This commit is contained in:
josc146
14
finetune/lora/v6/fla/models/linear_attn/__init__.py
vendored
Normal file
14
finetune/lora/v6/fla/models/linear_attn/__init__.py
vendored
Normal file
@@ -0,0 +1,14 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
|
||||
from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
|
||||
|
||||
from fla.models.linear_attn.configuration_linear_attn import \
|
||||
LinearAttentionConfig
|
||||
from fla.models.linear_attn.modeling_linear_attn import (
|
||||
LinearAttentionForCausalLM, LinearAttentionModel)
|
||||
|
||||
AutoConfig.register(LinearAttentionConfig.model_type, LinearAttentionConfig)
|
||||
AutoModel.register(LinearAttentionConfig, LinearAttentionModel)
|
||||
AutoModelForCausalLM.register(LinearAttentionConfig, LinearAttentionForCausalLM)
|
||||
|
||||
__all__ = ['LinearAttentionConfig', 'LinearAttentionForCausalLM', 'LinearAttentionModel']
|
||||
70
finetune/lora/v6/fla/models/linear_attn/configuration_linear_attn.py
vendored
Normal file
70
finetune/lora/v6/fla/models/linear_attn/configuration_linear_attn.py
vendored
Normal file
@@ -0,0 +1,70 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
|
||||
from typing import Optional
|
||||
|
||||
from transformers.configuration_utils import PretrainedConfig
|
||||
|
||||
|
||||
class LinearAttentionConfig(PretrainedConfig):
|
||||
|
||||
model_type = 'linear_attn'
|
||||
keys_to_ignore_at_inference = ['past_key_values']
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
vocab_size: int = 32000,
|
||||
hidden_size: int = 2048,
|
||||
expand_k: int = 1,
|
||||
expand_v: int = 1,
|
||||
hidden_ratio: Optional[int] = 4,
|
||||
intermediate_size: Optional[int] = None,
|
||||
num_hidden_layers: int = 24,
|
||||
num_heads: int = 4,
|
||||
attn_mode: str = "fused_chunk",
|
||||
feature_map: str = "elementwise_product",
|
||||
tie_feature_map_qk: bool = False,
|
||||
norm_q: bool = False,
|
||||
norm_k: bool = False,
|
||||
norm_feature_map: bool = False,
|
||||
hidden_act: str = "swish",
|
||||
max_position_embeddings: int = 2048,
|
||||
elementwise_affine: Optional[bool] = True,
|
||||
norm_eps: float = 1e-6,
|
||||
use_cache: bool = True,
|
||||
pad_token_id: int = None,
|
||||
bos_token_id: int = 1,
|
||||
eos_token_id: int = 2,
|
||||
tie_word_embeddings: bool = False,
|
||||
initializer_range: float = 0.02,
|
||||
fuse_cross_entropy: bool = True,
|
||||
**kwargs
|
||||
):
|
||||
self.vocab_size = vocab_size
|
||||
self.max_position_embeddings = max_position_embeddings
|
||||
self.hidden_size = hidden_size
|
||||
self.expand_k = expand_k
|
||||
self.expand_v = expand_v
|
||||
self.hidden_ratio = hidden_ratio
|
||||
self.intermediate_size = intermediate_size
|
||||
self.num_hidden_layers = num_hidden_layers
|
||||
self.num_heads = num_heads
|
||||
self.attn_mode = attn_mode
|
||||
self.feature_map = feature_map
|
||||
self.tie_feature_map_qk = tie_feature_map_qk
|
||||
self.norm_q = norm_q
|
||||
self.norm_k = norm_k
|
||||
self.norm_feature_map = norm_feature_map
|
||||
self.hidden_act = hidden_act
|
||||
self.elementwise_affine = elementwise_affine
|
||||
self.norm_eps = norm_eps
|
||||
self.use_cache = use_cache
|
||||
self.initializer_range = initializer_range
|
||||
self.fuse_cross_entropy = fuse_cross_entropy
|
||||
|
||||
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,
|
||||
)
|
||||
424
finetune/lora/v6/fla/models/linear_attn/modeling_linear_attn.py
vendored
Normal file
424
finetune/lora/v6/fla/models/linear_attn/modeling_linear_attn.py
vendored
Normal file
@@ -0,0 +1,424 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import math
|
||||
import warnings
|
||||
from typing import List, Optional, Tuple, Union
|
||||
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.utils.checkpoint
|
||||
from transformers.activations import ACT2FN
|
||||
from transformers.cache_utils import Cache, DynamicCache
|
||||
from transformers.modeling_outputs import (BaseModelOutputWithPast,
|
||||
CausalLMOutputWithPast)
|
||||
from transformers.modeling_utils import PreTrainedModel
|
||||
from transformers.utils import logging
|
||||
|
||||
from fla.layers.linear_attn import LinearAttention
|
||||
from fla.models.linear_attn.configuration_linear_attn import \
|
||||
LinearAttentionConfig
|
||||
from fla.modules import FusedCrossEntropyLoss, RMSNorm
|
||||
from fla.modules.activations import swiglu_linear
|
||||
|
||||
logger = logging.get_logger(__name__)
|
||||
|
||||
|
||||
class LinearAttentionMLP(nn.Module):
|
||||
def __init__(
|
||||
self,
|
||||
hidden_size: int,
|
||||
hidden_ratio: Optional[int] = None,
|
||||
intermediate_size: Optional[int] = None,
|
||||
hidden_act: str = 'swish'
|
||||
) -> LinearAttentionMLP:
|
||||
super().__init__()
|
||||
|
||||
self.hidden_size = hidden_size
|
||||
# the final number of params is `hidden_ratio * hidden_size^2`
|
||||
# `intermediate_size` is chosen to be a multiple of 256 closest to `2/3 * hidden_size * hidden_ratio`
|
||||
if hidden_ratio is None:
|
||||
hidden_ratio = 4
|
||||
if intermediate_size is None:
|
||||
intermediate_size = int(hidden_size * hidden_ratio * 2 / 3)
|
||||
intermediate_size = 256 * ((intermediate_size + 256 - 1) // 256)
|
||||
self.hidden_ratio = hidden_ratio
|
||||
self.intermediate_size = intermediate_size
|
||||
|
||||
self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size * 2, bias=False)
|
||||
self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
|
||||
self.act_fn = ACT2FN[hidden_act]
|
||||
|
||||
def forward(self, x):
|
||||
y = self.gate_proj(x)
|
||||
gate, y = y.chunk(2, -1)
|
||||
return swiglu_linear(gate, y, self.down_proj.weight, self.down_proj.bias)
|
||||
|
||||
|
||||
class LinearAttentionBlock(nn.Module):
|
||||
def __init__(self, config: LinearAttentionConfig, layer_idx: int):
|
||||
super().__init__()
|
||||
self.hidden_size = config.hidden_size
|
||||
|
||||
self.attn_norm = RMSNorm(hidden_size=config.hidden_size, eps=config.norm_eps)
|
||||
self.attn = LinearAttention(
|
||||
hidden_size=config.hidden_size,
|
||||
expand_k=config.expand_k,
|
||||
expand_v=config.expand_v,
|
||||
num_heads=config.num_heads,
|
||||
mode=config.attn_mode,
|
||||
feature_map=config.feature_map,
|
||||
tie_feature_map_qk=config.tie_feature_map_qk,
|
||||
norm_q=config.norm_q,
|
||||
norm_k=config.norm_k,
|
||||
do_feature_map_norm=config.norm_feature_map,
|
||||
elementwise_affine=config.elementwise_affine,
|
||||
norm_eps=config.norm_eps,
|
||||
layer_idx=layer_idx
|
||||
)
|
||||
self.mlp_norm = RMSNorm(hidden_size=config.hidden_size, eps=config.norm_eps)
|
||||
self.mlp = LinearAttentionMLP(
|
||||
hidden_size=config.hidden_size,
|
||||
hidden_ratio=config.hidden_ratio,
|
||||
intermediate_size=config.intermediate_size,
|
||||
hidden_act=config.hidden_act
|
||||
)
|
||||
|
||||
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,
|
||||
**kwargs,
|
||||
) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
|
||||
|
||||
residual = hidden_states
|
||||
# currently not supported
|
||||
attn_weights, present_key_value = None, None
|
||||
|
||||
hidden_states = self.attn_norm(hidden_states)
|
||||
hidden_states = self.attn(hidden_states)
|
||||
hidden_states, residual = self.mlp_norm(hidden_states, residual, True)
|
||||
hidden_states = self.mlp(hidden_states)
|
||||
hidden_states = residual + hidden_states
|
||||
|
||||
outputs = (hidden_states,)
|
||||
|
||||
if output_attentions:
|
||||
outputs += (attn_weights,)
|
||||
|
||||
if use_cache:
|
||||
outputs += (present_key_value,)
|
||||
|
||||
return outputs
|
||||
|
||||
|
||||
class LinearAttentionPreTrainedModel(PreTrainedModel):
|
||||
config_class = LinearAttentionConfig
|
||||
supports_gradient_checkpointing = True
|
||||
_no_split_modules = ['LinearAttentionBlock']
|
||||
|
||||
def __init__(self, *inputs, **kwargs):
|
||||
super().__init__(*inputs, **kwargs)
|
||||
|
||||
def _init_weights(
|
||||
self,
|
||||
module: nn.Module,
|
||||
rescale_prenorm_residual: bool = True,
|
||||
num_residuals_per_layer: int = 2,
|
||||
):
|
||||
if isinstance(module, (nn.Linear, nn.Conv1d)):
|
||||
# Slightly different from the TF version which uses truncated_normal for initialization
|
||||
# cf https://github.com/pytorch/pytorch/pull/5617
|
||||
nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
|
||||
if module.bias is not None:
|
||||
nn.init.zeros_(module.bias)
|
||||
elif isinstance(module, nn.Embedding):
|
||||
nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
|
||||
if module.padding_idx is not None:
|
||||
module.weight.data[module.padding_idx].zero_()
|
||||
|
||||
if rescale_prenorm_residual:
|
||||
# Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
|
||||
# > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
|
||||
# > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
|
||||
# > -- GPT-2 :: https://openai.com/blog/better-language-models/
|
||||
#
|
||||
# Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
|
||||
for name, p in module.named_parameters():
|
||||
if name in ["o_proj.weight", "down_proj.weight"]:
|
||||
# Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
|
||||
# Following Pytorch init, except scale by 1/sqrt(2 * n_layer)
|
||||
# We need to reinit p since this code could be called multiple times
|
||||
# Having just p *= scale would repeatedly scale it down
|
||||
with torch.no_grad():
|
||||
p /= math.sqrt(num_residuals_per_layer * self.config.num_hidden_layers)
|
||||
|
||||
|
||||
class LinearAttentionModel(LinearAttentionPreTrainedModel):
|
||||
|
||||
def __init__(self, config: LinearAttentionConfig):
|
||||
super().__init__(config)
|
||||
self.padding_idx = config.pad_token_id
|
||||
self.vocab_size = config.vocab_size
|
||||
|
||||
self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
|
||||
self.layers = nn.ModuleList(
|
||||
[LinearAttentionBlock(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
|
||||
)
|
||||
self.norm = RMSNorm(config.hidden_size, eps=config.norm_eps)
|
||||
|
||||
self.gradient_checkpointing = False
|
||||
|
||||
self.post_init()
|
||||
|
||||
def get_input_embeddings(self):
|
||||
return self.embeddings
|
||||
|
||||
def set_input_embeddings(self, value):
|
||||
self.embeddings = value
|
||||
|
||||
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]:
|
||||
if output_attentions:
|
||||
warnings.warn(
|
||||
"`LinearAttentionModel` does not support output attention weights now, "
|
||||
"so `output_attentions` is set to `False`."
|
||||
)
|
||||
output_attentions = False
|
||||
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 if not self.training else False)
|
||||
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:
|
||||
_, seq_length = input_ids.shape[:2]
|
||||
elif inputs_embeds is not None:
|
||||
_, seq_length = inputs_embeds.shape[:2]
|
||||
else:
|
||||
raise ValueError("You have to specify either input_ids or inputs_embeds")
|
||||
|
||||
past_key_values_length = 0
|
||||
if use_cache:
|
||||
use_legacy_cache = not isinstance(past_key_values, Cache)
|
||||
if use_legacy_cache:
|
||||
past_key_values = DynamicCache.from_legacy_cache(past_key_values)
|
||||
past_key_values_length = past_key_values.get_usable_length(seq_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)
|
||||
|
||||
if inputs_embeds is None:
|
||||
inputs_embeds = self.embeddings(input_ids)
|
||||
|
||||
# embed positions
|
||||
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 = None
|
||||
|
||||
for decoder_layer in self.layers:
|
||||
if output_hidden_states:
|
||||
all_hidden_states += (hidden_states,)
|
||||
|
||||
if self.gradient_checkpointing and self.training:
|
||||
layer_outputs = self._gradient_checkpointing_func(
|
||||
decoder_layer.__call__,
|
||||
hidden_states,
|
||||
attention_mask,
|
||||
position_ids,
|
||||
past_key_values,
|
||||
output_attentions,
|
||||
use_cache,
|
||||
)
|
||||
else:
|
||||
layer_outputs = decoder_layer(
|
||||
hidden_states,
|
||||
attention_mask=attention_mask,
|
||||
position_ids=position_ids,
|
||||
past_key_value=past_key_values,
|
||||
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 = None
|
||||
if use_cache:
|
||||
next_cache = next_decoder_cache.to_legacy_cache() if use_legacy_cache else next_decoder_cache
|
||||
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 LinearAttentionForCausalLM(LinearAttentionPreTrainedModel):
|
||||
_tied_weights_keys = ["lm_head.weight"]
|
||||
|
||||
def __init__(self, config):
|
||||
super().__init__(config)
|
||||
self.model = LinearAttentionModel(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.embeddings
|
||||
|
||||
def set_input_embeddings(self, value):
|
||||
self.model.embeddings = 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 generate(self, *args, **kwargs):
|
||||
try:
|
||||
return super().generate(*args, **kwargs)
|
||||
except AttributeError as exc:
|
||||
# Expected exception: "AttributeError: '(object name)' object has no attribute 'past_key_values'"
|
||||
if 'past_key_values' in str(exc):
|
||||
raise AttributeError(
|
||||
f"You tried to call `generate` with a decoding strategy that manipulates `past_key_values`, "
|
||||
f"which is not supported for {self.__class__.__name__}. "
|
||||
f"Try another generation strategy instead. "
|
||||
f"For the available generation strategies, check this doc: "
|
||||
f"https://huggingface.co/docs/transformers/en/generation_strategies#decoding-strategies"
|
||||
)
|
||||
else:
|
||||
raise exc
|
||||
|
||||
def prepare_inputs_for_generation(
|
||||
self,
|
||||
input_ids: torch.LongTensor = None,
|
||||
state: Optional[torch.Tensor] = None,
|
||||
inputs_embeds: Optional[torch.FloatTensor] = None,
|
||||
**kwargs
|
||||
):
|
||||
# only last token for inputs_ids if the state is passed along.
|
||||
if state is not None:
|
||||
input_ids = input_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 state is None:
|
||||
model_inputs = {"inputs_embeds": inputs_embeds}
|
||||
else:
|
||||
model_inputs = {"input_ids": input_ids}
|
||||
model_inputs["state"] = state
|
||||
return model_inputs
|
||||
|
||||
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:
|
||||
if self.config.fuse_cross_entropy:
|
||||
loss_fct = FusedCrossEntropyLoss(inplace_backward=True)
|
||||
else:
|
||||
loss_fct = nn.CrossEntropyLoss()
|
||||
# Enable model parallelism
|
||||
labels = labels.to(logits.device)
|
||||
labels = torch.cat((labels[..., 1:], torch.full_like(labels[:, :1], loss_fct.ignore_index)), 1)
|
||||
loss = loss_fct(logits.view(-1, self.config.vocab_size), labels.view(-1))
|
||||
|
||||
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,
|
||||
)
|
||||
Reference in New Issue
Block a user