SmolLM3Backbone model

[source]

SmolLM3Backbone class

keras_hub.models.SmolLM3Backbone(
    vocabulary_size,
    hidden_dim,
    intermediate_dim,
    num_layers,
    num_attention_heads,
    num_key_value_heads,
    attention_bias,
    attention_dropout,
    rope_layer_enabled_list,
    layer_types,
    mlp_bias,
    layer_norm_epsilon,
    max_position_embeddings,
    rope_theta,
    partial_rotary_factor,
    **kwargs
)

SmolLM3 core network with hyperparameters.

This network implements a Transformer-based decoder network, SmolLM3, as described in the SmolLM3 model architecture. It includes the embedding lookups and transformer layers.

The default constructor gives a fully customizable, randomly initialized SmolLM3 model with any number of layers, heads, and embedding dimensions. To load preset architectures and weights, use the from_preset constructor.

Arguments

• vocabulary_size: int. The size of the token vocabulary.
• hidden_dim: int. The size of the transformer hidden state at the end of each transformer layer.
• intermediate_dim: int. The output dimension of the first Dense layer in the MLP network of each transformer layer.
• num_layers: int. The number of transformer layers.
• num_attention_heads: int. The number of attention heads for each transformer layer.
• num_key_value_heads: int. The number of key-value heads for grouped query attention in each transformer layer.
• attention_bias: bool. Whether to use bias in the query, key, value, and output projection layers in the attention blocks.
• attention_dropout: float. Dropout probability for the attention layers.
• rope_layer_enabled_list: list of bool. List indicating whether RoPE (Rotary Position Embedding) is enabled for each layer. Typically, some layers may disable RoPE for architectural variations.
• layer_types: list of str. List of layer types for each transformer layer (e.g., "attention" or other custom types).
• mlp_bias: bool. Whether to use bias in the MLP (feedforward) layers.
• layer_norm_epsilon: float. Epsilon value for layer normalization layers to prevent division by zero.
• max_position_embeddings: int. The maximum sequence length that this model might ever be used with.
• rope_theta: float. The base period of the RoPE embeddings.
• partial_rotary_factor: float. The percentage of hidden dimensions to rotate in RoPE. A value of 1.0 rotates all dimensions, while values less than 1.0 only rotate a subset.

Examples

input_data = {
    "token_ids": np.ones(shape=(1, 12), dtype="int32"),
    "padding_mask": np.array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0]]),
}

# Pretrained SmolLM3 decoder.
model = keras_hub.models.SmolLM3Backbone.from_preset(
    "hf://HuggingFaceTB/SmolLM3-3B"
)
model(input_data)

# Randomly initialized SmolLM3 decoder with custom config.
model = keras_hub.models.SmolLM3Backbone(
    vocabulary_size=49152,
    hidden_dim=576,
    intermediate_dim=1536,
    num_layers=30,
    num_attention_heads=9,
    num_key_value_heads=3,
    attention_bias=False,
    attention_dropout=0.0,
    rope_layer_enabled_list=[True] * 30,
    layer_types=["attention"] * 30,
    mlp_bias=False,
    layer_norm_epsilon=1e-5,
    max_position_embeddings=2048,
    rope_theta=10000.0,
    partial_rotary_factor=1.0,
)
model(input_data)

[source]

from_preset method

SmolLM3Backbone.from_preset(preset, load_weights=True, **kwargs)

Instantiate a keras_hub.models.Backbone from a model preset.

A preset is a directory of configs, weights and other file assets used to save and load a pre-trained model. The preset can be passed as a one of:

1. a built-in preset identifier like 'bert_base_en'
2. a Kaggle Models handle like 'kaggle://user/bert/keras/bert_base_en'
3. a Hugging Face handle like 'hf://user/bert_base_en'
4. a ModelScope handle like 'modelscope://user/bert_base_en'
5. a path to a local preset directory like './bert_base_en'

This constructor can be called in one of two ways. Either from the base class like keras_hub.models.Backbone.from_preset(), or from a model class like keras_hub.models.GemmaBackbone.from_preset(). If calling from the base class, the subclass of the returning object will be inferred from the config in the preset directory.

For any Backbone subclass, you can run cls.presets.keys() to list all built-in presets available on the class.

Arguments

• preset: string. A built-in preset identifier, a Kaggle Models handle, a Hugging Face handle, or a path to a local directory.
• load_weights: bool. If True, the weights will be loaded into the model architecture. If False, the weights will be randomly initialized.

Examples

# Load a Gemma backbone with pre-trained weights.
model = keras_hub.models.Backbone.from_preset(
    "gemma_2b_en",
)

# Load a Bert backbone with a pre-trained config and random weights.
model = keras_hub.models.Backbone.from_preset(
    "bert_base_en",
    load_weights=False,
)

token_embedding property

keras_hub.models.SmolLM3Backbone.token_embedding

A keras.layers.Embedding instance for embedding token ids.

This layer embeds integer token ids to the hidden dim of the model.