# 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 typing as tp
from functools import cached_property
import chex
import jax
import jax.numpy as jnp
from eformer import common_types
from eformer.escale import apply_logical_sharding
from flax import nnx as nn
from easydel.infra.base_module import EasyDeLBaseModule
from easydel.infra.factory import TaskType, register_module
from easydel.infra.loss_utils import auxiliary_load_balancing_loss_func
from easydel.infra.modeling_outputs import (
AttentionLayerOutput,
DecoderLayerOutput,
MoeCausalLMOutput,
MoeModelOutput,
)
from easydel.infra.utils import (
auto_remat,
block_wise_ffn,
get_dot_general_by_bits,
)
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 easydel.layers.norms import RMSNorm as FlaxGrok1RMSNorm
from .grok_1_configuration import Grok1Config
[docs]class Grok1Attention(AttentionModule):
"""Grok-1 Attention module.
This module implements the multi-head attention mechanism with rotary position embeddings
used in the Grok-1 model.
Attributes:
config (Grok1Config): Configuration object for the model.
layer_index (int): The index of the current layer.
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: Grok1Config,
layer_index: int,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__(config=config)
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
self.layer_index = layer_index
self.hidden_size = config.hidden_size
self.head_dim = self.config.hidden_size // self.config.num_attention_heads
self.num_key_value_groups = (
self.config.num_attention_heads // self.config.num_key_value_heads
)
if self.num_key_value_groups == 1:
assert self.config.num_attention_heads == self.config.num_key_value_heads
self.q_proj = ParallelLinear(
config.hidden_size,
config.num_attention_heads * self.head_dim,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.k_proj = ParallelLinear(
config.hidden_size,
config.num_key_value_heads * self.head_dim,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.v_proj = ParallelLinear(
config.hidden_size,
config.num_key_value_heads * self.head_dim,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.o_proj = ParallelLinear(
config.num_attention_heads * self.head_dim,
config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.rotary = self.config.get_basic_rope(
self.dtype,
self.head_dim,
self.head_dim,
True,
)
self.attention_performer = FlexibleAttentionModule(
base_config=config,
softmax_scale=self.head_dim**-0.5,
dropout_prob=config.attention_dropout,
)
self.resid_dropout = nn.Dropout(rate=config.resid_pdrop)
def _merge_heads(self, hidden_states):
"""
Merges the attention heads into a single hidden state tensor.
Args:
hidden_states (chex.Array): The hidden states with separate head dimensions.
Returns:
chex.Array: The hidden states with merged head dimensions.
"""
return hidden_states.reshape(hidden_states.shape[:2] + (self.hidden_size,))
def __call__(
self,
hidden_states: chex.Array,
attention_mask: chex.Array,
position_ids: chex.Array,
causal_mask: tp.Optional[chex.Array | bool],
mode: common_types.RUNTIME_MODE_TYPES, # type:ignore
cache_view: tp.Optional[TransformerCacheView | PagedAttentionCacheView] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
segment_ids: tp.Optional[chex.Array] = None,
output_attentions: bool = False,
fcm_mask: tp.Optional[chex.Array] = None,
frequencies: tp.Optional[chex.Array] = None,
):
"""Forward pass of the Grok1Attention 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.
cache_view (tp.Optional[TransformerCacheView | PagedAttentionCacheView], optional): Cache view for key/value states.
cache_metadata (tp.Optional[TransformerMetadata | PagedAttentionMetadata], optional): Metadata for cache handling.
segment_ids (tp.Optional[chex.Array], optional): Segment IDs for segment-based attention.
output_attentions (bool, optional): Whether to return attention weights.
fcm_mask (tp.Optional[chex.Array], optional): Forward causal mask.
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.
"""
batch_size, sequence_length = hidden_states.shape[:2]
query_states, key_states, value_states = (
self.q_proj(hidden_states),
self.k_proj(hidden_states),
self.v_proj(hidden_states),
)
query_states = query_states.reshape(
batch_size,
sequence_length,
self.config.num_attention_heads,
self.head_dim,
)
key_states = key_states.reshape(
batch_size,
sequence_length,
self.config.num_key_value_heads,
self.head_dim,
)
value_states = value_states.reshape(
batch_size,
sequence_length,
self.config.num_key_value_heads,
self.head_dim,
)
(
query_states,
key_states,
value_states,
) = self.apply_qkv_shardings(query_states, key_states, value_states)
query_states, key_states = self.rotary(
positions=position_ids,
query=query_states,
key=key_states,
frequencies=frequencies,
)
(
key_states,
value_states,
attention_mask,
init_attention_bias,
cache_view,
) = self.concatenate(
query=query_states,
key=key_states,
value=value_states,
cache_view=cache_view,
cache_metadata=cache_metadata,
attention_mask=attention_mask,
causal_mask=causal_mask,
fcm_mask=fcm_mask,
)
attentions = self.attention_performer.forward(
query_states=query_states,
key_states=key_states,
value_states=value_states,
mode=mode,
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.o_proj(attn_output)
attn_output = self.resid_dropout(attn_output)
return AttentionLayerOutput(
attention_output=attn_output,
attention_weight=attentions.attention_weights if output_attentions else None,
cache_view=cache_view,
)
[docs]class Grok1BLockSparseMLP(nn.Module):
"""Grok-1 Block Sparse MLP module.
This module implements the specific MLP structure used within the sparse Mixture of Experts
layer in the Grok-1 model.
Attributes:
config (Grok1Config): 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: Grok1Config,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__()
self.config = config
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
self.linear = ParallelLinear(
config.hidden_size,
config.intermediate_size,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.linear_1 = ParallelLinear(
config.intermediate_size,
config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.linear_v = ParallelLinear(
config.hidden_size,
config.intermediate_size,
dtype=dtype,
param_dtype=param_dtype,
use_bias=False,
kernel_init=jax.nn.initializers.normal(config.initializer_range),
precision=self.precision,
rngs=rngs,
**get_dot_general_by_bits(config.bits, config.easy_method),
)
def __call__(self, hidden_states: jnp.ndarray) -> jnp.ndarray:
"""Forward pass of the Grok1BLockSparseMLP module.
Args:
hidden_states (chex.Array): Input hidden states.
Returns:
chex.Array: Output hidden states after processing through the block sparse MLP.
"""
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
hidden_states = self.linear_1(
nn.gelu(self.linear(hidden_states)) * self.linear_v(hidden_states)
)
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
return hidden_states
[docs]class Grok1SparseMoeBlock(nn.Module):
"""Grok-1 Sparse Mixture of Experts (MoE) block.
This module implements the sparse MoE layer used in Grok-1. It routes tokens
to a subset of experts based on learned gating weights.
Attributes:
config (Grok1Config): 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: Grok1Config,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__()
self.config = config
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
self.gate = ParallelLinear(
self.config.hidden_size,
self.config.num_experts,
use_bias=False,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
kernel_init=nn.initializers.normal(),
)
self.experts = [
Grok1BLockSparseMLP(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
for i in range(self.config.num_experts)
]
def __call__(self, hidden_states: chex.Array) -> tp.Tuple[chex.Array, chex.Array]:
"""Forward pass of the Grok1SparseMoeBlock.
Args:
hidden_states (chex.Array): Input hidden states.
Returns:
tp.Tuple[chex.Array, chex.Array]: A tuple containing the output hidden states
and the router logits.
"""
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
router_logits = self.gate(hidden_states).astype(
jnp.promote_types(self.dtype, jnp.float32)
)
routing_weights, selected_experts = jax.lax.top_k(
router_logits, k=self.config.num_experts_per_tok
)
routing_weights = jax.nn.softmax(
routing_weights.astype(jnp.promote_types(self.dtype, jnp.float32)), axis=-1
)
final_hidden_state = jnp.zeros_like(hidden_states)
for index in range(self.config.num_experts):
expert_layer_output = (
block_wise_ffn(
self.layers[index],
hidden_states,
self.config.scan_mlp_chunk_size,
)
if self.config.use_scan_mlp
else self.layers[index](hidden_states)
)
expert_layer_output_exp = (
jnp.sum(
jnp.multiply(
selected_experts == index,
routing_weights,
),
axis=-1,
)[:, :, None]
* expert_layer_output
)
final_hidden_state += expert_layer_output_exp
return (final_hidden_state, router_logits)
[docs]class Grok1DecoderLayer(nn.Module):
"""Grok-1 Transformer Decoder Layer.
This module represents a single decoder layer in the Grok-1 model,
combining self-attention and a sparse MoE block with residual connections
and layer normalization.
Attributes:
config (Grok1Config): 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: Grok1Config,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__()
self.config = config
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
attn_block = Grok1Attention
mlp_block = Grok1SparseMoeBlock
attn_block, mlp_block = auto_remat(
attn_block,
mlp_block,
policy=config.gradient_checkpointing,
)
self.attn = attn_block(
config=self.config,
layer_index=self.layer_index,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.moe_block = mlp_block(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.pre_attn_norm = FlaxGrok1RMSNorm(
dim=self.config.hidden_size,
eps=self.config.rms_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.post_attn_norm = FlaxGrok1RMSNorm(
dim=self.config.hidden_size,
eps=self.config.rms_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.pre_moe_norm = FlaxGrok1RMSNorm(
dim=self.config.hidden_size,
eps=self.config.rms_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.post_moe_norm = FlaxGrok1RMSNorm(
dim=self.config.hidden_size,
eps=self.config.rms_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
def __call__(
self,
hidden_states: chex.Array,
attention_mask: chex.Array,
position_ids: chex.Array,
causal_mask: tp.Optional[chex.Array | bool],
mode: common_types.RUNTIME_MODE_TYPES, # type:ignore
cache_view: tp.Optional[TransformerCacheView | PagedAttentionCacheView] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
segment_ids: tp.Optional[chex.Array] = None,
output_attentions: bool = False,
output_router_logits: bool = False,
fcm_mask: tp.Optional[chex.Array] = None,
frequencies: tp.Optional[chex.Array] = None,
) -> DecoderLayerOutput:
"""Forward pass of the Grok1DecoderLayer 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.
cache_view (tp.Optional[TransformerCacheView | PagedAttentionCacheView], optional): Cache view for key/value states.
cache_metadata (tp.Optional[TransformerMetadata | PagedAttentionMetadata], optional): Metadata for cache handling.
segment_ids (tp.Optional[chex.Array], optional): Segment IDs for segment-based attention.
output_attentions (bool, optional): Whether to return attention weights. Defaults to False.
output_router_logits (bool, optional): Whether to return router logits from the MoE layer. Defaults to False.
fcm_mask (tp.Optional[chex.Array], optional): Forward causal mask. Defaults to None.
frequencies (tp.Optional[chex.Array], optional): Precomputed rotary frequencies.
Returns:
tp.Tuple[chex.Array, tp.Optional[chex.Array], tp.Optional[chex.Array]]: A tuple containing the
output hidden states, optionally the attention weights, and optionally the router logits.
"""
residual = hidden_states
hidden_states = self.pre_attn_norm(hidden_states)
attn_outputs = self.attn(
hidden_states,
attention_mask,
position_ids,
causal_mask,
mode,
cache_view,
cache_metadata,
segment_ids,
output_attentions,
fcm_mask,
frequencies,
)
hidden_states = attn_outputs.attention_output
hidden_states = self.post_attn_norm(hidden_states)
hidden_states = residual + hidden_states
residual = hidden_states
hidden_states = self.pre_moe_norm(hidden_states)
hidden_states, router_logits = self.moe_block(hidden_states)
hidden_states = self.post_moe_norm(hidden_states)
hidden_states = residual + hidden_states
return DecoderLayerOutput(
hidden_states=hidden_states,
attention_weight=attn_outputs.attention_weight,
router_logits=router_logits if output_router_logits else None,
cache_view=attn_outputs.cache_view,
)
[docs]@register_module(
TaskType.BASE_MODULE,
config=Grok1Config,
model_type="grok-1",
)
class Grok1Model(EasyDeLBaseModule):
"""Grok-1 model implementation.
This class implements the main Grok-1 transformer model architecture, consisting of
an embedding layer, multiple Grok1DecoderLayer layers (with sparse MoE), and a final
RMS normalization layer.
Attributes:
config (Grok1Config): 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: Grok1Config,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.embed_tokens = nn.Embed(
self.config.vocab_size,
self.config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.layers = [
Grok1DecoderLayer(
layer_index=layer_index,
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
for layer_index in range(self.config.num_hidden_layers)
]
self.norm = FlaxGrok1RMSNorm(
self.config.hidden_size,
eps=self.config.rms_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
@cached_property
def frequencies(self):
return self.config.get_basic_frequencies(
head_size=self.config.hidden_size // self.config.num_attention_heads,
base=self.config.rope_theta,
)
def __call__(
self,
input_ids: tp.Optional[chex.Array] = None,
inputs_embeds: tp.Optional[chex.Array] = None,
attention_mask: tp.Optional[chex.Array] = None,
position_ids: tp.Optional[chex.Array] = None,
segment_ids: tp.Optional[chex.Array] = None,
output_attentions: tp.Optional[bool] = None,
output_hidden_states: tp.Optional[bool] = None,
output_router_logits: tp.Optional[bool] = None,
mode: tp.Optional[common_types.RUNTIME_MODE_TYPES] = None, # type:ignore
past_key_values: tp.Optional[TransformerCache | PagedAttentionCache] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
) -> MoeModelOutput:
"""Forward pass through the Grok1Model.
Args:
input_ids (chex.Array, optional): Input token IDs, shape (batch_size, sequence_length).
inputs_embeds (chex.Array, optional): Input embeddings, shape (batch_size, sequence_length, hidden_size).
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.
segment_ids (chex.Array, optional): Segment token indices for segment embeddings.
output_attentions (bool, optional): Whether to return attention weights.
output_hidden_states (bool, optional): Whether to return hidden states of all layers.
output_router_logits (bool, optional): Whether to return router logits from MoE layers.
past_key_values (TransformerCache | PagedAttentionCache, optional): Cache containing precomputed key/value states.
cache_metadata (TransformerMetadata | PagedAttentionMetadata, optional): Metadata for cache handling.
Returns:
MoeModelOutput: Model outputs (last hidden state, optional hidden states, optional attentions, optional router logits)
"""
output_attentions = (
output_attentions
if output_attentions is not None
else self.config.output_attentions
)
output_router_logits = (
output_router_logits
if output_router_logits is not None
else self.config.output_router_logits
)
output_hidden_states = (
output_hidden_states
if output_hidden_states is not None
else self.config.output_hidden_states
)
all_hidden_states = () if output_hidden_states else None
all_self_attns = () if output_attentions else None
all_router_logits = () if output_router_logits else None
if output_router_logits is None:
output_router_logits = self.config.output_router_logits
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.wte(input_ids.astype("i4"))
batch_size, sequence_length = inputs_embeds.shape[:2]
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)
hidden_states = inputs_embeds
if mode is None:
mode = (
common_types.MODE_DECODE
if sequence_length == 1 and past_key_values is not None
else common_types.MODE_TRAIN
)
if past_key_values is None:
past_key_values = TransformerCache.init_empty(len(self.layers))
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
for idx, block in enumerate(self.layers):
if output_hidden_states:
all_hidden_states += (hidden_states,)
layer_outputs = block(
hidden_states=hidden_states,
attention_mask=attention_mask,
position_ids=position_ids,
output_attentions=output_attentions,
output_router_logits=output_router_logits,
mode=mode,
cache_view=past_key_values.view[idx],
cache_metadata=cache_metadata,
frequencies=self.frequencies,
causal_mask=self.causal_mask,
segment_ids=segment_ids,
)
hidden_states = layer_outputs.hidden_states
if output_attentions:
all_self_attns += (layer_outputs.attention_weight,)
if output_router_logits:
all_router_logits += (layer_outputs.router_logits,)
past_key_values[idx] = layer_outputs.cache_view
hidden_states = self.norm(hidden_states)
if output_hidden_states:
all_hidden_states += (hidden_states,)
return MoeModelOutput(
last_hidden_state=hidden_states,
hidden_states=all_hidden_states,
attentions=all_self_attns,
router_logits=all_router_logits,
past_key_values=past_key_values,
)
[docs]@register_module(
TaskType.CAUSAL_LM,
config=Grok1Config,
model_type="grok-1",
)
class Grok1ForCausalLM(EasyDeLBaseModule):
"""Grok-1 model with a language modeling head.
This model extends the base Grok1Model by adding a linear layer on top to
predict the next token in a sequence, making it suitable for causal language
modeling tasks. It also includes handling for the Mixture of Experts auxiliary loss.
Attributes:
config (Grok1Config): 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: Grok1Config,
dtype: jnp.dtype = jnp.float32,
param_dtype: jnp.dtype = jnp.float32,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.model = Grok1Model(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.lm_head = ParallelLinear(
config.hidden_size,
config.vocab_size,
dtype=self.dtype,
rngs=rngs,
param_dtype=self.param_dtype,
precision=self.precision,
use_bias=False,
kernel_init=nn.initializers.normal(config.initializer_range),
**get_dot_general_by_bits(config.bits, config.easy_method),
)
self.output_multiplier_scale = self.config.output_multiplier_scale
def __call__(
self,
input_ids: tp.Optional[chex.Array] = None,
inputs_embeds: tp.Optional[chex.Array] = None,
attention_mask: tp.Optional[chex.Array] = None,
position_ids: tp.Optional[chex.Array] = None,
segment_ids: tp.Optional[chex.Array] = None,
output_attentions: tp.Optional[bool] = None,
output_hidden_states: tp.Optional[bool] = None,
output_router_logits: tp.Optional[bool] = None,
mode: tp.Optional[common_types.RUNTIME_MODE_TYPES] = None, # type:ignore
past_key_values: tp.Optional[TransformerCache | PagedAttentionCache] = None,
cache_metadata: tp.Optional[TransformerMetadata | PagedAttentionMetadata] = None,
) -> MoeCausalLMOutput | tp.Tuple:
"""Forward pass through the Grok1ForCausalLM model.
Args:
input_ids (chex.Array, optional): Input token IDs, shape (batch_size, sequence_length).
inputs_embeds (chex.Array, optional): Input embeddings, shape (batch_size, sequence_length, hidden_size).
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.
segment_ids (chex.Array, optional): Segment token indices for segment embeddings.
output_attentions (bool, optional): Whether to return attention weights.
output_hidden_states (bool, optional): Whether to return hidden states of all layers.
output_router_logits (bool, optional): Whether to return router logits from MoE layers.
past_key_values (TransformerCache | PagedAttentionCache, optional): Cache containing precomputed key/value states.
cache_metadata (TransformerMetadata | PagedAttentionMetadata, optional): Metadata for cache handling.
Returns:
MoeCausalLMOutput: Model outputs (logits, optional auxiliary loss, optional hidden states, optional attentions, optional router logits)
"""
if output_router_logits is None:
output_router_logits = self.config.output_router_logits
outputs = self.model(
input_ids=input_ids,
attention_mask=attention_mask,
position_ids=position_ids,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
output_router_logits=output_router_logits,
mode=mode,
past_key_values=past_key_values,
cache_metadata=cache_metadata,
segment_ids=segment_ids,
)
logits = self.lm_head(outputs.last_hidden_state)
logits = logits * self.output_multiplier_scale
aux_loss = None
if output_router_logits and outputs.router_logits is not None:
aux_loss = auxiliary_load_balancing_loss_func(
gate_logits=outputs.router_logits,
num_experts=self.num_experts,
top_k=self.num_experts_per_tok,
attention_mask=attention_mask,
)
aux_loss += aux_loss * self.config.router_aux_loss_coef
return MoeCausalLMOutput(
aux_loss=aux_loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
router_logits=outputs.router_logits,
past_key_values=outputs.past_key_values,
)