# Copyright 2023 The EASYDEL Author @erfanzar (Erfan Zare Chavoshi).
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import functools
import typing as tp
import chex
import jax
from flax import nnx as nn
from jax import numpy as jnp
from easydel.infra.base_module import EasyDeLBaseModule
from easydel.infra.factory import TaskType, register_module
from easydel.infra.modeling_outputs import BaseModelOutput, CausalLMOutput
from easydel.infra.utils import (
ACT2FN,
auto_remat,
control_mlp_sharding,
)
from easydel.layers.attention import AttentionModule, FlexibleAttentionModule
from easydel.layers.caching import (
PagedAttentionCache,
PagedAttentionCacheView,
PagedAttentionMetadata,
TransformerCache,
TransformerCacheView,
TransformerMetadata,
)
from easydel.layers.linear import ParallelLinear
from .gpt_neox_configuration import GPTNeoXConfig as GPTNeoXConfig
[docs]class GPTNeoXAttention(AttentionModule):
"""GPT-NeoX Attention module.
This module implements the attention mechanism used in the GPT-NeoX model,
including rotary position embeddings and parallel linear layers for QKV.
Attributes:
config (GPTNeoXConfig): Configuration object for the model.
dtype (jnp.dtype): Data type for computations.
param_dtype (jnp.dtype): Data type for parameters.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations.
rngs (nn.Rngs): Random number generators.
"""
def __init__(
self,
config: GPTNeoXConfig,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
) -> None:
self.config = config
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
self.head_dim = self.config.hidden_size // self.config.num_attention_heads
self.rotary = self.config.get_basic_rope(
dtype=dtype,
head_size=self.head_dim,
rotary_dim=int(self.head_dim * self.config.rotary_pct),
base=self.config.rotary_emb_base,
)
self.query_key_value = ParallelLinear(
config.hidden_size,
3 * config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.dense = ParallelLinear(
config.hidden_size,
config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.attention_performer = FlexibleAttentionModule(
base_config=config,
softmax_scale=self.head_dim**-0.5,
dropout_prob=config.attention_dropout,
)
def _split_heads(self, hidden_states):
return hidden_states.reshape(
hidden_states.shape[:2] + (self.config.num_attention_heads, self.head_dim)
)
def __call__(
self,
hidden_states: chex.Array,
attention_mask: chex.Array,
position_ids: chex.Array,
causal_mask: tp.Optional[chex.Array] = None,
segment_ids: tp.Optional[chex.Array] = None,
cache_view: tp.Optional[TransformerCacheView | PagedAttentionCacheView] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
output_attentions: bool = False,
frequencies: tp.Optional[chex.Array] = None,
):
"""Forward pass of the GPTNeoXAttention module.
Args:
hidden_states (chex.Array): Input hidden states.
attention_mask (chex.Array): Mask to apply on the attention scores.
position_ids (chex.Array): Position indices for the tokens.
causal_mask (chex.Array, optional): Causal mask for ensuring autoregressive behavior.
segment_ids (tp.Optional[chex.Array], optional): Segment IDs for segment-based attention.
cache_view (tp.Optional[TransformerCacheView | PagedAttentionCacheView], optional): Cache view for key/value states.
cache_metadata (tp.Optional[TransformerMetadata | PagedAttentionMetadata], optional): Metadata for cache handling.
output_attentions (bool, optional): Whether to return attention weights.
frequencies (tp.Optional[chex.Array], optional): Precomputed rotary frequencies.
Returns:
tp.Tuple[chex.Array, tp.Optional[chex.Array]]: A tuple containing the attention output and optionally the attention weights.
"""
query, key, value = jnp.split(
self.query_key_value(hidden_states),
indices_or_sections=3,
axis=-1,
)
query = self._split_heads(query)
key = self._split_heads(key)
value = self._split_heads(value)
query, key = self.rotary(
positions=position_ids,
query=query,
key=key,
frequencies=frequencies,
)
(
key,
value,
attention_mask,
init_attention_bias,
) = self.concatenate(
query=query,
key=key,
cache_view=cache_view,
value=value,
attention_mask=attention_mask,
causal_mask=causal_mask,
fcm_mask=None,
)
attentions = self.attention_performer.forward(
query_states=query,
key_states=key,
value_states=value,
bias=None,
cache_metadata=cache_metadata,
cache_view=cache_view,
init_bias=init_attention_bias,
attention_mask=attention_mask,
segment_ids=segment_ids,
causal=True,
dropout_rng=self.rngs.params(),
)
attn_output = self.shard_attention_prod(
self._merge_heads(attentions.attention_outputs)
)
attn_output = self.dense(attn_output)
return attn_output, attentions.attention_weights
[docs]class GPTNeoXMlp(nn.Module):
"""GPT-NeoX MLP module.
This module implements the feed-forward network used in the GPT-NeoX model.
Attributes:
config (GPTNeoXConfig): Configuration object for the model.
dtype (jnp.dtype): Data type for computations.
param_dtype (jnp.dtype): Data type for parameters.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations.
rngs (nn.Rngs): Random number generators.
"""
def __init__(
self,
config: GPTNeoXConfig,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
) -> None:
self.config = config
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.dense_h_to_4h = ParallelLinear(
self.config.hidden_size,
self.config.intermediate_size,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.dense_4h_to_h = ParallelLinear(
self.config.intermediate_size,
self.config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.act = ACT2FN[self.config.hidden_act]
def __call__(self, hidden_states):
"""Forward pass of the GPTNeoXMlp module.
Args:
hidden_states (chex.Array): Input hidden states.
Returns:
chex.Array: Output hidden states after processing through the MLP.
"""
hidden_states = control_mlp_sharding(
hidden_states,
self.config.partition_axis,
)
return self.dense_4h_to_h(self.act(self.dense_h_to_4h(hidden_states)))
[docs]class GPTNeoXBlock(nn.Module):
"""GPT-NeoX Transformer block.
This module represents a single transformer block in the GPT-NeoX model,
containing self-attention and MLP sub-layers with residual connections
and layer normalization. It supports both standard and parallel residual connections.
Attributes:
config (GPTNeoXConfig): Configuration object for the model.
dtype (jnp.dtype): Data type for computations.
param_dtype (jnp.dtype): Data type for parameters.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations.
rngs (nn.Rngs): Random number generators.
"""
def __init__(
self,
config: GPTNeoXConfig,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
) -> None:
self.config = config
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
self.use_parallel_residual = config.use_parallel_residual
attn_block = GPTNeoXAttention
mlp_block = GPTNeoXMlp
attn_block, mlp_block = auto_remat(
attn_block,
mlp_block,
policy=config.gradient_checkpointing,
)
self.input_layernorm = nn.LayerNorm(
config.hidden_size,
epsilon=config.layer_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.post_attention_layernorm = nn.LayerNorm(
config.hidden_size,
epsilon=config.layer_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.attention = GPTNeoXAttention(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.mlp = GPTNeoXMlp(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
def __call__(
self,
hidden_states: chex.Array,
attention_mask: chex.Array,
position_ids: chex.Array,
causal_mask: tp.Optional[chex.Array] = None,
segment_ids: tp.Optional[chex.Array] = None,
cache_view: tp.Optional[TransformerCacheView | PagedAttentionCacheView] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
output_attentions: bool = False,
frequencies: tp.Optional[chex.Array] = None,
):
"""Forward pass of the GPTNeoXBlock module.
Args:
hidden_states (chex.Array): Input hidden states.
attention_mask (chex.Array): Mask to apply on the attention scores.
position_ids (chex.Array): Position indices for the tokens.
causal_mask (chex.Array, optional): Causal mask for ensuring autoregressive behavior.
segment_ids (tp.Optional[chex.Array], optional): Segment IDs for segment-based attention.
cache_view (tp.Optional[TransformerCacheView | PagedAttentionCacheView], optional): Cache view for key/value states.
cache_metadata (tp.Optional[TransformerMetadata | PagedAttentionMetadata], optional): Metadata for cache handling.
output_attentions (bool, optional): Whether to return attention weights.
frequencies (tp.Optional[chex.Array], optional): Precomputed rotary frequencies.
Returns:
tp.Tuple[chex.Array, tp.Optional[chex.Array]]: A tuple containing the output hidden states and optionally the attention weights.
"""
attn_out = self.attention(
self.input_layernorm(hidden_states),
attention_mask,
position_ids,
causal_mask,
segment_ids,
cache_view,
cache_metadata,
output_attentions,
frequencies,
)
attn = attn_out[0]
if self.use_parallel_residual:
mlp = self.mlp(self.post_attention_layernorm(hidden_states))
hidden_states = mlp + hidden_states + attn
else:
hidden_states = attn + hidden_states
hidden_states = (
self.mlp(self.post_attention_layernorm(hidden_states)) + hidden_states
)
return (hidden_states,) + attn_out[1:]
[docs]@register_module(
TaskType.BASE_MODULE,
config=GPTNeoXConfig,
model_type="gpt_neox",
)
class GPTNeoXModel(EasyDeLBaseModule):
"""GPT-NeoX model implementation.
This class implements the main GPT-NeoX transformer model architecture, consisting of
an embedding layer, multiple GPTNeoXBlock layers, and a final layer normalization.
Attributes:
config (GPTNeoXConfig): Configuration object for the model.
dtype (jnp.dtype): Data type for computations.
param_dtype (jnp.dtype): Data type for parameters.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations.
rngs (nn.Rngs): Random number generators.
"""
def __init__(
self,
config: GPTNeoXConfig,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: tp.Optional[tp.Union[str, jax.lax.Precision]] = None,
*,
rngs: nn.Rngs,
):
super().__init__(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.embed_in = nn.Embed(
self.config.vocab_size,
self.config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.emb_dropout = nn.Dropout(config.hidden_dropout, rngs=rngs)
self.layers = [
GPTNeoXBlock(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
for i in range(config.num_hidden_layers)
]
self.final_layer_norm = nn.LayerNorm(
config.hidden_size,
epsilon=self.config.layer_norm_eps,
dtype=self.dtype,
param_dtype=param_dtype,
rngs=rngs,
)
@functools.cached_property
def frequencies(self):
head_dim = self.config.hidden_size // self.config.num_attention_heads
return self.config.get_basic_frequencies(
head_size=head_dim,
rotary_dim=int(head_dim * self.config.rotary_pct),
base=self.config.rotary_emb_base,
)
def __call__(
self,
input_ids: tp.Optional[chex.Array] = None,
attention_mask: tp.Optional[chex.Array] = None,
position_ids: tp.Optional[chex.Array] = None,
past_key_values: tp.Optional[TransformerCache | PagedAttentionCache] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
inputs_embeds: tp.Optional[chex.Array] = None,
segment_ids: tp.Optional[chex.Array] = None,
extra_embedding: tp.Optional[chex.Array] = None,
output_attentions: bool = False,
output_hidden_states: bool = False,
return_dict: bool = True,
):
"""Forward pass through the GPTNeoXModel.
Args:
input_ids (chex.Array, optional): Input token IDs, shape (batch_size, sequence_length).
attention_mask (chex.Array, optional): Mask to avoid attention on padding tokens.
position_ids (chex.Array, optional): Indices of positions of each input sequence token.
past_key_values (TransformerCache | PagedAttentionCache, optional): Cache containing precomputed key/value states.
cache_metadata (TransformerMetadata | PagedAttentionMetadata, optional): Metadata for cache handling.
inputs_embeds (chex.Array, optional): Input embeddings, shape (batch_size, sequence_length, hidden_size).
segment_ids (chex.Array, optional): Segment token indices for segment embeddings.
extra_embedding (chex.Array, optional): Additional embedding to add to input embeddings.
output_attentions (bool, optional): Whether to return attention weights.
output_hidden_states (bool, optional): Whether to return hidden states of all layers.
return_dict (bool, optional): Whether to return a model output object or a tuple.
Returns:
Union[BaseModelOutput, Tuple]: Model outputs (last hidden state, optional hidden states, optional attentions)
"""
all_attentions = () if output_attentions else None
all_hidden_states = () if output_hidden_states else None
if (input_ids is None) ^ (inputs_embeds is not None):
raise ValueError(
"You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one"
)
if inputs_embeds is None:
inputs_embeds = self.embed_in(input_ids.astype("i4"))
batch_size, sequence_length, _ = inputs_embeds.shape
if attention_mask is None:
attention_mask = jnp.ones((batch_size, sequence_length), "b1")
else:
if attention_mask.dtype != jnp.bool:
attention_mask = jnp.astype(attention_mask == 1, "b1")
if position_ids is None:
position_ids = jnp.broadcast_to(
jnp.clip(jnp.cumsum(attention_mask, axis=-1) - 1, a_min=0),
(batch_size, sequence_length),
).astype(jnp.int32)
assert sequence_length <= self.config.max_position_embeddings, (
f"Maximum Position Embedding Reached ! (Excepted <= {self.config.max_position_embeddings} got {sequence_length})"
)
hidden_states = self.emb_dropout(
inputs_embeds + extra_embedding if extra_embedding is not None else inputs_embeds
)
if past_key_values is None:
past_key_values = TransformerCache.init_empty(len(self.layers))
for idx, block in enumerate(self.layers):
if output_hidden_states:
all_hidden_states += (hidden_states,)
hidden_states, attn_weight = block(
hidden_states=hidden_states,
attention_mask=attention_mask,
position_ids=position_ids,
cache_view=past_key_values.views[idx],
cache_metadata=cache_metadata,
segment_ids=segment_ids,
causal_mask=self.causal_mask,
frequencies=self.frequencies,
output_attentions=output_attentions,
)
if output_attentions:
all_attentions += (attn_weight,)
hidden_states = self.final_layer_norm(hidden_states)
if output_hidden_states:
all_hidden_states += (hidden_states,)
outputs = (
hidden_states,
all_hidden_states,
all_attentions,
)
if return_dict:
return BaseModelOutput(
last_hidden_state=hidden_states,
hidden_states=outputs[1],
attentions=outputs[2],
)
return tuple([v for v in outputs if v is not None])
[docs]@register_module(
TaskType.CAUSAL_LM,
config=GPTNeoXConfig,
model_type="gpt_neox",
)
class GPTNeoXForCausalLM(EasyDeLBaseModule):
"""GPT-NeoX model with a language modeling head.
This model extends the base GPTNeoXModel by adding a linear layer on top to
predict the next token in a sequence, making it suitable for causal language
modeling tasks.
Attributes:
config (GPTNeoXConfig): Configuration object for the model.
dtype (jnp.dtype): Data type for computations.
param_dtype (jnp.dtype): Data type for parameters.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations.
rngs (nn.Rngs): Random number generators.
"""
def __init__(
self,
config: GPTNeoXConfig,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: tp.Optional[tp.Union[str, jax.lax.Precision]] = None,
*,
rngs: nn.Rngs,
):
super().__init__(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.gpt_neox = GPTNeoXModel(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.lm_head = ParallelLinear(
config.hidden_size,
config.vocab_size,
use_bias=False,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
def __call__(
self,
input_ids,
attention_mask: tp.Optional[chex.Array] = None,
position_ids: tp.Optional[chex.Array] = None,
past_key_values: tp.Optional[TransformerCache | PagedAttentionCache] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
inputs_embeds: tp.Optional[chex.Array] = None,
segment_ids: tp.Optional[chex.Array] = None,
extra_embedding: tp.Optional[chex.Array] = None,
output_attentions: bool = False,
output_hidden_states: bool = False,
return_dict: bool = True,
):
"""Forward pass through the GPTNeoXForCausalLM model.
Args:
input_ids (chex.Array, optional): Input token IDs, shape (batch_size, sequence_length).
attention_mask (chex.Array, optional): Mask to avoid attention on padding tokens.
position_ids (chex.Array, optional): Indices of positions of each input sequence token.
past_key_values (TransformerCache | PagedAttentionCache, optional): Cache containing precomputed key/value states.
cache_metadata (TransformerMetadata | PagedAttentionMetadata, optional): Metadata for cache handling.
inputs_embeds (chex.Array, optional): Input embeddings, shape (batch_size, sequence_length, hidden_size).
segment_ids (chex.Array, optional): Segment token indices for segment embeddings.
extra_embedding (chex.Array, optional): Additional embedding to add to input embeddings.
output_attentions (bool, optional): Whether to return attention weights.
output_hidden_states (bool, optional): Whether to return hidden states of all layers.
return_dict (bool, optional): Whether to return a model output object or a tuple.
Returns:
Union[CausalLMOutput, Tuple]: Model outputs (logits, optional hidden states, optional attentions)
"""
outputs = self.gpt_neox(
input_ids=input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
past_key_values=past_key_values,
cache_metadata=cache_metadata,
inputs_embeds=inputs_embeds,
segment_ids=segment_ids,
extra_embedding=extra_embedding,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
hidden_states = outputs[0]
if self.config.tie_word_embeddings:
lm_logits = jax.lax.dot_general(
hidden_states,
self.gpt_neox.embed_in.embedding.value.T,
(((hidden_states.ndim - 1), (0,)), ((), ())),
)
else:
lm_logits = self.lm_head(hidden_states)
lm_logits = lm_logits.astype(jnp.float32)
if not return_dict:
return (lm_logits,) + outputs[1:]
return CausalLMOutput(
logits=lm_logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
past_key_values=outputs.past_key_values,
)