Source code for easydel.modules.cohere2.modeling_cohere2

# 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.


from functools import partial

import chex
import jax
import jax.numpy as jnp
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.ad_checkpoint import checkpoint_name
from jaxtyping import Array, 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,
    SequenceClassifierOutput,
)
from easydel.infra.utils import ArrayParam, auto_remat, block_wise_ffn, get_dot_general_by_bits
from easydel.layers.attention import FlexibleAttentionModule
from easydel.layers.attention_unified import UnifiedAttention
from easydel.layers.base_modules import BaseCausalLMModule, BaseSequenceClassificationModule
from easydel.layers.caching import (
    RaggedPagesCache,
    RaggedPagesCacheView,
    RaggedPagesMetadata,
    TransformerCache,
    TransformerCacheView,
    TransformerMetadata,
)
from easydel.layers.linear import ColumnParallelLinear, RowParallelLinear

from .cohere2_configuration import Cohere2Config


[docs]class Cohere2LayerNorm(nn.Module): """Cohere Layer Normalization. Attributes: dim (Union[int, tuple]): The dimension(s) to normalize over. eps (float): A small epsilon value to prevent division by zero. dtype (jnp.dtype): The data type for computation. param_dtype (jnp.dtype): The data type for the parameters. rngs (Optional[nn.Rngs]): Random number generators. """ kernel_init = staticmethod(nn.initializers.ones) def __init__( self, dim: int | tuple, eps: float = 1e-6, dtype: jnp.dtype = jnp.bfloat16, param_dtype: jnp.dtype = jnp.bfloat16, rngs: nn.Rngs = None, ): super().__init__() if rngs is None: rngs = nn.Rngs(0) self.dim = dim self.eps = eps self.dtype = dtype self.param_dtype = param_dtype self.kernel = ArrayParam.bound( shape=(self.dim,) if isinstance(self.dim, int) else self.dim, dtype=self.param_dtype, init_method="ones", key=rngs.params(), ) def _norm(self, x: jnp.ndarray) -> jnp.ndarray: """Computes the Layer Normalization for a given input tensor.""" mean = jnp.mean(x, -1, keepdims=True) variance = jnp.mean(jnp.pow((x - mean), 2), -1, keepdims=True) return (x - mean) * jax.lax.rsqrt(variance + self.eps) def __call__(self, x: jnp.ndarray) -> jnp.ndarray: """Applies Layer Normalization to the input tensor. Args: x (jnp.ndarray): The input tensor. Returns: jnp.ndarray: The normalized output tensor. """ if self.dtype in [ jnp.float8_e4m3b11fnuz, jnp.float8_e4m3fn, jnp.float8_e4m3fnuz, jnp.float8_e5m2, jnp.float8_e5m2fnuz, ]: x = x.astype(jnp.float32) else: x = x.astype(jnp.promote_types(self.dtype, jnp.float32)) output = self._norm(x).astype(self.dtype) weight = self.kernel.value.astype(self.dtype) return output * weight
[docs]class Cohere2Attention(UnifiedAttention): """Cohere2 Attention with layer-specific sliding window and conditional RoPE. Inherits from UnifiedAttention with Cohere2-specific customizations: - Layer-specific sliding window (only applies to sliding_attention layers) - Conditional RoPE application (only when sliding window is enabled) """ def __init__( self, config: Cohere2Config, dtype: jnp.dtype = jnp.bfloat16, param_dtype: jnp.dtype = jnp.bfloat16, precision: jax.lax.PrecisionLike = None, *, rngs: nn.Rngs, layer_idx: int, ) -> None: """Initialize Cohere2Attention with layer-specific configuration. Args: config: Model configuration layer_idx: Layer index for determining sliding window usage dtype: Data type for computations param_dtype: Data type for parameters precision: JAX precision setting rngs: Random number generators """ super().__init__( config, dtype, param_dtype, precision, rngs=rngs, layer_idx=layer_idx, attention_type="standard", causal=True, sliding_window=config.sliding_window if config.layer_types[layer_idx] == "sliding_attention" else None, ) def _create_rotary(self, config: Cohere2Config, dtype: jnp.dtype): """Create Cohere2-specific rotary embedding layer.""" return config.get_basic_rope(dtype, self.head_dim, self.head_dim, False) def _create_attention_performer(self, config: Cohere2Config, rngs: nn.Rngs): """Create attention performer with Cohere2's attention dropout.""" return FlexibleAttentionModule( rngs=rngs, dropout_prob=config.attention_dropout, base_config=config, softmax_scale=self.head_dim**-0.5, ) def _apply_rotary(self, query_states, key_states, position_ids, frequencies): """Override to apply RoPE only when sliding window is enabled (Cohere2-specific).""" if self.sliding_window is not None: return self.rotary( query=query_states, key=key_states, positions=position_ids, frequencies=frequencies, ) return query_states, key_states
[docs]class Cohere2MLP(nn.Module): """Feed-forward network used in Cohere v2 decoder layers.""" def __init__( self, config: Cohere2Config, dtype: jnp.dtype = jnp.bfloat16, param_dtype: jnp.dtype = jnp.bfloat16, precision: jax.lax.PrecisionLike = None, *, rngs: nn.Rngs, ): self.config = config self.dtype = dtype self.param_dtype = param_dtype self.precision = precision column_parallel_linear = partial( ColumnParallelLinear, 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), ) row_parallel_linear = partial( RowParallelLinear, 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.gate_proj = column_parallel_linear(config.hidden_size, config.intermediate_size) self.down_proj = row_parallel_linear(config.intermediate_size, config.hidden_size) self.up_proj = column_parallel_linear(config.hidden_size, config.intermediate_size) def __call__( self, hidden_states: Float[Array, "batch seq_len hidden_dim"] ) -> Float[Array, "batch seq_len hidden_dim"]: hidden_states = apply_logical_sharding( hidden_states, dynamic_axes=common_types.HiddenStateSharding, partition_manager=self.config.partition_manager, ) gate = jax.nn.silu(checkpoint_name(self.gate_proj(hidden_states), name="mlp_gate")) up = checkpoint_name(self.up_proj(hidden_states), name="mlp_up") hidden_states = checkpoint_name(self.down_proj(gate * up), name="mlp_down") hidden_states = apply_logical_sharding( hidden_states, dynamic_axes=common_types.HiddenStateSharding, partition_manager=self.config.partition_manager, ) return hidden_states
[docs]class Cohere2Block(nn.Module): """Cohere v2 transformer block combining attention and MLP.""" def __init__( self, config: Cohere2Config, dtype: jnp.dtype = jnp.bfloat16, param_dtype: jnp.dtype = jnp.bfloat16, precision: jax.lax.PrecisionLike = None, *, rngs: nn.Rngs, layer_idx: int, ) -> None: super().__init__() self.config = config self.layer_idx = layer_idx self.dtype = dtype self.param_dtype = param_dtype self.precision = precision self.rngs = rngs attn_block = Cohere2Attention mlp_block = Cohere2MLP 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, layer_idx=layer_idx, dtype=dtype, param_dtype=param_dtype, precision=precision, rngs=rngs, ) self.mlp = mlp_block( config, dtype=dtype, param_dtype=param_dtype, precision=precision, rngs=rngs, ) self.input_layernorm = Cohere2LayerNorm( self.config.hidden_size, eps=self.config.layer_norm_eps, dtype=dtype, param_dtype=param_dtype, rngs=rngs, ) self.is_sliding = (layer_idx + 1) % self.config.sliding_window_pattern != 0 self.sliding_window = config.sliding_window 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 module block. 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): Causal mask for ensuring autoregressive behavior. segment_ids (tp.Optional[chex.Array]): Segment IDs for segment-based attention (optional). deterministic (bool): If True, disables dropout for deterministic behavior. init_cache (bool): If True, initializes cache for caching keys and values. output_attentions (bool): If True, outputs attention weights alongside the hidden states. fcm_mask (tp.Optional[chex.Array]): fcm mask to be combined with attn mask and causal mask. Returns: tp.Tuple[chex.Array, chex.Array]: A tuple containing the attention output and 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, ) feed_forward_input = hidden_states if self.config.use_scan_mlp: feed_forward_hidden_states = block_wise_ffn( self.mlp, feed_forward_input, self.config.scan_mlp_chunk_size, ) else: feed_forward_hidden_states = self.mlp(feed_forward_input) hidden_states = attn_outputs.attention_output + feed_forward_hidden_states + residual 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, router_logits=None, gate_loss=None, cache_view=attn_outputs.cache_view, )
[docs]@register_module(TaskType.BASE_MODULE, config=Cohere2Config, model_type="cohere2") class Cohere2Model(EasyDeLBaseModule): """Decoder-only Cohere v2 model with embeddings, blocks, and final norm.""" def __init__( self, config: Cohere2Config, dtype: jnp.dtype = jnp.bfloat16, param_dtype: jnp.dtype = jnp.bfloat16, precision: jax.lax.PrecisionLike = None, *, rngs: nn.Rngs, ): super().__init__( config=config, dtype=dtype, param_dtype=param_dtype, precision=precision, rngs=rngs, ) 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, embedding_init=nn.initializers.normal(stddev=config.initializer_range), dtype=dtype, param_dtype=param_dtype, rngs=rngs, ) self.layers = [ Cohere2Block( config=config, layer_idx=idx, dtype=dtype, param_dtype=param_dtype, precision=precision, rngs=rngs, ) for idx in range(config.num_hidden_layers) ] self.norm = Cohere2LayerNorm( self.config.hidden_size, eps=self.config.layer_norm_eps, dtype=dtype, param_dtype=param_dtype, ) 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: 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_tokens(input_ids.astype("i4")) 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.norm(hidden_states) if output_hidden_states: all_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) -> nn.Module: """ Returns the encoder part of the model's graph definition. For Cohere2Model (decoder-only), this is not applicable. """ raise NotImplementedError("Cohere2Model is a decoder-only model and does not have a separate encoder.")
[docs] def get_decoder(self) -> nn.Module: """ Returns the decoder part of the model's graph definition. For Cohere2Model, this is the model itself. """ return self
[docs] def get_lm_head(self) -> nn.Module: """ Returns the language model head of the module. Cohere2Model does not include the lm_head. """ raise NotImplementedError("Cohere2Model does not include the language model head. See Cohere2ForCausalLM.")
[docs] def get_embedding(self) -> nn.Module: """ Returns the embedding layer of the module. """ return self.embed_tokens
[docs]@register_module(TaskType.CAUSAL_LM, config=Cohere2Config, model_type="cohere2") class Cohere2ForCausalLM(BaseCausalLMModule[Cohere2Model, Cohere2Config]): """Cohere2 model with a Causal Language Modeling head.""" _task_type = TaskType.CAUSAL_LM _model_type = "cohere2" _config_class = Cohere2Config def __init__( self, config: Cohere2Config, 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=Cohere2Model, base_model_name="model", dtype=dtype, param_dtype=param_dtype, precision=precision, rngs=rngs, lm_head_bias=False, ) self.logit_scale = self.config.logit_scale 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, apply_lm_head: bool = True, output_attentions: bool | None = None, output_hidden_states: bool | None = None, ) -> CausalLMOutput: """ Forward pass through the Cohere module. Args: input_ids (chex.Array): Input tensor containing token IDs. attention_mask (chex.Array): Mask for attention. position_ids (chex.Array): Positional indices. segment_ids (tp.Optional[chex.Array]): Segment IDs for different input parts. inputs_embeds (tp.Optional[chex.Array]): Embedded input tensor. output_attentions (tp.Optional[bool]): If True, output attention weights. output_hidden_states (tp.Optional[bool]): If True, output hidden states. init_cache (bool): If True, initialize cache for decoding. deterministic (bool): If True, disable dropout. Returns: CausalLMOutput | tp.Tuple: Model output, either as a named tuple or a standard tuple. """ outputs = self.model( input_ids=input_ids, attention_mask=attention_mask, mask_info=mask_info, position_ids=position_ids, output_attentions=output_attentions, output_hidden_states=output_hidden_states, mode=mode, past_key_values=past_key_values, cache_metadata=cache_metadata, inputs_embeds=inputs_embeds, ) hidden_states = outputs.last_hidden_state hidden_states = apply_logical_sharding( hidden_states, dynamic_axes=common_types.HiddenStateSharding, partition_manager=self.config.partition_manager, ) lm_logits = None if apply_lm_head: lm_logits = self.apply_lm_head(hidden_states) 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 apply_lm_head(self, hidden_states: chex.Array) -> chex.Array: """ Applies the language model head to the hidden states. Args: hidden_states (chex.Array): The last hidden states from the model. Returns: chex.Array: The logits after applying the language model head. """ lm_logits = self.lm_head(hidden_states) if self.logit_scale is not None: lm_logits *= self.logit_scale return lm_logits
[docs] def get_encoder(self) -> nn.Module: """ Returns the encoder part of the model's graph definition. For Cohere2ForCausalLM (decoder-only), this is not applicable. """ raise NotImplementedError("Cohere2ForCausalLM is a decoder-only model and does not have a separate encoder.")
[docs] def get_decoder(self) -> nn.Module: """ Returns the decoder part of the model's graph definition. For Cohere2ForCausalLM, this is the underlying Cohere2Model. """ return self.model.get_decoder() # self.model is the Cohere2Model instance
[docs] def get_lm_head(self) -> nn.Module: """ Returns the language model head of the module. """ return self.lm_head
[docs] def get_embedding(self) -> nn.Module: """ Returns the embedding layer of the module. """ # Access the embedding layer through the decoder (Cohere2Model) return self.model.get_embedding() # Leverages Cohere2Model's get_embedding
[docs]@register_module(TaskType.SEQUENCE_CLASSIFICATION, config=Cohere2Config, model_type="cohere2") class Cohere2ForSequenceClassification(BaseSequenceClassificationModule[Cohere2Model, Cohere2Config]): """Cohere2 model for sequence classification.""" _task_type = TaskType.SEQUENCE_CLASSIFICATION _model_type = "cohere2" _config_class = Cohere2Config def __init__( self, config: Cohere2Config, 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=Cohere2Model, base_model_name="model", dtype=dtype, param_dtype=param_dtype, precision=precision, rngs=rngs, classifier_name="score", classifier_bias=False, ) 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, ) -> SequenceClassifierOutput: transformer_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 = transformer_outputs.last_hidden_state logits = self.score(hidden_states) if input_ids is not None: batch_size = input_ids.shape[0] else: batch_size = inputs_embeds.shape[0] if self.config.pad_token_id is None and batch_size != 1: raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.") if self.config.pad_token_id is None: sequence_lengths = -1 else: if input_ids is not None: sequence_lengths = jnp.argmax(jnp.equal(input_ids, self.config.pad_token_id).astype("i4"), -1) - 1 sequence_lengths = sequence_lengths % input_ids.shape[-1] else: sequence_lengths = -1 pooled_logits = logits[jnp.arange(batch_size), sequence_lengths] return SequenceClassifierOutput( logits=pooled_logits, past_key_values=past_key_values, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, )
[docs] def get_encoder(self) -> nn.Module: """ Returns the encoder part of the model's graph definition. For Cohere2ForSequenceClassification (decoder-only), this is not applicable. """ raise NotImplementedError( "Cohere2ForSequenceClassification is a decoder-only model and does not have a separate encoder." )
[docs] def get_decoder(self) -> nn.Module: """ Returns the decoder part of the model's graph definition. For Cohere2ForSequenceClassification, this is the underlying Cohere2Model. """ return self.model # self.model is the Cohere2Model instance
[docs] def get_lm_head(self) -> nn.Module: """ Returns the language model head of the module. Cohere2ForSequenceClassification uses a classification head instead. """ raise NotImplementedError( "Cohere2ForSequenceClassification uses a classification head (self.score), not an lm_head." )
[docs] def get_embedding(self) -> nn.Module: """ Returns the embedding layer of the module. """ # Access the embedding layer through the decoder (Cohere2Model) return self.model.get_embedding() # Leverages Cohere2Model's get_embedding