easydel.modules.qwen3.modeling_qwen3_flax

class easydel.modules.qwen3.modeling_qwen3_flax.Qwen3Attention(*args: Any, **kwargs: Any)

Bases: AttentionModule

Qwen3 Attention module.

This module implements the multi-head attention mechanism used in the Qwen3 model. It supports Grouped Query Attention (GQA) and Rotary Position Embeddings (RoPE).

config

Configuration object for the model.

Type

Qwen3Config

dtype

Data type for computations.

Type

jnp.dtype

param_dtype

Data type for parameters.

Type

jnp.dtype

precision

Precision setting for JAX operations.

Type

jax.lax.PrecisionLike

rngs

Random number generators.

Type

nn.Rngs

hidden_size

Dimensionality of the hidden states.

Type

int

head_dim

Dimensionality of each attention head.

Type

int

num_key_value_groups

Number of query head groups for each key/value head.

Type

int

q_proj

Linear layer for query projection.

Type

ParallelLinear

k_proj

Linear layer for key projection.

Type

ParallelLinear

v_proj

Linear layer for value projection.

Type

ParallelLinear

o_proj

Linear layer for the output projection.

Type

ParallelLinear

attention_performer

Module to perform the core attention computation.

Type

FlexibleAttentionModule

rotary

Rotary position embedding module.

Type

RoPE

class easydel.modules.qwen3.modeling_qwen3_flax.Qwen3DecoderLayer(*args: Any, **kwargs: Any)

Bases: Module

Qwen3 Transformer Decoder Layer.

This module represents a single decoder layer in the Qwen3 model, combining self-attention and MLP sub-layers with residual connections and RMS normalization.

config

Configuration object for the model. layer_idx (int): The index of the layer in the model.

Type

Qwen3Config

dtype

Data type for computations.

Type

jnp.dtype

param_dtype

Data type for parameters.

Type

jnp.dtype

precision

Precision setting for JAX operations.

Type

jax.lax.PrecisionLike

rngs

Random number generators.

Type

nn.Rngs

input_layernorm

RMS normalization applied before the attention layer.

Type

RMSNorm

self_attn

The self-attention module.

Type

Qwen3Attention

mlp

The feed-forward (MLP) module.

Type

Qwen3MLP

post_attention_layernorm

RMS normalization applied after the attention layer and before the MLP layer.

Type

RMSNorm

class easydel.modules.qwen3.modeling_qwen3_flax.Qwen3ForCausalLM(*args: Any, **kwargs: Any)

Bases: EasyDeLBaseModule

Qwen3 model with a Causal Language Modeling head.

This model consists of the base Qwen3 transformer (Qwen3Model) followed by a linear layer (lm_head) that projects the transformer’s output hidden states to the vocabulary size, producing logits for next token prediction. Optionally, the input token embeddings can be tied to the output projection layer.

config

Configuration object for the model.

Type

Qwen3Config

dtype

Data type for computation.

Type

jnp.dtype

param_dtype

Data type for parameters.

Type

jnp.dtype

precision

Precision setting for JAX operations.

Type

jax.lax.PrecisionLike

rngs

Random number generators.

Type

nn.Rngs

model

The core Qwen3 transformer model.

Type

Qwen3Model

lm_head

The linear layer for projecting hidden states to vocabulary logits.

Type

ParallelLinear

class easydel.modules.qwen3.modeling_qwen3_flax.Qwen3ForSequenceClassification(*args: Any, **kwargs: Any)

Bases: EasyDeLBaseModule

Qwen3 model with a Sequence Classification head.

This model consists of the base Qwen3 transformer (Qwen3Model) followed by a linear layer (score) that projects the transformer’s output hidden states (typically the hidden state of the last token or a pooled representation) to the number of classes for classification.

config

Configuration object for the model.

Type

Qwen3Config

dtype

Data type for computation.

Type

jnp.dtype

param_dtype

Data type for parameters.

Type

jnp.dtype

precision

Precision setting for JAX operations.

Type

jax.lax.PrecisionLike

rngs

Random number generators.

Type

nn.Rngs

model

The core Qwen3 transformer model.

Type

Qwen3Model

score

The linear layer for classification.

Type

ParallelLinear

class easydel.modules.qwen3.modeling_qwen3_flax.Qwen3MLP(*args: Any, **kwargs: Any)

Bases: Module

Qwen3 MLP module.

This module implements the feed-forward network (MLP) used in the Qwen3 model. It uses a Gated Linear Unit (GLU) structure with SiLU activation.

config

Configuration object for the model.

Type

Qwen3Config

dtype

Data type for computations.

Type

jnp.dtype

param_dtype

Data type for parameters.

Type

jnp.dtype

precision

Precision setting for JAX operations.

Type

jax.lax.PrecisionLike

gate_proj

Linear layer for the GLU gate.

Type

ParallelLinear

down_proj

Linear layer for the down projection.

Type

ParallelLinear

up_proj

Linear layer for the GLU value.

Type

ParallelLinear

act_fn

Activation function (SiLU).

Type

callable

class easydel.modules.qwen3.modeling_qwen3_flax.Qwen3Model(*args: Any, **kwargs: Any)

Bases: EasyDeLBaseModule

The base Qwen3 model transformer.

This class represents the core transformer architecture of the Qwen3 model, consisting of an embedding layer, multiple Qwen3DecoderLayer layers, and a final RMS normalization layer.

config

Configuration object for the model.

Type

Qwen3Config

dtype

Data type for computation.

Type

jnp.dtype

param_dtype

Data type for parameters.

Type

jnp.dtype

precision

Precision setting for JAX operations.

Type

jax.lax.PrecisionLike

rngs

Random number generators.

Type

nn.Rngs

embed_tokens

Embedding layer for input tokens.

Type

nn.Embed

layers

List of decoder layers.

Type

tp.List[Qwen3DecoderLayer]

norm

Final layer normalization.

Type

RMSNorm

gradient_checkpointing

Gradient checkpointing configuration.

Type

EasyDeLGradientCheckPointers