1.0 / (base ** (torch.arange(0, n_elem, 2, dtype=dtype, device=device) / n_elem))
# Create position indexes `[0, 1, ..., seq_len - 1]`
seq_idx = torch.arange(seq_len, dtype=dtype, device=device)
# Calculate the product of position index and $\theta_i$
idx_theta = torch.outer(seq_idx, theta).float()
cache = torch.stack([torch.cos(idx_theta), torch.sin(idx_theta)], dim=-1)
# this is to mimic the behaviour of complex32, else we will get different results
if dtype in (torch.float16, torch.bfloat16, torch.int8):
cache = cache.bfloat16() if dtype == torch.bfloat16 else cache.half()
return cache
def get_causal_mask(self, input_ids, past_key_values, attention_mask=None) -> torch.BoolTensor:
batch_size, seq_length = input_ids.shape
# B x L x L
causal_mask = torch.ones(batch_size, seq_length, seq_length, device=input_ids.device)
causal_mask.tril_()
past_length = 0
if past_key_values:
past_length = past_key_values[0][0].shape[0]
if past_length:
causal_mask = torch.cat((torch.ones(batch_size, seq_length, past_length,
device=input_ids.device), causal_mask), dim=-1)
if attention_mask is not None:
causal_mask = causal_mask * attention_mask.unsqueeze(1)
if not past_length and attention_mask is not None:
causal_mask -= attention_mask.unsqueeze(-1) - 1
causal_mask = (causal_mask < 0.5).bool()
causal_mask.unsqueeze_(1)
return causal_mask
def get_alibi_mask(self, tensor, seq_length_with_past):
if self.training:
slopes = torch.Tensor(_get_interleave(self.num_heads))
alibi = slopes.unsqueeze(1).unsqueeze(1) * torch.arange(seq_length_with_past).unsqueeze(0).unsqueeze(0).expand(
self.num_heads,
-1, -1)
alibi = alibi.view(self.num_heads, 1, seq_length_with_past)
mask = _buffered_future_mask(tensor, seq_length_with_past, alibi, self.num_heads)
else:
if self.first_run:
self.first_run = False
self.register_buffer("future_mask", _gen_alibi_mask(self.num_heads, self.model_cache_seq_len).to(tensor), persistent=False)
if seq_length_with_past > self.model_cache_seq_len:
self.model_cache_seq_len = seq_length_with_past
self.register_buffer("future_mask", _gen_alibi_mask(self.num_heads, self.model_cache_seq_len).to(tensor), persistent=False)
mask = self.future_mask[:self.num_heads, :seq_length_with_past, :seq_length_with_past]
return mask
def forward(
self,
input_ids,
position_ids: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.BoolTensor] = None,
past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
inputs_embeds: Optional[torch.Tensor] = None,
use_cache: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
):
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
batch_size, seq_length = input_ids.shape
seq_length_with_past = seq_length
# -> word embedding
if inputs_embeds is None:
inputs_embeds = self.word_embeddings(input_ids)
# [b s h] --> [s b h].
inputs_embeds = inputs_embeds.transpose(0, 1).contiguous()
if self.prefix_size is not None:
if past_key_values is None:
past_key_values = self.get_prompt(batch_size=batch_size, device=input_ids.device,
dtype=inputs_embeds.dtype)
if attention_mask is not None:
attention_mask = torch.cat([attention_mask.new_ones((batch_size, self.prefix_size)),
attention_mask], dim=-1)
if past_key_values is not None:
past_key_values_length = past_key_values[0][0].shape[0]
seq_length_with_past = seq_length_with_past + past_key_values_length
full_attention_mask = None
rotary_pos_emb=None
if self.pos_emb_impl == "alibi":
if self.training:
if self.alibi_mask is None or self.alibi_mask.shape[-1] != seq_length_with_past:
self.alibi_mask = self.get_alibi_mask(inputs_embeds, seq_length_with_past)
alibi_mask = self.alibi_mask
else:
alibi_mask = self.get_alibi_mask(inputs_embeds, seq_length_with_past)
if attention_mask is not None:
if len(attention_mask.shape) == 2:
expanded_mask = attention_mask.to(alibi_mask.dtype)
expanded_mask = torch.tril(torch.gt(expanded_mask[:, :, None] * expanded_mask[:, None, :], 0)
) * torch.eq(expanded_mask[:, :, None] - expanded_mask[:, None, :], 0)
else:
expanded_mask = attention_mask
src_len, tgt_len = alibi_mask.size()[-2:]
expanded_mask = expanded_mask.unsqueeze(1).expand(batch_size, 1, src_len, tgt_len).to(alibi_mask.dtype)
# Target sizes: [1, 1, 41, 41]. Tensor sizes: [1, 1, 8, 8]
inverted_mask = 1.0 - expanded_mask
inverted_mask = inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(alibi_mask.dtype).min)
full_attention_mask = inverted_mask + alibi_mask.unsqueeze(0)
else:
full_attention_mask = alibi_mask
elif self.pos_emb_impl == "rope":
if (attention_mask is not None and not attention_mask.all()) or (past_key_values and seq_length != 1):
# B x 1 x L x L
full_attention_mask = self.get_causal_mask(input_ids, past_key_values, attention_mask)
# Rotary positional embeddings
rotary_pos_emb = self.get_rotary_tensor(self.max_seq_len, self.head_dim, dtype=inputs_embeds.dtype, device=inputs_embeds.device)
if position_ids is not None:
rotary_pos_emb = rotary_pos_emb[position_ids]
else:
rotary_pos_emb = rotary_pos_emb[None, :seq_length]
rotary_pos_emb = rotary_pos_emb.transpose(0, 1).contiguous()
else:
raise NotImplementedError("position embedding type: {} not supported!".format(self.pos_emb_impl))
# Run encoder.
hidden_states, presents, all_hidden_states, all_self_attentions = self.encoder(
inputs_embeds,
full_attention_mask,
rotary_pos_emb=rotary_pos_emb,
kv_caches=past_key_values,
use_cache=use_cache,
output_hidden_states=output_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 BaseModelOutputWithPast(
last_hidden_state=hidden_states,
past_key_values=presents,
hidden_states=all_hidden_states,
attentions=all_self_attentions,
)
class BatGPTForCausalLM(BatGPTPreTrainedModel):
def __init__(self, config: BatGPTConfig, device=None):
super().__init__(config)
self.max_sequence_length = config.max_length
self.model = BatGPTModel(config, device=device)
self.lm_head = module_init(nn.Linear, config.empty_init, config.hidden_size, config.vocab_size, bias=False,
dtype=config.torch_dtype, device=device)
self.config = config
def get_input_embeddings(self):
return self.model.get_input_embeddings()
def _update_model_kwargs_for_generation(
self,
outputs: ModelOutput,
model_kwargs: Dict[str, Any],
is_encoder_decoder: bool = False,
standardize_cache_format: bool = False,
) -> Dict[str, Any]:
# update past_key_values
model_kwargs["past_key_values"] = self._extract_past_from_model_output(
outputs, standardize_cache_format=standardize_cache_format
)
# update attention mask
if "attention_mask" in model_kwargs:
attention_mask = model_kwargs["attention_mask"]
model_kwargs["attention_mask"] = torch.cat(
[attention_mask, attention_mask.new_ones((attention_mask.shape[0], 1))], dim=-1
)
# update position ids
if "position_ids" in model_kwargs:
position_ids = model_kwargs["position_ids"]
new_position_id = position_ids[..., -1:].clone()
new_position_id += 1
model_kwargs["position_ids"] = torch.cat(
[position_ids, new_position_id], dim=-1
)
model_kwargs["is_first_forward"] = False
return model_kwargs
def prepare_inputs_for_generation(
self,
input_ids: torch.LongTensor,
past_key_values: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.Tensor] = None,
position_ids: Optional[torch.Tensor] = None,
is_first_forward: bool = True,
**kwargs
) -> dict:
# only last token for input_ids if past is not None
if position_ids is None:
position_ids = _build_position_ids(input_ids, device=input_ids.device)
if not is_first_forward:
position_ids = position_ids[..., -1:]
input_ids = input_ids[:, -1:]
return {
"input_ids": input_ids,
"past_key_values": past_key_values,
"position_ids": position_ids,
"attention_mask": attention_mask,
"return_last_logit": True
}
def forward(
self,
input_ids: Optional[torch.Tensor] = None,
position_ids: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.Tensor] = None,
past_key_values: Optional[Tuple[torch.FloatTensor]] = 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,
return_last_logit: Optional[bool] = False,
):
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
encodings = self.model(
input_ids=input_ids,
position_ids=position_ids,
attention_mask=attention_mask,
past_key_values=past_key_values,
inputs_embeds=inputs_embeds,
use_cache=use_cache,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
hidden_states = encodings[0]
if return_last_logit:
hidden_states = hidden_states[-1:]
lm_logits = self.lm_head(hidden_states)
lm_logits = lm_logits.transpose(0, 1).contiguous()
loss = None
if labels is not None:
lm_logits = lm_logits.to(torch.float32)
# Shift so that tokens < n predict n
shift_logits = lm_logits[..., :-1, :].contiguous()
shift_labels = labels[..., 1:].contiguous().to(shift_logits.device)
# Flatten the tokens
loss_fct = CrossEntropyLoss(ignore_index=-100)
loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
lm_logits = lm_logits.to(hidden_states.dtype)
loss = loss.to(hidden_states.dtype)
if not return_dict:
output = (lm_logits,) + encodings[1:]
return ((loss,) + output) if loss is not None else output
return CausalLMOutputWithPast(
loss=loss,
logits=lm_logits,
past_key_values=encodings.past_key_values,
hidden_states=encodings.hidden_states,
attentions=encodings.attentions,
)
@staticmethod
def _reorder_cache(
past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
"""
This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
[`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
beam_idx at every generation step.
Output shares the same memory storage as `past`.
"""
return tuple(
(
layer_past[0].index_select(1, beam_idx.to(layer_past[0].device)),
layer_past[1].index_select(1, beam_idx.to(layer_past[1].device)),
)
for layer_past in past
)
def process_response(self, response):
response = response.strip()
return response
def build_inputs(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, system_prompt = None):
inputs = tokenizer.build_inputs(query, history=history, system_prompt=system_prompt)
inputs = inputs.to(self.device)
return inputs
def build_stream_inputs(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, system_prompt = None):
inputs = tokenizer.build_stream_inputs(query, history=history, system_prompt=system_prompt)
inputs = inputs.to(self.device)
return inputs
@torch.no_grad()
def chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, system_prompt=None, max_length: int = 8192, num_beams=1,
do_sample=True, top_p=0.8, temperature=0.8, logits_processor=None, **kwargs):
if history is None:
history = []
if logits_processor is None:
logits_processor = LogitsProcessorList()
logits_processor.append(InvalidScoreLogitsProcessor())
gen_kwargs = {"max_length": max_length, "num_beams": num_beams, "do_sample": do_sample, "top_p": top_p,
"temperature": temperature, **kwargs} #, "logits_processor": logits_processor
inputs = self.build_inputs(tokenizer, query, history=history, system_prompt=system_prompt)
outputs = self.generate(**inputs, **gen_kwargs)
outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):]
response = tokenizer.decode(outputs, skip_special_tokens=True) #
response = self.process_response(response)
history = history + [(query, response)]
return response, history
@torch.no_grad()
def stream_chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, system_prompt=None, past_key_values=None,
max_length: int = 8192, do_sample=True, top_p=0.8, temperature=0.8, logits_processor=None,
return_past_key_values=False, **kwargs):
if history is None:
history = []
if logits_processor is None:
logits_processor = LogitsProcessorList()
logits_processor.append(InvalidScoreLogitsProcessor())
gen_kwargs = {"max_length": max_length, "do_sample": do_sample, "top_p": top_p,
"temperature": temperature, "logits_processor": logits_processor, **kwargs}
if past_key_values is None and not return_past_key_values:
inputs = self.build_inputs(tokenizer, query, history=history, system_prompt=system_prompt)
else:
inputs = self.build_stream_inputs(tokenizer, query, history=history, system_prompt=system_prompt)
if past_key_values is not None:
past_length = past_key_values[0][0].shape[0]
if self.model.prefix_size is not None:
past_length -= self.transformer.prefix_size
inputs.position_ids += past_length
attention_mask = inputs.attention_mask
attention_mask = torch.cat((attention_mask.new_ones(1, past_length), attention_mask), dim=1)
inputs['attention_mask'] = attention_mask
for outputs in self.stream_generate(**inputs, past_key_values=past_key_values,
return_past_key_values=return_past_key_values, **gen_kwargs):
if return_past_key_values:
outputs, past_key_values = outputs
outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):]
response = tokenizer.decode(outputs)
if response and response[-1] != "�":
response = self.process_response(response)
new_history = history + [(query, response)]
if return_past_key_values:
yield response, new_history, past_key_values
else:
yield response, new_history
@torch.no_grad()
def stream_generate(
self,
input_ids,
generation_config: Optional[GenerationConfig] = None,
logits_processor: Optional[LogitsProcessorList] = None,
stopping_criteria: Optional[StoppingCriteriaList] = None,
prefix_allowed_tokens_fn: Optional[Callable[[int, torch.Tensor], List[int]]] = None,
return_past_key_values=False,
**kwargs,
):
batch_size, input_ids_seq_length = input_ids.shape[0], input_ids.shape[-1]
if generation_config is None:
generation_config = self.generation_config
generation_config = copy.deepcopy(generation_config)
model_kwargs = generation_config.update(**kwargs)
bos_token_id, eos_token_id = generation_config.bos_token_id, generation_config.eos_token_id
if isinstance(eos_token_id, int):
eos_token_id = [eos_token_id]
has_default_max_length = kwargs.get("max_length") is None and generation_config.max_length is not None
if has_default_max_length and generation_config.max_new_tokens is None:
warnings.warn(
f"Using `max_length`'s default ({generation_config.max_length}) to control the generation length. "
"This behaviour is deprecated and will be removed from the config in v5 of Transformers -- we"
" recommend using `max_new_tokens` to control the maximum length of the generation.",
UserWarning,
)
elif generation_config.max_new_tokens is not None:
generation_config.max_length = generation_config.max_new_tokens + input_ids_seq_length
if not has_default_max_length:
logger.warn(
f"Both `max_new_tokens` (={generation_config.max_new_tokens}) and `max_length`(="
f"{generation_config.max_length}) seem to have been set. `max_new_tokens` will take precedence. "
"Please refer to the documentation for more information. "
"(https://huggingface.co/docs/transformers/main/en/main_classes/text_generation)",
UserWarning,
)
if input_ids_seq_length >= generation_config.max_length:
input_ids_string = "decoder_input_ids" if self.config.is_encoder_decoder else "input_ids"
logger.warning(
f"Input length of {input_ids_string} is {input_ids_seq_length}, but `max_length` is set to"
f" {generation_config.max_length}. This can lead to unexpected behavior. You should consider"
" increasing `max_new_tokens`."
)
# 2. Set generation parameters if not already defined
logits_processor = logits_processor if logits_processor is not None else LogitsProcessorList()
stopping_criteria = stopping_criteria if stopping_criteria is not None else StoppingCriteriaList()
logits_processor = self._get_logits_processor(
generation_config=generation_config,
input_ids_seq_length=input_ids_seq_length,
encoder_input_ids=input_ids,
prefix_allowed_tokens_fn=prefix_allowed_tokens_fn,
logits_processor=logits_processor,
)
stopping_criteria = self._get_stopping_criteria(
generation_config=generation_config, stopping_criteria=stopping_criteria
)
logits_warper = self._get_logits_warper(generation_config)
unfinished_sequences = input_ids.new(input_ids.shape[0]).fill_(1)
scores = None
while True:
model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
# forward pass to get next token
outputs = self(
**model_inputs,
return_dict=True,
output_attentions=False,
output_hidden_states=False,
)
next_token_logits = outputs.logits[:, -1, :]
# pre-process distribution
next_token_scores = logits_processor(input_ids, next_token_logits)
next_token_scores = logits_warper(input_ids, next_token_scores)
# sample
probs = nn.functional.softmax(next_token_scores, dim=-1)
if generation_config.do_sample:
next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
else:
next_tokens = torch.argmax(probs, dim=-1)
# update generated ids, model inputs, and length for next step
input_ids = torch.cat([input_ids, next_tokens[:, None]], dim=-1)
model_kwargs = self._update_model_kwargs_for_generation(
outputs, model_kwargs, is_encoder_decoder=self.config.is_encoder_decoder
)
unfinished_sequences = unfinished_sequences.mul((sum(next_tokens != i for i in eos_token_id)).long())
if return_past_key_values:
yield input_ids, outputs.past_key_values
else:
yield input_ids
# stop when each sentence is finished, or if we exceed the maximum length
if unfinished_sequences.max() == 0 or stopping_criteria(input_ids, scores):
break