# Copyright 2025 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
from typing import ClassVar
import jax.lax
from chex import Array
from eformer import common_types
from eformer.escale import apply_logical_sharding
from ejkernel.types import MaskInfo
from flax import nnx as nn
from jax import numpy as jnp
from jax.ad_checkpoint import checkpoint_name
from jaxtyping import Bool, Float, Int
from easydel.infra.base_module import EasyDeLBaseModule
from easydel.infra.factory import TaskType, register_module
from easydel.infra.modeling_outputs import BaseModelOutput, CausalLMOutput, DecoderLayerOutput
from easydel.infra.utils import ACT2FN, auto_remat, block_wise_ffn
from easydel.layers.attention_unified import UnifiedAttention
from easydel.layers.base_modules import BaseCausalLMModule
from easydel.layers.caching import (
RaggedPagesCache,
RaggedPagesCacheView,
RaggedPagesMetadata,
TransformerCache,
TransformerCacheView,
TransformerMetadata,
)
from easydel.layers.linear import ColumnParallelLinear, RowParallelLinear
from .phi_configuration import PhiConfig
[docs]class PhiMLP(nn.Module):
"""Phi MLP module.
This module implements the feed-forward network (MLP) used in the Phi model.
It consists of two linear projections with a GELU activation in between.
Attributes:
config (PhiConfig): Configuration object for the model.
layer_idx (int, optional): 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.
fc1 (ParallelLinear): First linear projection layer (up-projection).
fc2 (ParallelLinear): Second linear projection layer (down-projection).
act (callable): Activation function.
"""
def __init__(
self,
config: PhiConfig,
layer_idx: int | None = None,
dtype: jnp.dtype = jnp.bfloat16,
param_dtype: jnp.dtype = jnp.bfloat16,
precision: jax.lax.Precision | None = None,
*,
rngs: nn.Rngs,
):
"""Initializes the PhiMLP module.
Args:
config (PhiConfig): The configuration object for the Phi model.
layer_idx (int, optional): Index of the current layer. Defaults to None.
dtype (jnp.dtype): Data type for computation. Defaults to jnp.float32.
param_dtype (jnp.dtype): Data type for parameters. Defaults to jnp.float32.
precision (jax.lax.PrecisionLike, optional): Precision setting for JAX operations. Defaults to None.
rngs (nn.Rngs): Random number generators.
"""
self.config = config
self.layer_idx = layer_idx
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
self.fc1 = ColumnParallelLinear(
config.hidden_size,
config.intermediate_size,
kernel_init=nn.initializers.normal(config.initializer_range),
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.fc2 = RowParallelLinear(
config.intermediate_size,
config.hidden_size,
kernel_init=nn.initializers.normal(config.initializer_range),
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.act = ACT2FN[self.config.hidden_act]
def __call__(
self, hidden_states: Float[Array, "batch seq_len hidden_dim"]
) -> Float[Array, "batch seq_len hidden_dim"]:
"""Forward pass of the PhiMLP module.
Args:
hidden_states: Input hidden states.
Returns:
Output hidden states after MLP transformation.
"""
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
gate = checkpoint_name(self.act(self.fc1(hidden_states)), "mlp_gate")
hidden_states = checkpoint_name(self.fc2(gate), "mlp_down")
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
return checkpoint_name(hidden_states, "mlp_output")
[docs]class PhiAttention(UnifiedAttention):
"""Phi Attention with Q/K normalization.
Inherits Q/K normalization from QKNormAttention.
Features:
- Uses LayerNorm instead of RMSNorm
- Standard LayerNorm on full hidden_size (not per-head)
- Partial RoPE (partial_rotary_factor)
- Custom bias configuration
"""
norms_mapping: ClassVar[dict[str, str]] = {
"query_normalization": "q_layernorm",
"key_normalization": "k_layernorm",
}
projection_mapping: ClassVar[dict[str, str]] = {
"query_projection": "q_proj",
"key_projection": "k_proj",
"value_projection": "v_proj",
"output_projection": "dense",
}
def __init__(
self,
config: PhiConfig,
layer_idx: int | None = None,
dtype: jnp.dtype = jnp.bfloat16,
param_dtype: jnp.dtype = jnp.bfloat16,
precision: jax.lax.Precision | None = None,
*,
rngs: nn.Rngs,
):
self.qk_layernorm = config.qk_layernorm
config.attention_bias = True
super().__init__(
config,
dtype,
param_dtype,
precision,
rngs=rngs,
layer_idx=layer_idx if layer_idx is not None else -1,
attention_type="standard",
causal=True,
use_qk_norm=config.qk_layernorm,
)
self.layer_idx = layer_idx
self.attention_dropout = config.attention_dropout
self.partial_rotary_factor = config.partial_rotary_factor
self.rotary_emb_dim = int(config.partial_rotary_factor * self.head_dim)
self.is_causal = True
def _create_q_norm(self, config, dtype, param_dtype, rngs):
"""Override to use standard LayerNorm on hidden_size if qk_layernorm is enabled."""
return nn.LayerNorm(
config.hidden_size,
epsilon=config.layer_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
use_bias=True,
rngs=rngs,
)
def _create_k_norm(self, config, dtype, param_dtype, rngs):
"""Override to use standard LayerNorm on hidden_size if qk_layernorm is enabled."""
return nn.LayerNorm(
config.hidden_size,
epsilon=config.layer_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
use_bias=True,
rngs=rngs,
)
def _create_rotary(self, config, dtype):
"""Override for partial RoPE."""
return config.get_basic_rope(
dtype,
head_size=int(config.partial_rotary_factor * (config.hidden_size // config.num_attention_heads)),
rotary_dim=int(config.partial_rotary_factor * (config.hidden_size // config.num_attention_heads)),
)
def _preprocess_qkv(self, query_states, key_states, value_states):
if self.use_qk_norm:
return self.query_normalization(query_states), self.key_normalization(key_states), value_states
return query_states, key_states, value_states
[docs]class PhiDecoderLayer(nn.Module):
"""Phi Transformer Decoder Layer.
This module represents a single decoder layer in the Phi model,
combining self-attention and MLP sub-layers with residual connections
and layer normalization.
Attributes:
config (PhiConfig): Configuration object for the model.
layer_idx (int, optional): 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.
input_layernorm (nn.LayerNorm): Layer normalization applied before the attention and MLP blocks.
resid_dropout (nn.Dropout): Dropout applied to the residual connection after the MLP block.
self_attn (PhiAttention): The self-attention module.
mlp (PhiMLP): The feed-forward (MLP) module.
"""
def __init__(
self,
config: PhiConfig,
layer_idx: int | None = None,
dtype: jnp.dtype = jnp.bfloat16,
param_dtype: jnp.dtype = jnp.bfloat16,
precision: jax.lax.Precision | None = None,
*,
rngs: nn.Rngs,
):
"""Initializes the PhiDecoderLayer.
Args:
config (PhiConfig): The configuration object for the Phi model.
layer_idx (int, optional): Index of the current layer. Defaults to None.
dtype (jnp.dtype): Data type for computation. Defaults to jnp.float32.
param_dtype (jnp.dtype): Data type for parameters. Defaults to jnp.float32.
precision (jax.lax.PrecisionLike, optional): Precision setting for JAX operations. Defaults to None.
rngs (nn.Rngs): Random number generators.
"""
self.config = config
self.layer_idx = layer_idx
self.dtype = dtype
self.param_dtype = param_dtype
self.precision = precision
self.rngs = rngs
attn_block = PhiAttention
mlp_block = PhiMLP
attn_block, mlp_block = auto_remat(
attn_block,
mlp_block,
policy=config.gradient_checkpointing,
save_names=config.gradient_checkpointing_targets,
exclude_names=config.gradient_checkpointing_targets,
)
self.self_attn = attn_block(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
layer_idx=layer_idx,
)
self.mlp = mlp_block(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
layer_idx=layer_idx,
)
self.input_layernorm = nn.LayerNorm(
config.hidden_size,
epsilon=config.layer_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.resid_dropout = nn.Dropout(self.config.resid_pdrop)
def __call__(
self,
hidden_states: Float[Array, "batch seq_len hidden_dim"],
mask_info: MaskInfo | None,
position_ids: Int[Array, "batch seq_len"],
mode: common_types.RUNTIME_MODE_TYPES, # type:ignore
cache_view: TransformerCacheView | RaggedPagesCacheView | None = None,
cache_metadata: TransformerMetadata | RaggedPagesMetadata | None = None,
output_attentions: bool = False,
frequencies: Float[Array, "seq_len head_dim"] | None = None,
):
"""Forward pass of the PhiDecoderLayer 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. Shape: (batch_size, sequence_length).
causal_mask (tp.Optional[chex.Array | bool]): Causal mask for ensuring autoregressive behavior.
cache_view (tp.Optional[TransformerCacheView | RaggedPagesCacheView]): Cache view for attention KVs.
cache_metadata (tp.Optional[TransformerMetadata | RaggedPagesMetadata]): Metadata for paged attention.
segment_ids (tp.Optional[chex.Array]): Segment IDs for segment-based attention (optional).
output_attentions (bool): Whether to return attention weights. Default is False.
fcm_mask (tp.Optional[chex.Array]): Flash Chunking Mask (FCM) for attention.
frequencies (tp.Optional[chex.Array]): Precomputed rotary frequency embeddings.
Returns:
tp.Tuple[chex.Array, tp.Optional[chex.Array]]:
A tuple containing the output hidden states and optionally the attention weights.
"""
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
attn_outputs = self.self_attn(
hidden_states,
mask_info,
position_ids,
mode,
cache_view,
cache_metadata,
output_attentions,
frequencies,
)
if self.config.use_scan_mlp:
feed_forward_hidden_states = block_wise_ffn(
self.mlp,
hidden_states,
self.config.scan_mlp_chunk_size,
)
else:
feed_forward_hidden_states = self.mlp(hidden_states)
feed_forward_hidden_states = self.resid_dropout(feed_forward_hidden_states)
hidden_states = checkpoint_name(
self.resid_dropout(attn_outputs.attention_output) + feed_forward_hidden_states + residual, "residual"
)
hidden_states = checkpoint_name(hidden_states, "layer_output")
hidden_states = apply_logical_sharding(
hidden_states,
dynamic_axes=common_types.HiddenStateSharding,
partition_manager=self.config.partition_manager,
)
return DecoderLayerOutput(
hidden_states=hidden_states,
attention_weight=attn_outputs.attention_weight,
cache_view=attn_outputs.cache_view,
)
[docs]@register_module(TaskType.BASE_MODULE, config=PhiConfig, model_type="phi")
class PhiModel(EasyDeLBaseModule):
"""The base Phi model transformer.
This class represents the core transformer architecture of the Phi model,
consisting of an embedding layer, multiple PhiDecoderLayer layers,
and a final layer normalization.
Attributes:
config (PhiConfig): Configuration object for the model.
dtype (jnp.dtype): Data type for computation.
param_dtype (jnp.dtype): Data type for parameters.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations.
rngs (nn.Rngs): Random number generators.
embed_tokens (nn.Embed): Embedding layer for input tokens.
layers (tp.List[PhiDecoderLayer]): List of decoder layers.
final_layernorm (nn.LayerNorm): Final layer normalization.
embed_dropout (nn.Dropout): Dropout layer applied after embeddings.
gradient_checkpointing (EasyDeLGradientCheckPointers): Gradient checkpointing configuration.
"""
def __init__(
self,
config: PhiConfig,
dtype: jnp.dtype = jnp.bfloat16,
param_dtype: jnp.dtype = jnp.bfloat16,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
"""Initializes the PhiModel.
Args:
config (PhiConfig): The configuration object for the Phi model.
dtype (jnp.dtype): Data type for computation. Defaults to jnp.float32.
param_dtype (jnp.dtype): Data type for parameters. Defaults to jnp.float32.
precision (jax.lax.PrecisionLike): Precision setting for JAX operations. Defaults to None.
rngs (nn.Rngs): Random number generators.
"""
super().__init__(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
)
self.padding_idx = config.pad_token_id
self.vocab_size = config.vocab_size
embed_block = auto_remat(
nn.Embed,
policy=config.gradient_checkpointing,
save_names=config.gradient_checkpointing_targets,
exclude_names=config.gradient_checkpointing_targets,
)
self.embed_tokens = embed_block(
config.vocab_size,
config.hidden_size,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
self.embed_dropout = nn.Dropout(config.embd_pdrop, rngs=rngs)
self.layers = [
PhiDecoderLayer(
config=config,
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
layer_idx=idx,
rngs=rngs,
)
for idx in range(self.config.num_hidden_layers)
]
self.final_layernorm = nn.LayerNorm(
config.hidden_size,
epsilon=config.layer_norm_eps,
dtype=dtype,
param_dtype=param_dtype,
rngs=rngs,
)
@functools.cached_property
def frequencies(self):
return self.config.get_basic_frequencies(
head_size=int(
self.config.partial_rotary_factor * (self.config.hidden_size // self.config.num_attention_heads)
),
rotary_dim=int(
self.config.partial_rotary_factor * (self.config.hidden_size // self.config.num_attention_heads)
),
)
def __call__(
self,
input_ids: Int[Array, "batch seq_len"] | None = None,
inputs_embeds: Float[Array, "batch seq_len hidden_dim"] | None = None,
attention_mask: Bool[Array, "batch seq_len"] | None = None,
mask_info: MaskInfo | None = None,
position_ids: Int[Array, "batch seq_len"] | None = None,
mode: common_types.RUNTIME_MODE_TYPES | None = None, # type:ignore
past_key_values: TransformerCache | RaggedPagesCache | None = None,
cache_metadata: TransformerMetadata | RaggedPagesMetadata | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
) -> BaseModelOutput:
"""Forward pass of the PhiModel.
Args:
input_ids (tp.Optional[chex.Array]): Input token IDs. Shape: (batch_size, sequence_length).
inputs_embeds (tp.Optional[chex.Array]): Input embeddings.
Either `input_ids` or `inputs_embeds` must be provided.
attention_mask (tp.Optional[chex.Array]): Mask to avoid performing attention on padding token indices.
Shape: (batch_size, sequence_length).
position_ids (tp.Optional[chex.Array]): Position indices for the tokens.
Shape: (batch_size, sequence_length).
segment_ids (tp.Optional[chex.Array]): Segment IDs (unused).
output_attentions (tp.Optional[bool]): Whether to return attention weights.
Defaults to `config.output_attentions`.
output_hidden_states (tp.Optional[bool]): Whether to return hidden states for all layers.
Defaults to `config.output_hidden_states`.
past_key_values (tp.Optional[TransformerCache | RaggedPagesCache]):
Precomputed key/value states for attention.
cache_metadata (tp.Optional[TransformerMetadata | RaggedPagesMetadata]): Metadata for paged attention.
Returns:
BaseModelOutput: The model's output.
returns a `BaseModelOutput` object containing `last_hidden_state`, `hidden_states` (optional),
and `attentions` (optional).
Raises:
ValueError: If neither `input_ids` nor `inputs_embeds` is provided.
"""
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 = checkpoint_name(self.embed_tokens(input_ids.astype("i4")), "embeddings")
sequence_length = inputs_embeds.shape[1]
all_attentions = () if output_attentions else None
all_hidden_states = () if output_hidden_states else None
assert sequence_length <= self.config.max_position_embeddings, (
f"Maximum Position Embedding Reached ! "
f"(Excepted <= {self.config.max_position_embeddings} got {sequence_length})"
)
mask_info = MaskInfo.dynamic_init(
mask_info=mask_info,
input_ids=input_ids,
inputs_embeds=inputs_embeds,
attention_mask=attention_mask,
)
if position_ids is None:
position_ids = mask_info.q_position_ids
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,
mask_info=mask_info,
position_ids=position_ids,
mode=mode,
cache_view=past_key_values.views[idx],
cache_metadata=cache_metadata,
output_attentions=output_attentions,
frequencies=self.frequencies,
)
hidden_states = layer_outputs.hidden_states
if output_attentions:
all_attentions += (layer_outputs.attention_weight,)
past_key_values[idx] = layer_outputs.cache_view
hidden_states = self.final_layernorm(hidden_states)
hidden_states = checkpoint_name(hidden_states, "model_output")
if output_hidden_states:
all_hidden_states += (hidden_states,)
return BaseModelOutput(
last_hidden_state=hidden_states,
hidden_states=all_hidden_states,
attentions=all_attentions,
past_key_values=past_key_values,
)
[docs] def get_encoder(self):
"""
Returns the encoder part of the model's graph definition.
Decoder-Only models don't have an encoder.
"""
raise NotImplementedError("This is a decoder-only model and does not have an encoder.")
[docs] def get_decoder(self):
"""
Returns the decoder part of the model's graph definition.
"""
return self
[docs] def get_lm_head(self):
"""
Returns the language model head of the module.
Base Models don't have a Language Model Head.
"""
raise NotImplementedError("The base model does not have a language model head.")
[docs] def get_embedding(self):
"""
Returns the embedding layer of the module.
"""
return self.embed_tokens
[docs]@register_module(TaskType.CAUSAL_LM, config=PhiConfig, model_type="phi")
class PhiForCausalLM(BaseCausalLMModule[PhiModel, PhiConfig]):
"""Phi model with a Causal Language Modeling head."""
_task_type = TaskType.CAUSAL_LM
_model_type = "phi"
_config_class = PhiConfig
def __init__(
self,
config: PhiConfig,
dtype: jnp.dtype = jnp.bfloat16,
param_dtype: jnp.dtype = jnp.bfloat16,
precision: jax.lax.PrecisionLike = None,
*,
rngs: nn.Rngs,
):
super().__init__(
config=config,
base_model_class=PhiModel,
base_model_name="model",
dtype=dtype,
param_dtype=param_dtype,
precision=precision,
rngs=rngs,
lm_head_bias=True,
)
def __call__(
self,
input_ids: Int[Array, "batch seq_len"] | None = None,
inputs_embeds: Float[Array, "batch seq_len hidden_dim"] | None = None,
attention_mask: Bool[Array, "batch seq_len"] | None = None,
mask_info: MaskInfo | None = None,
position_ids: Int[Array, "batch seq_len"] | None = None,
output_attentions: bool | None = None,
output_hidden_states: bool | None = None,
mode: common_types.RUNTIME_MODE_TYPES | None = None, # type:ignore
past_key_values: TransformerCache | RaggedPagesCache | None = None,
cache_metadata: TransformerMetadata | RaggedPagesMetadata | None = None,
apply_lm_head: bool = True,
) -> CausalLMOutput:
"""Forward pass of the PhiForCausalLM model.
Args:
input_ids (tp.Optional[chex.Array]): Input token IDs. Shape: (batch_size, sequence_length).
inputs_embeds (tp.Optional[chex.Array]): Input embeddings.
Either `input_ids` or `inputs_embeds` must be provided.
attention_mask (tp.Optional[chex.Array]): Mask to avoid performing attention on padding token indices.
Shape: (batch_size, sequence_length).
position_ids (tp.Optional[chex.Array]): Position indices for the tokens.
Shape: (batch_size, sequence_length).
segment_ids (tp.Optional[chex.Array]): Segment IDs (unused).
output_attentions (tp.Optional[bool]): Whether to return attention weights.
Defaults to `config.output_attentions`.
output_hidden_states (tp.Optional[bool]): Whether to return hidden states for all layers.
Defaults to `config.output_hidden_states`.
past_key_values (tp.Optional[TransformerCache | RaggedPagesCache]):
Precomputed key/value states for attention.
cache_metadata (tp.Optional[TransformerMetadata | RaggedPagesMetadata]): Metadata for paged attention.
Returns:
CausalLMOutput: The model's output.
returns a `CausalLMOutput` object containing `logits`, `hidden_states` (optional),
and `attentions` (optional).
"""
outputs = self.model(
input_ids=input_ids,
attention_mask=attention_mask,
mask_info=mask_info,
position_ids=position_ids,
mode=mode,
past_key_values=past_key_values,
cache_metadata=cache_metadata,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
inputs_embeds=inputs_embeds,
)
hidden_states = outputs.last_hidden_state
lm_logits = None
if apply_lm_head:
lm_logits = checkpoint_name(self.apply_lm_head(hidden_states), "lm_head_output")
return CausalLMOutput(
logits=lm_logits,
hidden_states=outputs.hidden_states,
last_hidden_state=outputs.last_hidden_state,
attentions=outputs.attentions,
past_key_values=outputs.past_key_values,
)
[docs] def get_encoder(self):
"""
Returns the encoder part of the model's graph definition.
Decoder-Only models don't have an encoder.
"""
raise NotImplementedError("This is a decoder-only model and does not have an encoder.")
[docs] def get_decoder(self):
"""
Returns the decoder part of the model's graph definition.
"""
return self.model.get_decoder()
[docs] def get_lm_head(self):
"""
Returns the language model head of the module.
"""
return self.lm_head
[docs] def get_embedding(self):
"""
Returns the embedding layer of the module.
"""
return self.model.get_embedding()