openelm

mindnlp.transformers.models.openelm.modeling_openelm

OpenELM config

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMDecoderLayer

Bases: Module

The OpenELMDecoderLayer class represents a single layer of the OpenELM decoder model. It is designed to be used in the OpenELMDecoder model for generating high-quality sequence predictions.

This class inherits from the nn.Module class, which provides a base class for all neural network cells in MindSpore.

ATTRIBUTE	DESCRIPTION
attn	An instance of the OpenELMMultiHeadCausalAttention class responsible for performing multi-head causal attention operations. TYPE: OpenELMMultiHeadCausalAttention
ffn	An instance of the OpenELMFeedForwardNetwork class responsible for applying feed-forward neural network operations. TYPE: OpenELMFeedForwardNetwork
ffn_norm	An instance of the OpenELMRMSNorm class responsible for normalizing the output of the feed-forward network. TYPE: OpenELMRMSNorm
attn_norm	An instance of the OpenELMRMSNorm class responsible for normalizing the output of the attention layer. TYPE: OpenELMRMSNorm

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMDecoderLayer(nn.Module):

    """
    The `OpenELMDecoderLayer` class represents a single layer of the OpenELM decoder model.
    It is designed to be used in the OpenELMDecoder model for generating high-quality sequence predictions.

    This class inherits from the `nn.Module` class, which provides a base class for all neural network cells in MindSpore.

    Attributes:
        attn (OpenELMMultiHeadCausalAttention): An instance of the `OpenELMMultiHeadCausalAttention` class responsible
            for performing multi-head causal attention operations.
        ffn (OpenELMFeedForwardNetwork): An instance of the `OpenELMFeedForwardNetwork` class responsible
            for applying feed-forward neural network operations.
        ffn_norm (OpenELMRMSNorm): An instance of the `OpenELMRMSNorm` class responsible
            for normalizing the output of the feed-forward network.
        attn_norm (OpenELMRMSNorm): An instance of the `OpenELMRMSNorm` class responsible
            for normalizing the output of the attention layer.
    """
    def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
        """Initialize an instance of the OpenELMDecoderLayer class.

        Args:
            self: The instance of the OpenELMDecoderLayer class.
            config (OpenELMConfig): The configuration object for OpenELM.
                It specifies the model configuration settings.
            layer_idx (int): The index of the current layer in the decoder stack.
                It is used for identifying the layer position.

        Returns:
            None.

        Raises:
            None.
        """
        super().__init__()
        self.attn = OpenELMMultiHeadCausalAttention(config=config, layer_idx=layer_idx)
        self.ffn = OpenELMFeedForwardNetwork(config=config, layer_idx=layer_idx)
        self.ffn_norm = OpenELMRMSNorm(
            num_features=config.model_dim,
        )
        self.attn_norm = OpenELMRMSNorm(
            num_features=config.model_dim,
        )

    def forward(
        self,
        hidden_states: mindspore.Tensor,
        attention_mask: Optional[mindspore.Tensor] = None,
        position_ids: Optional[mindspore.Tensor] = None,
        past_key_value: Optional[Tuple[mindspore.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cache_position: Optional[mindspore.Tensor] = None,
        **kwargs,
    ) -> Tuple[
        mindspore.Tensor, Optional[Tuple[mindspore.Tensor, mindspore.Tensor]]
    ]:
        """
        Args:
            hidden_states (`mindspore.Tensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
            attention_mask (`mindspore.Tensor`, *optional*):
                attention mask of size `(batch_size, sequence_length)` if flash attention is used or `(batch_size, 1,
                query_sequence_length, key_sequence_length)` if default attention is used.
            output_attentions (`bool`, *optional*):
                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                returned tensors for more detail.
            use_cache (`bool`, *optional*):
                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
                (see `past_key_values`).
            past_key_value (`Tuple(mindspore.Tensor)`, *optional*): cached past key and value projection states
        """
        residual = hidden_states
        hidden_states = self.attn_norm(hidden_states)

        # Self Attention
        hidden_states, self_attn_weights, present_key_value = self.attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
            cache_position=cache_position,
            **kwargs,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.ffn_norm(hidden_states)
        hidden_states = self.ffn(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMDecoderLayer.__init__(config, layer_idx)

Initialize an instance of the OpenELMDecoderLayer class.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMDecoderLayer class.
config	The configuration object for OpenELM. It specifies the model configuration settings. TYPE: OpenELMConfig
layer_idx	The index of the current layer in the decoder stack. It is used for identifying the layer position. TYPE: int

RETURNS	DESCRIPTION
None	None.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
    """Initialize an instance of the OpenELMDecoderLayer class.

    Args:
        self: The instance of the OpenELMDecoderLayer class.
        config (OpenELMConfig): The configuration object for OpenELM.
            It specifies the model configuration settings.
        layer_idx (int): The index of the current layer in the decoder stack.
            It is used for identifying the layer position.

    Returns:
        None.

    Raises:
        None.
    """
    super().__init__()
    self.attn = OpenELMMultiHeadCausalAttention(config=config, layer_idx=layer_idx)
    self.ffn = OpenELMFeedForwardNetwork(config=config, layer_idx=layer_idx)
    self.ffn_norm = OpenELMRMSNorm(
        num_features=config.model_dim,
    )
    self.attn_norm = OpenELMRMSNorm(
        num_features=config.model_dim,
    )

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMDecoderLayer.forward(hidden_states, attention_mask=None, position_ids=None, past_key_value=None, output_attentions=False, use_cache=False, cache_position=None, **kwargs)

PARAMETER	DESCRIPTION
hidden_states	input to the layer of shape (batch, seq_len, embed_dim) TYPE: `mindspore.Tensor`
attention_mask	attention mask of size (batch_size, sequence_length) if flash attention is used or (batch_size, 1, query_sequence_length, key_sequence_length) if default attention is used. TYPE: `mindspore.Tensor`, *optional* DEFAULT: None
output_attentions	Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail. TYPE: `bool`, *optional* DEFAULT: False
use_cache	If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values). TYPE: `bool`, *optional* DEFAULT: False
past_key_value	cached past key and value projection states TYPE: `Tuple(mindspore.Tensor)`, *optional* DEFAULT: None

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(
    self,
    hidden_states: mindspore.Tensor,
    attention_mask: Optional[mindspore.Tensor] = None,
    position_ids: Optional[mindspore.Tensor] = None,
    past_key_value: Optional[Tuple[mindspore.Tensor]] = None,
    output_attentions: Optional[bool] = False,
    use_cache: Optional[bool] = False,
    cache_position: Optional[mindspore.Tensor] = None,
    **kwargs,
) -> Tuple[
    mindspore.Tensor, Optional[Tuple[mindspore.Tensor, mindspore.Tensor]]
]:
    """
    Args:
        hidden_states (`mindspore.Tensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
        attention_mask (`mindspore.Tensor`, *optional*):
            attention mask of size `(batch_size, sequence_length)` if flash attention is used or `(batch_size, 1,
            query_sequence_length, key_sequence_length)` if default attention is used.
        output_attentions (`bool`, *optional*):
            Whether or not to return the attentions tensors of all attention layers. See `attentions` under
            returned tensors for more detail.
        use_cache (`bool`, *optional*):
            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
            (see `past_key_values`).
        past_key_value (`Tuple(mindspore.Tensor)`, *optional*): cached past key and value projection states
    """
    residual = hidden_states
    hidden_states = self.attn_norm(hidden_states)

    # Self Attention
    hidden_states, self_attn_weights, present_key_value = self.attn(
        hidden_states=hidden_states,
        attention_mask=attention_mask,
        past_key_value=past_key_value,
        output_attentions=output_attentions,
        use_cache=use_cache,
        cache_position=cache_position,
        **kwargs,
    )
    hidden_states = residual + hidden_states

    # Fully Connected
    residual = hidden_states
    hidden_states = self.ffn_norm(hidden_states)
    hidden_states = self.ffn(hidden_states)
    hidden_states = residual + hidden_states

    outputs = (hidden_states,)

    if output_attentions:
        outputs += (self_attn_weights,)

    if use_cache:
        outputs += (present_key_value,)

    return outputs

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMFeedForwardNetwork

Bases: Module

The OpenELMFeedForwardNetwork class represents a feedforward network layer for the OpenELM model. This class inherits from nn.Module and implements the forward function of the feedforward network layer.

The init method initializes the OpenELMFeedForwardNetwork instance with the provided configuration and layer index. It calculates the intermediate dimensions based on the configuration, initializes the projection layers, and sets the activation function based on the configuration.

The extra_repr method returns a string representation of the instance, including the ffn_with_glu attribute.

The forward method implements the forward function of the feedforward network layer. It takes an input tensor of shape [batch size, sequence length, model dimension], applies the projection layers and activation functions based on the configuration, and returns a tensor of the same shape as the input.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMFeedForwardNetwork(nn.Module):

    """
    The OpenELMFeedForwardNetwork class represents a feedforward network layer for the OpenELM model.
    This class inherits from nn.Module and implements the forward function of the feedforward network layer.

    The __init__ method initializes the OpenELMFeedForwardNetwork instance with the provided configuration and layer index.
    It calculates the intermediate dimensions based on the configuration, initializes the projection layers, and sets
    the activation function based on the configuration.

    The extra_repr method returns a string representation of the instance, including the ffn_with_glu attribute.

    The forward method implements the forward function of the feedforward network layer.
    It takes an input tensor of shape [batch size, sequence length, model dimension], applies the projection layers and
    activation functions based on the configuration, and returns a tensor of the same shape as the input.

    """
    def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
        """
        Initializes an instance of the OpenELMFeedForwardNetwork class.

        Args:
            self: The instance of the class.
            config (OpenELMConfig): An instance of the OpenELMConfig class containing configuration settings.
            layer_idx (int): The index of the layer in the network.

        Returns:
            None.

        Raises:
            TypeError: If the input parameters are of incorrect types.
            ValueError: If the layer index is out of bounds or if there are any configuration issues.
            KeyError: If the activation function name specified in the config is not found in the ACT2FN dictionary.
        """
        super().__init__()
        ffn_multiplier = config.ffn_multipliers[layer_idx]
        intermediate_dim = int(
            make_divisible(
                ffn_multiplier * config.model_dim,
                divisor=config.ffn_dim_divisor,
            )
        )
        if config.ffn_with_glu:
            # FFN with Gated linear unit, as described in https://arxiv.org/abs/2002.05202v1.
            self.proj_1 = nn.Linear(
                in_channels=config.model_dim,
                out_channels=2 * intermediate_dim,
                bias=False,
            )
            self.proj_2 = nn.Linear(
                in_channels=intermediate_dim,
                out_channels=config.model_dim,
                bias=False,
            )
            self.ffn_with_glu = True
        else:
            # Standard FFN, as described in https://arxiv.org/abs/1706.03762
            self.proj_1 = nn.Linear(
                in_channels=config.model_dim,
                out_channels=intermediate_dim,
                bias=False,
            )
            self.proj_2 = nn.Linear(
                in_channels=intermediate_dim,
                out_channels=config.model_dim,
                bias=False,
            )
            self.ffn_with_glu = False

        self.act = ACT2FN[config.activation_fn_name]

    def extra_repr(self) -> str:
        """
        This method generates a string representation of the OpenELMFeedForwardNetwork object with additional
        information about the feedforward network configuration.

        Args:
            self (OpenELMFeedForwardNetwork): The instance of the OpenELMFeedForwardNetwork class.

        Returns:
            str: A string representation of the OpenELMFeedForwardNetwork object with the additional information
                about the feedforward network configuration including the ffn_with_glu attribute.

        Raises:
            None.
        """
        return super().extra_repr() + f"(ffn_with_glu) : {self.ffn_with_glu}"

    def forward(self, x: Tensor) -> Tensor:
        """Forward function of FFN layer.

        Args:
            x: Input tensor of the shape [batch size, sequence length, model dimension].

        Returns:
            A tensor of the same shape as the input.
        """
        if self.ffn_with_glu:
            y_12 = self.proj_1(x)
            y_1, y_2 = y_12.chunk(2, axis=-1)
            y = self.act(y_1) * y_2
            return self.proj_2(y)
        else:
            return self.proj_2(self.act(self.proj_1(x)))

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMFeedForwardNetwork.__init__(config, layer_idx)

Initializes an instance of the OpenELMFeedForwardNetwork class.

PARAMETER	DESCRIPTION
self	The instance of the class.
config	An instance of the OpenELMConfig class containing configuration settings. TYPE: OpenELMConfig
layer_idx	The index of the layer in the network. TYPE: int

RETURNS	DESCRIPTION
None	None.

RAISES	DESCRIPTION
TypeError	If the input parameters are of incorrect types.
ValueError	If the layer index is out of bounds or if there are any configuration issues.
KeyError	If the activation function name specified in the config is not found in the ACT2FN dictionary.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
    """
    Initializes an instance of the OpenELMFeedForwardNetwork class.

    Args:
        self: The instance of the class.
        config (OpenELMConfig): An instance of the OpenELMConfig class containing configuration settings.
        layer_idx (int): The index of the layer in the network.

    Returns:
        None.

    Raises:
        TypeError: If the input parameters are of incorrect types.
        ValueError: If the layer index is out of bounds or if there are any configuration issues.
        KeyError: If the activation function name specified in the config is not found in the ACT2FN dictionary.
    """
    super().__init__()
    ffn_multiplier = config.ffn_multipliers[layer_idx]
    intermediate_dim = int(
        make_divisible(
            ffn_multiplier * config.model_dim,
            divisor=config.ffn_dim_divisor,
        )
    )
    if config.ffn_with_glu:
        # FFN with Gated linear unit, as described in https://arxiv.org/abs/2002.05202v1.
        self.proj_1 = nn.Linear(
            in_channels=config.model_dim,
            out_channels=2 * intermediate_dim,
            bias=False,
        )
        self.proj_2 = nn.Linear(
            in_channels=intermediate_dim,
            out_channels=config.model_dim,
            bias=False,
        )
        self.ffn_with_glu = True
    else:
        # Standard FFN, as described in https://arxiv.org/abs/1706.03762
        self.proj_1 = nn.Linear(
            in_channels=config.model_dim,
            out_channels=intermediate_dim,
            bias=False,
        )
        self.proj_2 = nn.Linear(
            in_channels=intermediate_dim,
            out_channels=config.model_dim,
            bias=False,
        )
        self.ffn_with_glu = False

    self.act = ACT2FN[config.activation_fn_name]

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMFeedForwardNetwork.extra_repr()

This method generates a string representation of the OpenELMFeedForwardNetwork object with additional information about the feedforward network configuration.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMFeedForwardNetwork class. TYPE: OpenELMFeedForwardNetwork

RETURNS	DESCRIPTION
str	A string representation of the OpenELMFeedForwardNetwork object with the additional information about the feedforward network configuration including the ffn_with_glu attribute. TYPE: str

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def extra_repr(self) -> str:
    """
    This method generates a string representation of the OpenELMFeedForwardNetwork object with additional
    information about the feedforward network configuration.

    Args:
        self (OpenELMFeedForwardNetwork): The instance of the OpenELMFeedForwardNetwork class.

    Returns:
        str: A string representation of the OpenELMFeedForwardNetwork object with the additional information
            about the feedforward network configuration including the ffn_with_glu attribute.

    Raises:
        None.
    """
    return super().extra_repr() + f"(ffn_with_glu) : {self.ffn_with_glu}"

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMFeedForwardNetwork.forward(x)

Forward function of FFN layer.

PARAMETER	DESCRIPTION
x	Input tensor of the shape [batch size, sequence length, model dimension]. TYPE: Tensor

RETURNS	DESCRIPTION
Tensor	A tensor of the same shape as the input.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(self, x: Tensor) -> Tensor:
    """Forward function of FFN layer.

    Args:
        x: Input tensor of the shape [batch size, sequence length, model dimension].

    Returns:
        A tensor of the same shape as the input.
    """
    if self.ffn_with_glu:
        y_12 = self.proj_1(x)
        y_1, y_2 = y_12.chunk(2, axis=-1)
        y = self.act(y_1) * y_2
        return self.proj_2(y)
    else:
        return self.proj_2(self.act(self.proj_1(x)))

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM

Bases: OpenELMPreTrainedModel

This class represents a OpenELM model for Causal Language Modeling (LM). It is designed for generating text based on input sequences and predicting the next token in a sequence. The class includes methods for setting and getting input and output embeddings, setting the decoder, forwarding the model for generation, and preparing inputs for text generation. Additionally, it provides a static method for reordering cache during generation. The class inherits from OpenELMPreTrainedModel and implements functionality specific to Causal LM tasks.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMForCausalLM(OpenELMPreTrainedModel):

    """
    This class represents a OpenELM model for Causal Language Modeling (LM).
    It is designed for generating text based on input sequences and predicting the next token in a sequence.
    The class includes methods for setting and getting input and output embeddings, setting the decoder,
    forwarding the model for generation, and preparing inputs for text generation.
    Additionally, it provides a static method for reordering cache during generation.
    The class inherits from OpenELMPreTrainedModel and implements functionality specific to Causal LM tasks.
    """
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config: OpenELMConfig):
        """
        Initializes an instance of the OpenELMForCausalLM class.

        Args:
            self: The current object instance.
            config (OpenELMConfig): An instance of OpenELMConfig class containing the configuration settings
                for the OpenELM model.

        Returns:
            None

        Raises:
            None

        This method initializes the OpenELMForCausalLM object by setting its configuration, transformer, vocab_size,
        and lm_head attributes. The config parameter is an instance of OpenELMConfig class and is required to configure
        the OpenELM model.

        Attributes:
            self.transformer: An instance of the OpenELMModel class.
            self.vocab_size: An integer representing the size of the vocabulary used in the model.
            self.lm_head: An instance of the nn.Linear class or None depending on the value of
                config.share_input_output_layers.

        Note:
            The OpenELMModel and nn.Linear classes are imported from the appropriate libraries.

        Example:
            ```python
            >>> config = OpenELMConfig(vocab_size=10000, share_input_output_layers=False)
            >>> open_elm = OpenELMForCausalLM(config)
            ```
        """
        super().__init__(config)
        self.transformer = OpenELMModel(config)
        self.vocab_size = config.vocab_size
        if config.share_input_output_layers:
            self.lm_head = None
        else:
            self.lm_head = nn.Linear(config.model_dim, config.vocab_size, bias=False)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        """
        Retrieve the input embeddings from the OpenELMForCausalLM model.

        Args:
            self (OpenELMForCausalLM): The instance of OpenELMForCausalLM.

        Returns:
            token_embeddings: This method returns the input embeddings as a transformer token embeddings.

        Raises:
            None.
        """
        return self.transformer.token_embeddings

    def set_input_embeddings(self, value):
        """
        Sets the input embeddings for the OpenELMForCausalLM model.

        Args:
            self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
            value (torch.Tensor): The input embeddings to be set for the model. It should be a torch.Tensor object.

        Returns:
            None.

        Raises:
            None.
        """
        self.transformer.token_embeddings = value

    def get_output_embeddings(self):
        """
        Returns the output embeddings of the OpenELMForCausalLM model.

        Args:
            self: An instance of the OpenELMForCausalLM class.

        Returns:
            lm_head: The method returns the output embeddings of the OpenELMForCausalLM model.

        Raises:
            None.
        """
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        """
        This method sets the output embeddings for the OpenELMForCausalLM class.

        Args:
            self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
            new_embeddings (object): The new output embeddings to be set for the OpenELMForCausalLM instance.

        Returns:
            None.

        Raises:
            None.
        """
        self.lm_head = new_embeddings

    def set_decoder(self, decoder):
        """Set the decoder for the OpenELMForCausalLM instance.

        This method allows setting the decoder for the OpenELMForCausalLM instance.
        The decoder is used to transform the input data.

        Args:
            self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
            decoder: The decoder to be set. It should be compatible with the OpenELMForCausalLM instance.

        Returns:
            None.

        Raises:
            None.
        """
        self.transformer = decoder

    def get_decoder(self):
        """
        This method returns the transformer for OpenELMForCausalLM.

        Args:
            self (object): The instance of the OpenELMForCausalLM class.

        Returns:
            None: This method returns the transformer object associated with the OpenELMForCausalLM instance.

        Raises:
            None
        """
        return self.transformer

    def forward(
        self,
        input_ids: mindspore.Tensor = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        position_ids: Optional[mindspore.Tensor] = None,
        past_key_values: Optional[List[mindspore.Tensor]] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        labels: Optional[mindspore.Tensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[mindspore.Tensor] = None,
    ) -> Union[Tuple, CausalLMOutputWithPast]:
        """
        This method forwards a Causal Language Model for OpenELM.

        Args:
            self (object): The instance of the class.
            input_ids (mindspore.Tensor, optional): The input tensor containing token IDs. Default is None.
            attention_mask (mindspore.Tensor, optional): An optional tensor for masking tokens. Default is None.
            position_ids (mindspore.Tensor, optional): An optional tensor containing position IDs. Default is None.
            past_key_values (List[mindspore.Tensor], optional): A list of tensors representing past key values.
                Default is None.
            inputs_embeds (mindspore.Tensor, optional): An optional tensor of input embeddings. Default is None.
            labels (mindspore.Tensor, optional): An optional tensor containing labels. Default is None.
            use_cache (bool, optional): A flag indicating whether to use cache. Default is None.
            output_attentions (bool, optional): A flag indicating whether to output attentions. Default is None.
            output_hidden_states (bool, optional): A flag indicating whether to output hidden states. Default is None.
            return_dict (bool, optional): A flag indicating whether to return a dictionary. Default is None.
            cache_position (mindspore.Tensor, optional): An optional tensor for cache position. Default is None.

        Returns:
            Union[Tuple, CausalLMOutputWithPast]: The output of the method, which can be a tuple or an instance of
                CausalLMOutputWithPast. If return_dict is False, the return value includes loss, logits,
                past key values, hidden states, and attentions.

        Raises:
            None
        """
        output_attentions = (
            output_attentions
            if output_attentions is not None
            else self.config.output_attentions
        )
        output_hidden_states = (
            output_hidden_states
            if output_hidden_states is not None
            else self.config.output_hidden_states
        )
        return_dict = (
            return_dict if return_dict is not None else self.config.use_return_dict
        )
        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.transformer(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
            cache_position=cache_position,
        )

        hidden_states = outputs[0]
        if self.lm_head is None:
            # shared
            logits = ops.dense(
                hidden_states, weight=self.transformer.token_embeddings.weight
            )
        else:
            logits = self.lm_head(hidden_states)
        logits = logits[:, : self.config.vocab_size]
        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :]
            shift_labels = labels[..., 1:]
            # Flatten the tokens
            shift_logits = shift_logits.view(-1, self.config.vocab_size)
            shift_labels = shift_labels.view(-1)
            loss = ops.cross_entropy(shift_logits, shift_labels)

        if not return_dict:
            output = (logits,) + outputs[1:]
            return (loss,) + output if loss is not None else output

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

    def prepare_inputs_for_generation(
        self,
        input_ids,
        past_key_values=None,
        attention_mask=None,
        inputs_embeds=None,
        **kwargs,
    ):
        """
        Prepares the inputs for generation in the OpenELMForCausalLM class.

        Args:
            self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
            input_ids (Tensor): The input tensor of shape [batch_size, sequence_length] containing the token indices.
            past_key_values (Optional[Union[Cache, Tuple[Tensor]]]): The past key-value states.
                If provided, should be either an instance of Cache or a tuple containing tensors. Defaults to None.
            attention_mask (Optional[Tensor]): The attention mask tensor of shape [batch_size, sequence_length].
                If provided, it masks the attention scores. Defaults to None.
            inputs_embeds (Optional[Tensor]): The embedded inputs tensor of shape
                [batch_size, sequence_length, hidden_size]. If provided, it replaces input_ids. Defaults to None.

        Returns:
            model_inputs (Dict[str, Tensor]): A dictionary containing the model inputs for generation.
                It has the following keys:

                - 'inputs_embeds' (Tensor): The embedded inputs tensor.
                It is included if inputs_embeds is provided and past_key_values is None.
                - 'input_ids' (Tensor): The input tensor with token indices.
                It is included if inputs_embeds is None or past_key_values is not None.
                - 'position_ids' (Tensor): The token position indices tensor of shape [batch_size, sequence_length].
                - 'cache_position' (Tensor): The tensor containing the positions for caching of shape [sequence_length].
                - 'past_key_values' (Union[Cache, Tuple[Tensor]]): The past key-value states.
                - 'use_cache' (Optional[bool]): Whether to use cache for generation. Defaults to None.
                - 'attention_mask' (Optional[Tensor]): The attention mask tensor of shape [batch_size, sequence_length].
                It is included if attention_mask is provided.

        Raises:
            None
        """
        past_length = 0
        if past_key_values is not None:
            if isinstance(past_key_values, Cache):
                cache_length = past_key_values.get_seq_length()
                past_length = past_key_values.seen_tokens
                max_cache_length = past_key_values.get_max_length()
            else:
                cache_length = past_length = past_key_values[0][0].shape[2]
                max_cache_length = None

            # Keep only the unprocessed tokens:
            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
            # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
            # input)
            if (
                attention_mask is not None
                and attention_mask.shape[1] > input_ids.shape[1]
            ):
                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
            # input_ids based on the past_length.
            elif past_length < input_ids.shape[1]:
                input_ids = input_ids[:, past_length:]
            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.

            # If we are about to go beyond the maximum cache length, we need to crop the input attention mask.
            if (
                max_cache_length is not None
                and attention_mask is not None
                and cache_length + input_ids.shape[1] > max_cache_length
            ):
                attention_mask = attention_mask[:, -max_cache_length:]

        position_ids = kwargs.get("position_ids", None)
        if attention_mask is not None and position_ids is None:
            # create position_ids on the fly for batch generation
            position_ids = attention_mask.long().cumsum(-1) - 1
            position_ids = position_ids.masked_fill(attention_mask == 0, 1)
            if past_key_values:
                position_ids = position_ids[:, -input_ids.shape[1] :]

        if self.generation_config.cache_implementation == "static":
            # generation with static cache
            cache_position = kwargs.get("cache_position", None)
            if cache_position is None:
                past_length = 0
            else:
                past_length = cache_position[-1] + 1
            input_ids = input_ids[:, past_length:]
            position_ids = position_ids[:, past_length:]

        # we should only keep a `cache_position` in generate, and do +=1.
        # same goes for position ids. Could also help with continued generation.
        cache_position = ops.arange(
            past_length,
            past_length + position_ids.shape[-1],
        )

        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
        if inputs_embeds is not None and past_key_values is None:
            model_inputs = {"inputs_embeds": inputs_embeds}
        else:
            # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise
            # recompiles graphs as the stride of the inputs is a guard. Ref: https://github.com/huggingface/transformers/pull/29114
            # We could use `next_tokens` directly instead.
            model_inputs = {"input_ids": input_ids}

        model_inputs.update(
            {
                "position_ids": position_ids,
                "cache_position": cache_position,
                "past_key_values": past_key_values,
                "use_cache": kwargs.get("use_cache"),
                "attention_mask": attention_mask,
            }
        )
        return model_inputs

    @staticmethod
    def _reorder_cache(past_key_values, beam_idx):
        """
        Reorders the cache of past key values based on the provided beam index.

        Args:
            past_key_values (tuple): A tuple containing the past key values for each layer.
                Each element in the tuple represents the past key values for a layer.
            beam_idx (tensor): An index tensor specifying the order in which the cache should be reordered.

        Returns:
            None: This method does not return any value. Instead, it modifies the 'past_key_values' in place.

        Raises:
            ValueError: If the 'beam_idx' tensor is not valid or if the dimensions of 'past_key_values'
                are not as expected.
            IndexError: If the index specified in 'beam_idx' is out of range for the 'past_key_values'.
        """
        reordered_past = ()
        for layer_past in past_key_values:
            reordered_past += (
                tuple(
                    past_state.index_select(0, beam_idx)
                    for past_state in layer_past
                ),
            )
        return reordered_past

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.__init__(config)

Initializes an instance of the OpenELMForCausalLM class.

PARAMETER	DESCRIPTION
self	The current object instance.
config	An instance of OpenELMConfig class containing the configuration settings for the OpenELM model. TYPE: OpenELMConfig

RETURNS	DESCRIPTION
	None

This method initializes the OpenELMForCausalLM object by setting its configuration, transformer, vocab_size, and lm_head attributes. The config parameter is an instance of OpenELMConfig class and is required to configure the OpenELM model.

ATTRIBUTE	DESCRIPTION
self.transformer	An instance of the OpenELMModel class.
self.vocab_size	An integer representing the size of the vocabulary used in the model.
self.lm_head	An instance of the nn.Linear class or None depending on the value of config.share_input_output_layers.

Note

The OpenELMModel and nn.Linear classes are imported from the appropriate libraries.

Example

>>> config = OpenELMConfig(vocab_size=10000, share_input_output_layers=False)
>>> open_elm = OpenELMForCausalLM(config)

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(self, config: OpenELMConfig):
    """
    Initializes an instance of the OpenELMForCausalLM class.

    Args:
        self: The current object instance.
        config (OpenELMConfig): An instance of OpenELMConfig class containing the configuration settings
            for the OpenELM model.

    Returns:
        None

    Raises:
        None

    This method initializes the OpenELMForCausalLM object by setting its configuration, transformer, vocab_size,
    and lm_head attributes. The config parameter is an instance of OpenELMConfig class and is required to configure
    the OpenELM model.

    Attributes:
        self.transformer: An instance of the OpenELMModel class.
        self.vocab_size: An integer representing the size of the vocabulary used in the model.
        self.lm_head: An instance of the nn.Linear class or None depending on the value of
            config.share_input_output_layers.

    Note:
        The OpenELMModel and nn.Linear classes are imported from the appropriate libraries.

    Example:
        ```python
        >>> config = OpenELMConfig(vocab_size=10000, share_input_output_layers=False)
        >>> open_elm = OpenELMForCausalLM(config)
        ```
    """
    super().__init__(config)
    self.transformer = OpenELMModel(config)
    self.vocab_size = config.vocab_size
    if config.share_input_output_layers:
        self.lm_head = None
    else:
        self.lm_head = nn.Linear(config.model_dim, config.vocab_size, bias=False)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.forward(input_ids=None, attention_mask=None, position_ids=None, past_key_values=None, inputs_embeds=None, labels=None, use_cache=None, output_attentions=None, output_hidden_states=None, return_dict=None, cache_position=None)

This method forwards a Causal Language Model for OpenELM.

PARAMETER	DESCRIPTION
self	The instance of the class. TYPE: object
input_ids	The input tensor containing token IDs. Default is None. TYPE: Tensor DEFAULT: None
attention_mask	An optional tensor for masking tokens. Default is None. TYPE: Tensor DEFAULT: None
position_ids	An optional tensor containing position IDs. Default is None. TYPE: Tensor DEFAULT: None
past_key_values	A list of tensors representing past key values. Default is None. TYPE: List[Tensor] DEFAULT: None
inputs_embeds	An optional tensor of input embeddings. Default is None. TYPE: Tensor DEFAULT: None
labels	An optional tensor containing labels. Default is None. TYPE: Tensor DEFAULT: None
use_cache	A flag indicating whether to use cache. Default is None. TYPE: bool DEFAULT: None
output_attentions	A flag indicating whether to output attentions. Default is None. TYPE: bool DEFAULT: None
output_hidden_states	A flag indicating whether to output hidden states. Default is None. TYPE: bool DEFAULT: None
return_dict	A flag indicating whether to return a dictionary. Default is None. TYPE: bool DEFAULT: None
cache_position	An optional tensor for cache position. Default is None. TYPE: Tensor DEFAULT: None

RETURNS	DESCRIPTION
Union[Tuple, CausalLMOutputWithPast]	Union[Tuple, CausalLMOutputWithPast]: The output of the method, which can be a tuple or an instance of CausalLMOutputWithPast. If return_dict is False, the return value includes loss, logits, past key values, hidden states, and attentions.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(
    self,
    input_ids: mindspore.Tensor = None,
    attention_mask: Optional[mindspore.Tensor] = None,
    position_ids: Optional[mindspore.Tensor] = None,
    past_key_values: Optional[List[mindspore.Tensor]] = None,
    inputs_embeds: Optional[mindspore.Tensor] = None,
    labels: Optional[mindspore.Tensor] = None,
    use_cache: Optional[bool] = None,
    output_attentions: Optional[bool] = None,
    output_hidden_states: Optional[bool] = None,
    return_dict: Optional[bool] = None,
    cache_position: Optional[mindspore.Tensor] = None,
) -> Union[Tuple, CausalLMOutputWithPast]:
    """
    This method forwards a Causal Language Model for OpenELM.

    Args:
        self (object): The instance of the class.
        input_ids (mindspore.Tensor, optional): The input tensor containing token IDs. Default is None.
        attention_mask (mindspore.Tensor, optional): An optional tensor for masking tokens. Default is None.
        position_ids (mindspore.Tensor, optional): An optional tensor containing position IDs. Default is None.
        past_key_values (List[mindspore.Tensor], optional): A list of tensors representing past key values.
            Default is None.
        inputs_embeds (mindspore.Tensor, optional): An optional tensor of input embeddings. Default is None.
        labels (mindspore.Tensor, optional): An optional tensor containing labels. Default is None.
        use_cache (bool, optional): A flag indicating whether to use cache. Default is None.
        output_attentions (bool, optional): A flag indicating whether to output attentions. Default is None.
        output_hidden_states (bool, optional): A flag indicating whether to output hidden states. Default is None.
        return_dict (bool, optional): A flag indicating whether to return a dictionary. Default is None.
        cache_position (mindspore.Tensor, optional): An optional tensor for cache position. Default is None.

    Returns:
        Union[Tuple, CausalLMOutputWithPast]: The output of the method, which can be a tuple or an instance of
            CausalLMOutputWithPast. If return_dict is False, the return value includes loss, logits,
            past key values, hidden states, and attentions.

    Raises:
        None
    """
    output_attentions = (
        output_attentions
        if output_attentions is not None
        else self.config.output_attentions
    )
    output_hidden_states = (
        output_hidden_states
        if output_hidden_states is not None
        else self.config.output_hidden_states
    )
    return_dict = (
        return_dict if return_dict is not None else self.config.use_return_dict
    )
    # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
    outputs = self.transformer(
        input_ids=input_ids,
        attention_mask=attention_mask,
        position_ids=position_ids,
        past_key_values=past_key_values,
        inputs_embeds=inputs_embeds,
        use_cache=use_cache,
        output_attentions=output_attentions,
        output_hidden_states=output_hidden_states,
        return_dict=return_dict,
        cache_position=cache_position,
    )

    hidden_states = outputs[0]
    if self.lm_head is None:
        # shared
        logits = ops.dense(
            hidden_states, weight=self.transformer.token_embeddings.weight
        )
    else:
        logits = self.lm_head(hidden_states)
    logits = logits[:, : self.config.vocab_size]
    loss = None
    if labels is not None:
        # Shift so that tokens < n predict n
        shift_logits = logits[..., :-1, :]
        shift_labels = labels[..., 1:]
        # Flatten the tokens
        shift_logits = shift_logits.view(-1, self.config.vocab_size)
        shift_labels = shift_labels.view(-1)
        loss = ops.cross_entropy(shift_logits, shift_labels)

    if not return_dict:
        output = (logits,) + outputs[1:]
        return (loss,) + output if loss is not None else output

    return CausalLMOutputWithPast(
        loss=loss,
        logits=logits,
        past_key_values=outputs.past_key_values,
        hidden_states=outputs.hidden_states,
        attentions=outputs.attentions,
    )

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.get_decoder()

This method returns the transformer for OpenELMForCausalLM.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMForCausalLM class. TYPE: object

RETURNS	DESCRIPTION
None	This method returns the transformer object associated with the OpenELMForCausalLM instance.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def get_decoder(self):
    """
    This method returns the transformer for OpenELMForCausalLM.

    Args:
        self (object): The instance of the OpenELMForCausalLM class.

    Returns:
        None: This method returns the transformer object associated with the OpenELMForCausalLM instance.

    Raises:
        None
    """
    return self.transformer

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.get_input_embeddings()

Retrieve the input embeddings from the OpenELMForCausalLM model.

PARAMETER	DESCRIPTION
self	The instance of OpenELMForCausalLM. TYPE: OpenELMForCausalLM

RETURNS	DESCRIPTION
token_embeddings	This method returns the input embeddings as a transformer token embeddings.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def get_input_embeddings(self):
    """
    Retrieve the input embeddings from the OpenELMForCausalLM model.

    Args:
        self (OpenELMForCausalLM): The instance of OpenELMForCausalLM.

    Returns:
        token_embeddings: This method returns the input embeddings as a transformer token embeddings.

    Raises:
        None.
    """
    return self.transformer.token_embeddings

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.get_output_embeddings()

Returns the output embeddings of the OpenELMForCausalLM model.

PARAMETER	DESCRIPTION
self	An instance of the OpenELMForCausalLM class.

RETURNS	DESCRIPTION
lm_head	The method returns the output embeddings of the OpenELMForCausalLM model.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def get_output_embeddings(self):
    """
    Returns the output embeddings of the OpenELMForCausalLM model.

    Args:
        self: An instance of the OpenELMForCausalLM class.

    Returns:
        lm_head: The method returns the output embeddings of the OpenELMForCausalLM model.

    Raises:
        None.
    """
    return self.lm_head

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.prepare_inputs_for_generation(input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs)

Prepares the inputs for generation in the OpenELMForCausalLM class.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMForCausalLM class. TYPE: OpenELMForCausalLM
input_ids	The input tensor of shape [batch_size, sequence_length] containing the token indices. TYPE: Tensor
past_key_values	The past key-value states. If provided, should be either an instance of Cache or a tuple containing tensors. Defaults to None. TYPE: Optional[Union[Cache, Tuple[Tensor]]] DEFAULT: None
attention_mask	The attention mask tensor of shape [batch_size, sequence_length]. If provided, it masks the attention scores. Defaults to None. TYPE: Optional[Tensor] DEFAULT: None
inputs_embeds	The embedded inputs tensor of shape [batch_size, sequence_length, hidden_size]. If provided, it replaces input_ids. Defaults to None. TYPE: Optional[Tensor] DEFAULT: None

RETURNS

DESCRIPTION

model_inputs

A dictionary containing the model inputs for generation. It has the following keys: 'inputs_embeds' (Tensor): The embedded inputs tensor. It is included if inputs_embeds is provided and past_key_values is None. 'input_ids' (Tensor): The input tensor with token indices. It is included if inputs_embeds is None or past_key_values is not None. 'position_ids' (Tensor): The token position indices tensor of shape [batch_size, sequence_length]. 'cache_position' (Tensor): The tensor containing the positions for caching of shape [sequence_length]. 'past_key_values' (Union[Cache, Tuple[Tensor]]): The past key-value states. 'use_cache' (Optional[bool]): Whether to use cache for generation. Defaults to None. 'attention_mask' (Optional[Tensor]): The attention mask tensor of shape [batch_size, sequence_length]. It is included if attention_mask is provided. TYPE: Dict[str, Tensor]

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def prepare_inputs_for_generation(
    self,
    input_ids,
    past_key_values=None,
    attention_mask=None,
    inputs_embeds=None,
    **kwargs,
):
    """
    Prepares the inputs for generation in the OpenELMForCausalLM class.

    Args:
        self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
        input_ids (Tensor): The input tensor of shape [batch_size, sequence_length] containing the token indices.
        past_key_values (Optional[Union[Cache, Tuple[Tensor]]]): The past key-value states.
            If provided, should be either an instance of Cache or a tuple containing tensors. Defaults to None.
        attention_mask (Optional[Tensor]): The attention mask tensor of shape [batch_size, sequence_length].
            If provided, it masks the attention scores. Defaults to None.
        inputs_embeds (Optional[Tensor]): The embedded inputs tensor of shape
            [batch_size, sequence_length, hidden_size]. If provided, it replaces input_ids. Defaults to None.

    Returns:
        model_inputs (Dict[str, Tensor]): A dictionary containing the model inputs for generation.
            It has the following keys:

            - 'inputs_embeds' (Tensor): The embedded inputs tensor.
            It is included if inputs_embeds is provided and past_key_values is None.
            - 'input_ids' (Tensor): The input tensor with token indices.
            It is included if inputs_embeds is None or past_key_values is not None.
            - 'position_ids' (Tensor): The token position indices tensor of shape [batch_size, sequence_length].
            - 'cache_position' (Tensor): The tensor containing the positions for caching of shape [sequence_length].
            - 'past_key_values' (Union[Cache, Tuple[Tensor]]): The past key-value states.
            - 'use_cache' (Optional[bool]): Whether to use cache for generation. Defaults to None.
            - 'attention_mask' (Optional[Tensor]): The attention mask tensor of shape [batch_size, sequence_length].
            It is included if attention_mask is provided.

    Raises:
        None
    """
    past_length = 0
    if past_key_values is not None:
        if isinstance(past_key_values, Cache):
            cache_length = past_key_values.get_seq_length()
            past_length = past_key_values.seen_tokens
            max_cache_length = past_key_values.get_max_length()
        else:
            cache_length = past_length = past_key_values[0][0].shape[2]
            max_cache_length = None

        # Keep only the unprocessed tokens:
        # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
        # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
        # input)
        if (
            attention_mask is not None
            and attention_mask.shape[1] > input_ids.shape[1]
        ):
            input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
        # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
        # input_ids based on the past_length.
        elif past_length < input_ids.shape[1]:
            input_ids = input_ids[:, past_length:]
        # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.

        # If we are about to go beyond the maximum cache length, we need to crop the input attention mask.
        if (
            max_cache_length is not None
            and attention_mask is not None
            and cache_length + input_ids.shape[1] > max_cache_length
        ):
            attention_mask = attention_mask[:, -max_cache_length:]

    position_ids = kwargs.get("position_ids", None)
    if attention_mask is not None and position_ids is None:
        # create position_ids on the fly for batch generation
        position_ids = attention_mask.long().cumsum(-1) - 1
        position_ids = position_ids.masked_fill(attention_mask == 0, 1)
        if past_key_values:
            position_ids = position_ids[:, -input_ids.shape[1] :]

    if self.generation_config.cache_implementation == "static":
        # generation with static cache
        cache_position = kwargs.get("cache_position", None)
        if cache_position is None:
            past_length = 0
        else:
            past_length = cache_position[-1] + 1
        input_ids = input_ids[:, past_length:]
        position_ids = position_ids[:, past_length:]

    # we should only keep a `cache_position` in generate, and do +=1.
    # same goes for position ids. Could also help with continued generation.
    cache_position = ops.arange(
        past_length,
        past_length + position_ids.shape[-1],
    )

    # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
    if inputs_embeds is not None and past_key_values is None:
        model_inputs = {"inputs_embeds": inputs_embeds}
    else:
        # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise
        # recompiles graphs as the stride of the inputs is a guard. Ref: https://github.com/huggingface/transformers/pull/29114
        # We could use `next_tokens` directly instead.
        model_inputs = {"input_ids": input_ids}

    model_inputs.update(
        {
            "position_ids": position_ids,
            "cache_position": cache_position,
            "past_key_values": past_key_values,
            "use_cache": kwargs.get("use_cache"),
            "attention_mask": attention_mask,
        }
    )
    return model_inputs

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.set_decoder(decoder)

Set the decoder for the OpenELMForCausalLM instance.

This method allows setting the decoder for the OpenELMForCausalLM instance. The decoder is used to transform the input data.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMForCausalLM class. TYPE: OpenELMForCausalLM
decoder	The decoder to be set. It should be compatible with the OpenELMForCausalLM instance.

RETURNS	DESCRIPTION
	None.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def set_decoder(self, decoder):
    """Set the decoder for the OpenELMForCausalLM instance.

    This method allows setting the decoder for the OpenELMForCausalLM instance.
    The decoder is used to transform the input data.

    Args:
        self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
        decoder: The decoder to be set. It should be compatible with the OpenELMForCausalLM instance.

    Returns:
        None.

    Raises:
        None.
    """
    self.transformer = decoder

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.set_input_embeddings(value)

Sets the input embeddings for the OpenELMForCausalLM model.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMForCausalLM class. TYPE: OpenELMForCausalLM
value	The input embeddings to be set for the model. It should be a torch.Tensor object. TYPE: Tensor

RETURNS	DESCRIPTION
	None.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def set_input_embeddings(self, value):
    """
    Sets the input embeddings for the OpenELMForCausalLM model.

    Args:
        self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
        value (torch.Tensor): The input embeddings to be set for the model. It should be a torch.Tensor object.

    Returns:
        None.

    Raises:
        None.
    """
    self.transformer.token_embeddings = value

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMForCausalLM.set_output_embeddings(new_embeddings)

This method sets the output embeddings for the OpenELMForCausalLM class.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMForCausalLM class. TYPE: OpenELMForCausalLM
new_embeddings	The new output embeddings to be set for the OpenELMForCausalLM instance. TYPE: object

RETURNS	DESCRIPTION
	None.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def set_output_embeddings(self, new_embeddings):
    """
    This method sets the output embeddings for the OpenELMForCausalLM class.

    Args:
        self (OpenELMForCausalLM): The instance of the OpenELMForCausalLM class.
        new_embeddings (object): The new output embeddings to be set for the OpenELMForCausalLM instance.

    Returns:
        None.

    Raises:
        None.
    """
    self.lm_head = new_embeddings

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMModel

Bases: OpenELMPreTrainedModel

This class represents an OpenELM model for natural language processing tasks. It is designed to be used for tasks such as language modeling, text generation, and machine translation. The model architecture includes a transformer-based decoder with customizable layers and attention mechanisms.

The OpenELMModel class provides methods for initializing the model with configuration settings, accessing and updating input embeddings, and forwarding the model for inference or training. The forward method handles the main computation flow of the model, including processing input data, applying transformer layers, and generating model outputs. The class also includes helper methods for managing cache, attention masks, and normalization.

The OpenELMModel class is designed to be flexible and efficient, allowing for easy customization of the model architecture and behavior. It inherits from the OpenELMPreTrainedModel class, which provides additional functionality and pre-trained model weights.

For detailed information on each method and parameter, refer to the method docstrings within the class implementation.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMModel(OpenELMPreTrainedModel):

    """
    This class represents an OpenELM model for natural language processing tasks.
    It is designed to be used for tasks such as language modeling, text generation, and machine translation.
    The model architecture includes a transformer-based decoder with customizable layers and attention mechanisms.

    The OpenELMModel class provides methods for initializing the model with configuration settings, accessing and
    updating input embeddings, and forwarding the model for inference or training.
    The forward method handles the main computation flow of the model, including processing input data, applying
    transformer layers, and generating model outputs. The class also includes helper methods for managing cache,
    attention masks, and normalization.

    The OpenELMModel class is designed to be flexible and efficient, allowing for easy customization of the
    model architecture and behavior. It inherits from the OpenELMPreTrainedModel class, which provides
    additional functionality and pre-trained model weights.

    For detailed information on each method and parameter, refer to the method docstrings within the
    class implementation.
    """
    config_class = OpenELMConfig

    def __init__(self, config: OpenELMConfig):
        """
        Initializes an instance of the OpenELMModel class.

        Args:
            self: The instance of the class.
            config (OpenELMConfig):
                The configuration object containing the model settings.

                - Type: OpenELMConfig
                - Purpose: Specifies the parameters for the OpenELMModel.
                - Restrictions: Must be of type OpenELMConfig.

        Returns:
            None

        Raises:
            None
        """
        super().__init__(config)
        self.config = config

        self.token_embeddings = nn.Embedding(
            embedding_size=config.model_dim,
            vocab_size=config.vocab_size,
        )

        self.layers = nn.ModuleList([
            OpenELMDecoderLayer(config=config, layer_idx=layer_idx)
            for layer_idx in range(config.num_transformer_layers)]
        )
        self.norm = OpenELMRMSNorm(num_features=config.model_dim)
        if config.share_input_output_layers:
            self.classifier = None
        else:
            self.classifier = nn.Linear(
                in_channels=config.model_dim,
                out_channels=config.vocab_size,
                bias=False,
            )
        self.num_transformer_layers = config.num_transformer_layers
        self.gradient_checkpointing = False

        # Register a causal mask to separate causal and padding mask creation. Merging happens in the attention class.
        # NOTE: This is not friendly with TorchScript, ONNX, ExportedProgram serialization for very large `max_context_length`.
        causal_mask = ops.full(
            (config.max_context_length, config.max_context_length),
            fill_value=True,
            dtype=mindspore.bool_,
        )
        self.causal_mask = ops.triu(causal_mask, diagonal=1)

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        """
        Returns the input embeddings for the OpenELMModel.

        Args:
            self (OpenELMModel): An instance of the OpenELMModel class.

        Returns:
            None.

        Raises:
            None.

        This method retrieves the input embeddings from the OpenELMModel.
        The input embeddings are obtained from the token embeddings of the model.
        The token embeddings are stored in the `token_embeddings` attribute of the OpenELMModel instance.
        The input embeddings are used as input for further processing or analysis in the OpenELMModel.
        """
        return self.token_embeddings

    def set_input_embeddings(self, new_embeddings: mindspore.Tensor):
        """
        Set the input embeddings for the OpenELMModel.

        Args:
            self (OpenELMModel): The instance of the OpenELMModel class.
            new_embeddings (mindspore.Tensor): A tensor containing the new embeddings to be set as input.

        Returns:
            None.

        Raises:
            None.
        """
        self.token_embeddings = new_embeddings

    def forward(
        self,
        input_ids: mindspore.Tensor = None,
        attention_mask: Optional[mindspore.Tensor] = None,
        position_ids: Optional[mindspore.Tensor] = None,
        past_key_values: Optional[List[mindspore.Tensor]] = None,
        inputs_embeds: Optional[mindspore.Tensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        cache_position: Optional[mindspore.Tensor] = None,
    ) -> Union[Tuple, BaseModelOutputWithPast]:
        """
        Constructs the OpenELMModel.

        Args:
            self (OpenELMModel): The instance of the OpenELMModel class.
            input_ids (mindspore.Tensor, optional): The input tensor ids. Default: None.
            attention_mask (mindspore.Tensor, optional): The attention mask tensor. Default: None.
            position_ids (mindspore.Tensor, optional): The position ids tensor. Default: None.
            past_key_values (List[mindspore.Tensor], optional): The list of past key value tensors. Default: None.
            inputs_embeds (mindspore.Tensor, optional): The input embeddings tensor. Default: None.
            use_cache (bool, optional): Whether to use cache. Default: None.
            output_attentions (bool, optional): Whether to output attentions. Default: None.
            output_hidden_states (bool, optional): Whether to output hidden states. Default: None.
            return_dict (bool, optional): Whether to return a dictionary. Default: None.
            cache_position (mindspore.Tensor, optional): The cache position tensor. Default: None.

        Returns:
            Union[Tuple, BaseModelOutputWithPast]: The output tuple or BaseModelOutputWithPast object.

        Raises:
            ValueError: If both input_ids and inputs_embeds are specified or neither is specified.
            Warning: If use_cache=True is incompatible with gradient checkpointing.

        """
        output_attentions = (
            output_attentions
            if output_attentions is not None
            else self.config.output_attentions
        )
        output_hidden_states = (
            output_hidden_states
            if output_hidden_states is not None
            else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = (
            return_dict if return_dict is not None else self.config.use_return_dict
        )

        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 self.gradient_checkpointing and self.training and use_cache:
            logger.warning_once(
                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
            )
            use_cache = False

        if inputs_embeds is None:
            inputs_embeds = self.token_embeddings(input_ids)

        past_seen_tokens = 0
        if use_cache:  # kept for BC (cache positions)
            if not isinstance(past_key_values, StaticCache):
                past_key_values = DynamicCache.from_legacy_cache(past_key_values)
            past_seen_tokens = past_key_values.get_seq_length()

        if cache_position is None:
            cache_position = ops.arange(
                past_seen_tokens,
                past_seen_tokens + inputs_embeds.shape[1],
            )

        if position_ids is None:
            position_ids = cache_position.unsqueeze(0)

        causal_mask = self._update_causal_mask(attention_mask, inputs_embeds)

        # embed positions
        hidden_states = inputs_embeds

        # decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None
        next_decoder_cache = None

        for decoder_layer in self.layers:
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            if self.gradient_checkpointing and self.training:
                layer_outputs = self._gradient_checkpointing_func(
                    decoder_layer.__call__,
                    hidden_states,
                    causal_mask,
                    position_ids,
                    past_key_values,
                    output_attentions,
                    use_cache,
                    cache_position,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=causal_mask,
                    position_ids=position_ids,
                    past_key_value=past_key_values,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                    cache_position=cache_position,
                )

            hidden_states = layer_outputs[0]

            if use_cache:
                next_decoder_cache = layer_outputs[2 if output_attentions else 1]

            if output_attentions:
                all_self_attns += (layer_outputs[1],)

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        next_cache = None
        if use_cache:
            next_cache = (
                next_decoder_cache.to_legacy_cache()
                if isinstance(next_decoder_cache, Cache)
                else next_decoder_cache
            )
        if not return_dict:
            return tuple(
                v
                for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
                if v is not None
            )
        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=next_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
        )

    def _update_causal_mask(self, attention_mask, input_tensor):
        """
        Updates the causal mask used in the OpenELMModel for attention computations.

        Args:
            self (OpenELMModel): The instance of the OpenELMModel class.
            attention_mask (torch.Tensor): A tensor containing the attention mask.
                This mask is used to mask certain positions of the input tensor during attention computations.
                If the `_attn_implementation` attribute of the `config` object is set to 'flash_attention_2' and
                the attention_mask contains a 0.0 value, the attention_mask is returned as is. Otherwise,
                if the attention_mask is not provided or does not contain a 0.0 value, it is set to None.
            input_tensor (torch.Tensor): The input tensor to the model.
                It has shape (batch_size, seq_length) and represents the input sequences.

        Returns:
            None: The method updates the causal_mask attribute of the OpenELMModel instance in-place.

        Raises:
            None.
        """
        if self.config._attn_implementation == "flash_attention_2":
            if attention_mask is not None and 0.0 in attention_mask:
                return attention_mask
            return None

        batch_size, seq_length = input_tensor.shape[:2]
        dtype = input_tensor.dtype

        # support going beyond cached `max_position_embedding`
        if seq_length > self.causal_mask.shape[-1]:
            causal_mask = ops.full(
                (2 * self.causal_mask.shape[-1], 2 * self.causal_mask.shape[-1]),
                fill_value=1,
            )
            self.causal_mask = ops.triu(causal_mask, diagonal=1)

        # We use the current dtype to avoid any overflows
        min_dtype = finfo(dtype, 'min')
        causal_mask = (
            self.causal_mask[None, None, :, :].repeat(batch_size, 1, 1, 1).to(dtype)
            * min_dtype
        )

        causal_mask = causal_mask.to(dtype=dtype)
        if attention_mask is not None and attention_mask.ndim == 2:
            mask_length = attention_mask.shape[-1]
            padding_mask = causal_mask[..., :mask_length].eq(0.0) & attention_mask[
                :, None, None, :
            ].eq(0.0)
            causal_mask[..., :mask_length] = causal_mask[..., :mask_length].masked_fill(
                padding_mask, min_dtype
            )

        return causal_mask

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMModel.__init__(config)

Initializes an instance of the OpenELMModel class.

PARAMETER	DESCRIPTION
self	The instance of the class.
config	The configuration object containing the model settings. Type: OpenELMConfig Purpose: Specifies the parameters for the OpenELMModel. Restrictions: Must be of type OpenELMConfig. TYPE: OpenELMConfig

RETURNS	DESCRIPTION
	None

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(self, config: OpenELMConfig):
    """
    Initializes an instance of the OpenELMModel class.

    Args:
        self: The instance of the class.
        config (OpenELMConfig):
            The configuration object containing the model settings.

            - Type: OpenELMConfig
            - Purpose: Specifies the parameters for the OpenELMModel.
            - Restrictions: Must be of type OpenELMConfig.

    Returns:
        None

    Raises:
        None
    """
    super().__init__(config)
    self.config = config

    self.token_embeddings = nn.Embedding(
        embedding_size=config.model_dim,
        vocab_size=config.vocab_size,
    )

    self.layers = nn.ModuleList([
        OpenELMDecoderLayer(config=config, layer_idx=layer_idx)
        for layer_idx in range(config.num_transformer_layers)]
    )
    self.norm = OpenELMRMSNorm(num_features=config.model_dim)
    if config.share_input_output_layers:
        self.classifier = None
    else:
        self.classifier = nn.Linear(
            in_channels=config.model_dim,
            out_channels=config.vocab_size,
            bias=False,
        )
    self.num_transformer_layers = config.num_transformer_layers
    self.gradient_checkpointing = False

    # Register a causal mask to separate causal and padding mask creation. Merging happens in the attention class.
    # NOTE: This is not friendly with TorchScript, ONNX, ExportedProgram serialization for very large `max_context_length`.
    causal_mask = ops.full(
        (config.max_context_length, config.max_context_length),
        fill_value=True,
        dtype=mindspore.bool_,
    )
    self.causal_mask = ops.triu(causal_mask, diagonal=1)

    # Initialize weights and apply final processing
    self.post_init()

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMModel.forward(input_ids=None, attention_mask=None, position_ids=None, past_key_values=None, inputs_embeds=None, use_cache=None, output_attentions=None, output_hidden_states=None, return_dict=None, cache_position=None)

Constructs the OpenELMModel.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMModel class. TYPE: OpenELMModel
input_ids	The input tensor ids. Default: None. TYPE: Tensor DEFAULT: None
attention_mask	The attention mask tensor. Default: None. TYPE: Tensor DEFAULT: None
position_ids	The position ids tensor. Default: None. TYPE: Tensor DEFAULT: None
past_key_values	The list of past key value tensors. Default: None. TYPE: List[Tensor] DEFAULT: None
inputs_embeds	The input embeddings tensor. Default: None. TYPE: Tensor DEFAULT: None
use_cache	Whether to use cache. Default: None. TYPE: bool DEFAULT: None
output_attentions	Whether to output attentions. Default: None. TYPE: bool DEFAULT: None
output_hidden_states	Whether to output hidden states. Default: None. TYPE: bool DEFAULT: None
return_dict	Whether to return a dictionary. Default: None. TYPE: bool DEFAULT: None
cache_position	The cache position tensor. Default: None. TYPE: Tensor DEFAULT: None

RETURNS	DESCRIPTION
Union[Tuple, BaseModelOutputWithPast]	Union[Tuple, BaseModelOutputWithPast]: The output tuple or BaseModelOutputWithPast object.

RAISES	DESCRIPTION
ValueError	If both input_ids and inputs_embeds are specified or neither is specified.
Warning	If use_cache=True is incompatible with gradient checkpointing.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(
    self,
    input_ids: mindspore.Tensor = None,
    attention_mask: Optional[mindspore.Tensor] = None,
    position_ids: Optional[mindspore.Tensor] = None,
    past_key_values: Optional[List[mindspore.Tensor]] = None,
    inputs_embeds: Optional[mindspore.Tensor] = None,
    use_cache: Optional[bool] = None,
    output_attentions: Optional[bool] = None,
    output_hidden_states: Optional[bool] = None,
    return_dict: Optional[bool] = None,
    cache_position: Optional[mindspore.Tensor] = None,
) -> Union[Tuple, BaseModelOutputWithPast]:
    """
    Constructs the OpenELMModel.

    Args:
        self (OpenELMModel): The instance of the OpenELMModel class.
        input_ids (mindspore.Tensor, optional): The input tensor ids. Default: None.
        attention_mask (mindspore.Tensor, optional): The attention mask tensor. Default: None.
        position_ids (mindspore.Tensor, optional): The position ids tensor. Default: None.
        past_key_values (List[mindspore.Tensor], optional): The list of past key value tensors. Default: None.
        inputs_embeds (mindspore.Tensor, optional): The input embeddings tensor. Default: None.
        use_cache (bool, optional): Whether to use cache. Default: None.
        output_attentions (bool, optional): Whether to output attentions. Default: None.
        output_hidden_states (bool, optional): Whether to output hidden states. Default: None.
        return_dict (bool, optional): Whether to return a dictionary. Default: None.
        cache_position (mindspore.Tensor, optional): The cache position tensor. Default: None.

    Returns:
        Union[Tuple, BaseModelOutputWithPast]: The output tuple or BaseModelOutputWithPast object.

    Raises:
        ValueError: If both input_ids and inputs_embeds are specified or neither is specified.
        Warning: If use_cache=True is incompatible with gradient checkpointing.

    """
    output_attentions = (
        output_attentions
        if output_attentions is not None
        else self.config.output_attentions
    )
    output_hidden_states = (
        output_hidden_states
        if output_hidden_states is not None
        else self.config.output_hidden_states
    )
    use_cache = use_cache if use_cache is not None else self.config.use_cache
    return_dict = (
        return_dict if return_dict is not None else self.config.use_return_dict
    )

    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 self.gradient_checkpointing and self.training and use_cache:
        logger.warning_once(
            "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
        )
        use_cache = False

    if inputs_embeds is None:
        inputs_embeds = self.token_embeddings(input_ids)

    past_seen_tokens = 0
    if use_cache:  # kept for BC (cache positions)
        if not isinstance(past_key_values, StaticCache):
            past_key_values = DynamicCache.from_legacy_cache(past_key_values)
        past_seen_tokens = past_key_values.get_seq_length()

    if cache_position is None:
        cache_position = ops.arange(
            past_seen_tokens,
            past_seen_tokens + inputs_embeds.shape[1],
        )

    if position_ids is None:
        position_ids = cache_position.unsqueeze(0)

    causal_mask = self._update_causal_mask(attention_mask, inputs_embeds)

    # embed positions
    hidden_states = inputs_embeds

    # decoder layers
    all_hidden_states = () if output_hidden_states else None
    all_self_attns = () if output_attentions else None
    next_decoder_cache = None

    for decoder_layer in self.layers:
        if output_hidden_states:
            all_hidden_states += (hidden_states,)

        if self.gradient_checkpointing and self.training:
            layer_outputs = self._gradient_checkpointing_func(
                decoder_layer.__call__,
                hidden_states,
                causal_mask,
                position_ids,
                past_key_values,
                output_attentions,
                use_cache,
                cache_position,
            )
        else:
            layer_outputs = decoder_layer(
                hidden_states,
                attention_mask=causal_mask,
                position_ids=position_ids,
                past_key_value=past_key_values,
                output_attentions=output_attentions,
                use_cache=use_cache,
                cache_position=cache_position,
            )

        hidden_states = layer_outputs[0]

        if use_cache:
            next_decoder_cache = layer_outputs[2 if output_attentions else 1]

        if output_attentions:
            all_self_attns += (layer_outputs[1],)

    hidden_states = self.norm(hidden_states)

    # add hidden states from the last decoder layer
    if output_hidden_states:
        all_hidden_states += (hidden_states,)

    next_cache = None
    if use_cache:
        next_cache = (
            next_decoder_cache.to_legacy_cache()
            if isinstance(next_decoder_cache, Cache)
            else next_decoder_cache
        )
    if not return_dict:
        return tuple(
            v
            for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
            if v is not None
        )
    return BaseModelOutputWithPast(
        last_hidden_state=hidden_states,
        past_key_values=next_cache,
        hidden_states=all_hidden_states,
        attentions=all_self_attns,
    )

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMModel.get_input_embeddings()

Returns the input embeddings for the OpenELMModel.

PARAMETER	DESCRIPTION
self	An instance of the OpenELMModel class. TYPE: OpenELMModel

RETURNS	DESCRIPTION
	None.

This method retrieves the input embeddings from the OpenELMModel. The input embeddings are obtained from the token embeddings of the model. The token embeddings are stored in the token_embeddings attribute of the OpenELMModel instance. The input embeddings are used as input for further processing or analysis in the OpenELMModel.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def get_input_embeddings(self):
    """
    Returns the input embeddings for the OpenELMModel.

    Args:
        self (OpenELMModel): An instance of the OpenELMModel class.

    Returns:
        None.

    Raises:
        None.

    This method retrieves the input embeddings from the OpenELMModel.
    The input embeddings are obtained from the token embeddings of the model.
    The token embeddings are stored in the `token_embeddings` attribute of the OpenELMModel instance.
    The input embeddings are used as input for further processing or analysis in the OpenELMModel.
    """
    return self.token_embeddings

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMModel.set_input_embeddings(new_embeddings)

Set the input embeddings for the OpenELMModel.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMModel class. TYPE: OpenELMModel
new_embeddings	A tensor containing the new embeddings to be set as input. TYPE: Tensor

RETURNS	DESCRIPTION
	None.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def set_input_embeddings(self, new_embeddings: mindspore.Tensor):
    """
    Set the input embeddings for the OpenELMModel.

    Args:
        self (OpenELMModel): The instance of the OpenELMModel class.
        new_embeddings (mindspore.Tensor): A tensor containing the new embeddings to be set as input.

    Returns:
        None.

    Raises:
        None.
    """
    self.token_embeddings = new_embeddings

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMMultiHeadCausalAttention

Bases: Module

This class represents a multi-head causal attention mechanism for OpenELM models. It performs multi-head self-attention computation with optional key and query normalization and caching capabilities.

Inherits from nn.Module, this class provides functionality for processing input tensors through multi-head self-attention mechanism, with support for caching key-value pairs for efficient generation tasks.

The class initializes with configuration parameters and layer index, setting up projection layers, position embeddings, normalization options, and output projection layers. It also defines the number of query, key, and value heads, along with transformer dimensions and grouping information.

The 'forward' method performs the forward pass of multi-head self-attention, taking input hidden states, optional attention mask, cached key-value pairs, and other parameters. It computes queries, keys, and values, applies normalization if configured, updates cached key-value pairs if available, incorporates positional embeddings, and performs scaled dot-product attention calculation. Finally, it applies output projection and returns the attention output along with optional attention weights and updated cached key-value pairs.

Note

This class assumes the existence of certain related classes and functions like OpenELMConfig, OpenELMRotaryEmbedding, OpenELMRMSNorm, Cache, nn.Linear, and _scaled_dot_product_attention.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMMultiHeadCausalAttention(nn.Module):

    """
    This class represents a multi-head causal attention mechanism for OpenELM models.
    It performs multi-head self-attention computation with optional key and query normalization and caching capabilities.

    Inherits from nn.Module, this class provides functionality for processing input tensors through multi-head
    self-attention mechanism, with support for caching key-value pairs for efficient generation tasks.

    The class initializes with configuration parameters and layer index, setting up projection layers,
    position embeddings, normalization options, and output projection layers.
    It also defines the number of query, key, and value heads, along with transformer dimensions and grouping information.

    The 'forward' method performs the forward pass of multi-head self-attention, taking input hidden states,
    optional attention mask, cached key-value pairs, and other parameters.
    It computes queries, keys, and values, applies normalization if configured, updates cached key-value pairs
    if available, incorporates positional embeddings, and performs scaled dot-product attention calculation.
    Finally, it applies output projection and returns the attention output along with optional attention weights and
    updated cached key-value pairs.

    Note:
        This class assumes the existence of certain related classes and functions like OpenELMConfig,
        OpenELMRotaryEmbedding, OpenELMRMSNorm, Cache, nn.Linear, and _scaled_dot_product_attention.
    """
    def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
        '''
        Initializes an instance of the OpenELMMultiHeadCausalAttention class.

        Args:
            self: The instance of the class.
            config (OpenELMConfig): An instance of the OpenELMConfig class containing configuration parameters.
            layer_idx (int): The index of the layer.

        Returns:
            None

        Raises:
            None
        '''
        super().__init__()
        self.layer_idx = layer_idx
        head_dim = config.head_dim
        q_heads = config.num_query_heads[layer_idx]
        k_heads = config.num_kv_heads[layer_idx]
        v_heads = config.num_kv_heads[layer_idx]

        self.qkv_proj = nn.Linear(
            in_channels=config.model_dim,
            out_channels=(q_heads + k_heads + v_heads) * head_dim,
            bias=False,
        )

        self.pos_embedding = OpenELMRotaryEmbedding(
            model_dim=config.head_dim,
            max_seq_length=config.rope_max_length,
            freq_constant=config.rope_freq_constant,
        )

        if config.normalize_qk_projections:
            self.q_norm = OpenELMRMSNorm(
                num_features=config.head_dim,
            )
            self.k_norm = OpenELMRMSNorm(
                num_features=config.head_dim,
            )
        else:
            self.q_norm = None
            self.k_norm = None

        self.out_proj = nn.Linear(
            in_channels=q_heads * head_dim,
            out_channels=config.model_dim,
            bias=False,
        )

        self.head_dim = config.head_dim
        self.num_q_heads = q_heads
        self.num_k_heads = k_heads
        self.num_v_heads = v_heads
        self.transformer_dim = config.model_dim
        self.num_groups = self.num_q_heads // self.num_k_heads

    def extra_repr(self) -> str:
        """
        Returns a string representation of the OpenELMMultiHeadCausalAttention object, including the number of query,
        key, and value heads.

        Args:
            self (OpenELMMultiHeadCausalAttention): The instance of the OpenELMMultiHeadCausalAttention class.

        Returns:
            str: A string representation of the OpenELMMultiHeadCausalAttention object,
                including the number of query, key, and value heads.

        Raises:
            None.

        """
        return (
            super().extra_repr()
            + f"query_heads={self.num_q_heads}, key_heads={self.num_k_heads}, value_heads={self.num_v_heads}"
        )

    def forward(
        self,
        hidden_states: mindspore.Tensor,
        attention_mask: Optional[mindspore.Tensor] = None,
        past_key_value: Optional[Cache] = None,
        output_attentions: bool = False,
        use_cache: bool = False,
        cache_position: Optional[mindspore.Tensor] = None,
    ) -> Tuple[mindspore.Tensor, Optional[mindspore.Tensor], Optional[Tuple[mindspore.Tensor]]]:
        """
        Forward pass of multi-head self-attention.

        Args:
            hidden_states: Input tensor of the shape [batch size, sequence length, model dimension].
            past_key_value: Tensor storing the cached keys and values.
            output_attentions: output attention weights.
            use_cache: Specifies whether to use kv-cache for generation.
            cache_position: used for updating the kv-cache.

        Returns:
            The output of the same shape as the input, optionally with a tensor containing cached keys and values.
        """
        # scaled_dot_product_attention does not return attention weights, set output_attentions to False
        output_attentions = False
        batch_size, seq_length, d_model = hidden_states.shape

        # [B, S, d] --> [B, S, (q_h + k_h + v_h) * h]
        qkv = self.qkv_proj(hidden_states)
        # [B, S, (q_h + k_h + v_h) * h] --> [B, S, (q_h + k_h + v_h), h]
        qkv = qkv.reshape(
            batch_size,
            seq_length,
            self.num_q_heads + self.num_k_heads + self.num_v_heads,
            self.head_dim,
        )
        # [B, S, (q_h + k_h + v_h), h] --> [B, (q_h + k_h + v_h), S, h]
        qkv = qkv.swapaxes(1, 2)
        # [B, (q_h + k_h + v_h), S, h] --> [B, q_h, S h], [B, k_h, S, h], [B, v_h, S, h]
        queries, keys, values = qkv.split(
            [self.num_q_heads, self.num_k_heads, self.num_v_heads], axis=1
        )

        if self.q_norm is not None:
            queries = self.q_norm(queries)

        if self.k_norm is not None:
            keys = self.k_norm(keys)

        past_key_value = getattr(self, "past_key_value", past_key_value)

        if past_key_value is not None:
            # sin and cos are specific to RoPE models; position_ids needed for the static cache
            # cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
            cache_kwargs = {"cache_position": cache_position}
            keys, values = past_key_value.update(
                keys, values, self.layer_idx, cache_kwargs
            )

        # Add positional embedding
        queries, keys = self.pos_embedding(queries, keys)

        if self.num_groups != 1:
            # GQA
            # [B, k_h, S, h] --> [B, q_h, S, h]
            keys = keys.repeat_interleave(self.num_groups, dim=1)
            # [B, v_h, S, h] --> [B, q_h, S, h]
            values = values.repeat_interleave(self.num_groups, dim=1)

        causal_mask = attention_mask
        if attention_mask is not None and cache_position is not None:
            causal_mask = causal_mask[:, :, cache_position, : keys.shape[-2]]

        attn_output, _ = _scaled_dot_product_attention(
            queries,
            keys,
            values,
            attn_mask=causal_mask,
            dropout_p=0.,
            is_causal=False,
            is_training=self.training,
            dtype=queries.dtype
        )

        attn_output = attn_output.swapaxes(1, 2)
        attn_output = attn_output.reshape(
            batch_size, seq_length, self.num_q_heads * self.head_dim
        )
        attn_output = self.out_proj(attn_output)
        if not output_attentions:
            attn_weights = None
        return attn_output, attn_weights, past_key_value

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMMultiHeadCausalAttention.__init__(config, layer_idx)

Initializes an instance of the OpenELMMultiHeadCausalAttention class.

PARAMETER	DESCRIPTION
self	The instance of the class.
config	An instance of the OpenELMConfig class containing configuration parameters. TYPE: OpenELMConfig
layer_idx	The index of the layer. TYPE: int

RETURNS	DESCRIPTION
None	None

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
    '''
    Initializes an instance of the OpenELMMultiHeadCausalAttention class.

    Args:
        self: The instance of the class.
        config (OpenELMConfig): An instance of the OpenELMConfig class containing configuration parameters.
        layer_idx (int): The index of the layer.

    Returns:
        None

    Raises:
        None
    '''
    super().__init__()
    self.layer_idx = layer_idx
    head_dim = config.head_dim
    q_heads = config.num_query_heads[layer_idx]
    k_heads = config.num_kv_heads[layer_idx]
    v_heads = config.num_kv_heads[layer_idx]

    self.qkv_proj = nn.Linear(
        in_channels=config.model_dim,
        out_channels=(q_heads + k_heads + v_heads) * head_dim,
        bias=False,
    )

    self.pos_embedding = OpenELMRotaryEmbedding(
        model_dim=config.head_dim,
        max_seq_length=config.rope_max_length,
        freq_constant=config.rope_freq_constant,
    )

    if config.normalize_qk_projections:
        self.q_norm = OpenELMRMSNorm(
            num_features=config.head_dim,
        )
        self.k_norm = OpenELMRMSNorm(
            num_features=config.head_dim,
        )
    else:
        self.q_norm = None
        self.k_norm = None

    self.out_proj = nn.Linear(
        in_channels=q_heads * head_dim,
        out_channels=config.model_dim,
        bias=False,
    )

    self.head_dim = config.head_dim
    self.num_q_heads = q_heads
    self.num_k_heads = k_heads
    self.num_v_heads = v_heads
    self.transformer_dim = config.model_dim
    self.num_groups = self.num_q_heads // self.num_k_heads

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMMultiHeadCausalAttention.extra_repr()

Returns a string representation of the OpenELMMultiHeadCausalAttention object, including the number of query, key, and value heads.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMMultiHeadCausalAttention class. TYPE: OpenELMMultiHeadCausalAttention

RETURNS	DESCRIPTION
str	A string representation of the OpenELMMultiHeadCausalAttention object, including the number of query, key, and value heads. TYPE: str

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def extra_repr(self) -> str:
    """
    Returns a string representation of the OpenELMMultiHeadCausalAttention object, including the number of query,
    key, and value heads.

    Args:
        self (OpenELMMultiHeadCausalAttention): The instance of the OpenELMMultiHeadCausalAttention class.

    Returns:
        str: A string representation of the OpenELMMultiHeadCausalAttention object,
            including the number of query, key, and value heads.

    Raises:
        None.

    """
    return (
        super().extra_repr()
        + f"query_heads={self.num_q_heads}, key_heads={self.num_k_heads}, value_heads={self.num_v_heads}"
    )

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMMultiHeadCausalAttention.forward(hidden_states, attention_mask=None, past_key_value=None, output_attentions=False, use_cache=False, cache_position=None)

Forward pass of multi-head self-attention.

PARAMETER	DESCRIPTION
hidden_states	Input tensor of the shape [batch size, sequence length, model dimension]. TYPE: Tensor
past_key_value	Tensor storing the cached keys and values. TYPE: Optional[Cache] DEFAULT: None
output_attentions	output attention weights. TYPE: bool DEFAULT: False
use_cache	Specifies whether to use kv-cache for generation. TYPE: bool DEFAULT: False
cache_position	used for updating the kv-cache. TYPE: Optional[Tensor] DEFAULT: None

RETURNS	DESCRIPTION
Tuple[Tensor, Optional[Tensor], Optional[Tuple[Tensor]]]	The output of the same shape as the input, optionally with a tensor containing cached keys and values.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(
    self,
    hidden_states: mindspore.Tensor,
    attention_mask: Optional[mindspore.Tensor] = None,
    past_key_value: Optional[Cache] = None,
    output_attentions: bool = False,
    use_cache: bool = False,
    cache_position: Optional[mindspore.Tensor] = None,
) -> Tuple[mindspore.Tensor, Optional[mindspore.Tensor], Optional[Tuple[mindspore.Tensor]]]:
    """
    Forward pass of multi-head self-attention.

    Args:
        hidden_states: Input tensor of the shape [batch size, sequence length, model dimension].
        past_key_value: Tensor storing the cached keys and values.
        output_attentions: output attention weights.
        use_cache: Specifies whether to use kv-cache for generation.
        cache_position: used for updating the kv-cache.

    Returns:
        The output of the same shape as the input, optionally with a tensor containing cached keys and values.
    """
    # scaled_dot_product_attention does not return attention weights, set output_attentions to False
    output_attentions = False
    batch_size, seq_length, d_model = hidden_states.shape

    # [B, S, d] --> [B, S, (q_h + k_h + v_h) * h]
    qkv = self.qkv_proj(hidden_states)
    # [B, S, (q_h + k_h + v_h) * h] --> [B, S, (q_h + k_h + v_h), h]
    qkv = qkv.reshape(
        batch_size,
        seq_length,
        self.num_q_heads + self.num_k_heads + self.num_v_heads,
        self.head_dim,
    )
    # [B, S, (q_h + k_h + v_h), h] --> [B, (q_h + k_h + v_h), S, h]
    qkv = qkv.swapaxes(1, 2)
    # [B, (q_h + k_h + v_h), S, h] --> [B, q_h, S h], [B, k_h, S, h], [B, v_h, S, h]
    queries, keys, values = qkv.split(
        [self.num_q_heads, self.num_k_heads, self.num_v_heads], axis=1
    )

    if self.q_norm is not None:
        queries = self.q_norm(queries)

    if self.k_norm is not None:
        keys = self.k_norm(keys)

    past_key_value = getattr(self, "past_key_value", past_key_value)

    if past_key_value is not None:
        # sin and cos are specific to RoPE models; position_ids needed for the static cache
        # cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
        cache_kwargs = {"cache_position": cache_position}
        keys, values = past_key_value.update(
            keys, values, self.layer_idx, cache_kwargs
        )

    # Add positional embedding
    queries, keys = self.pos_embedding(queries, keys)

    if self.num_groups != 1:
        # GQA
        # [B, k_h, S, h] --> [B, q_h, S, h]
        keys = keys.repeat_interleave(self.num_groups, dim=1)
        # [B, v_h, S, h] --> [B, q_h, S, h]
        values = values.repeat_interleave(self.num_groups, dim=1)

    causal_mask = attention_mask
    if attention_mask is not None and cache_position is not None:
        causal_mask = causal_mask[:, :, cache_position, : keys.shape[-2]]

    attn_output, _ = _scaled_dot_product_attention(
        queries,
        keys,
        values,
        attn_mask=causal_mask,
        dropout_p=0.,
        is_causal=False,
        is_training=self.training,
        dtype=queries.dtype
    )

    attn_output = attn_output.swapaxes(1, 2)
    attn_output = attn_output.reshape(
        batch_size, seq_length, self.num_q_heads * self.head_dim
    )
    attn_output = self.out_proj(attn_output)
    if not output_attentions:
        attn_weights = None
    return attn_output, attn_weights, past_key_value

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMPreTrainedModel

Bases: PreTrainedModel

This class represents a pre-trained model for OpenELM. It is a subclass of PreTrainedModel and implements various methods and functionalities for training and inference.

The class contains an initialization method, '_init_weights', which is responsible for initializing the weights of the model. This method takes a 'cell' parameter, which represents the neural network cell.

The '_init_weights' method initializes the weights differently based on the type of the 'cell' parameter. If the 'cell' is an instance of 'nn.Linear', the weight is initialized using a normal distribution with a range defined by the 'initializer_range' attribute of the 'config' object. If the 'cell' has a bias, it is initialized to zeros.

If the 'cell' is an instance of 'nn.Embedding', the weight is initialized using a normal distribution with a range defined by the 'initializer_range' attribute of the 'config' object. If the 'cell' has a padding index, the weight corresponding to the padding index is set to zero.

If the 'cell' is an instance of 'OpenELMRMSNorm', the weight is initialized to ones.

Note

This class is designed specifically for OpenELM and inherits functionalities from the 'PreTrainedModel' class.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMPreTrainedModel(PreTrainedModel):

    """
    This class represents a pre-trained model for OpenELM.
    It is a subclass of PreTrainedModel and implements various methods and functionalities for training and inference.

    The class contains an initialization method, '_init_weights', which is responsible for initializing the weights
    of the model. This method takes a 'cell' parameter, which represents the neural network cell.

    The '_init_weights' method initializes the weights differently based on the type of the 'cell' parameter.
    If the 'cell' is an instance of 'nn.Linear', the weight is initialized using a normal distribution
    with a range defined by the 'initializer_range' attribute of the 'config' object.
    If the 'cell' has a bias, it is initialized to zeros.

    If the 'cell' is an instance of 'nn.Embedding', the weight is initialized using a normal distribution
    with a range defined by the 'initializer_range' attribute of the 'config' object.
    If the 'cell' has a padding index, the weight corresponding to the padding index is set to zero.

    If the 'cell' is an instance of 'OpenELMRMSNorm', the weight is initialized to ones.

    Note:
        This class is designed specifically for OpenELM and inherits functionalities from the 'PreTrainedModel' class.

    """
    config_class = OpenELMConfig
    base_model_prefix = "transformer"
    supports_gradient_checkpointing = False
    _no_split_modules = ["OpenELMDecoderLayer"]
    _skip_keys_device_placement = "past_key_values"

    def _init_weights(self, cell: nn.Module) -> None:
        """Initialize the weights."""
        if isinstance(cell, nn.Linear):
            # Slightly different from the TF version which uses truncated_normal for initialization
            # cf https://github.com/pytorch/pytorch/pull/5617
            cell.weight.initialize(Normal(self.config.initializer_range))
            if cell.bias is not None:
                cell.bias.initialize('zeros')
        elif isinstance(cell, nn.Embedding):
            weight = np.random.normal(0.0, self.config.initializer_range, cell.weight.shape)
            if cell.padding_idx:
                weight[cell.padding_idx] = 0

            cell.weight.set_data(Tensor(weight, cell.weight.dtype))

        elif isinstance(cell, OpenELMRMSNorm):
            cell.weight.initialize('ones')

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRMSNorm

Bases: Module

This class represents the OpenELMRMSNorm normalization layer, which can be used for normalizing input tensors.

ATTRIBUTE	DESCRIPTION
eps	A small value added to the denominator for numerical stability. TYPE: float
weight	Learnable scaling parameter. TYPE: Parameter

METHOD	DESCRIPTION
__init__	Initialize the OpenELMRMSNorm normalization layer.
_norm	Apply the OpenELMRMSNorm normalization to the input tensor.
forward	Forward pass through the OpenELMRMSNorm layer.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMRMSNorm(nn.Module):

    """
    This class represents the OpenELMRMSNorm normalization layer, which can be used for normalizing input tensors. 

    Attributes:
        eps (float): A small value added to the denominator for numerical stability.
        weight (nn.Parameter): Learnable scaling parameter.

    Methods:
        __init__:
            Initialize the OpenELMRMSNorm normalization layer.

        _norm:
            Apply the OpenELMRMSNorm normalization to the input tensor.

        forward:
            Forward pass through the OpenELMRMSNorm layer.

    """
    def __init__(self, num_features: int, eps: float = 1e-6):
        """
        Initialize the OpenELMRMSNorm normalization layer.

        Args:
            dim (int): The dimension of the input tensor.
            eps (float, optional): A small value added to the denominator for numerical stability. Default is 1e-6.

        Attributes:
            eps (float): A small value added to the denominator for numerical stability.
            weight (nn.Parameter): Learnable scaling parameter.

        """
        super().__init__()
        self.eps = eps
        self.weight = Parameter(ops.ones(num_features))
        self.num_features = num_features

    def _norm(self, x: Tensor) -> Tensor:
        """
        Apply the OpenELMRMSNorm normalization to the input tensor.

        Args:
            x (mindspore.Tensor): The input tensor.

        Returns:
            mindspore.Tensor: The normalized tensor.

        """
        return x * ops.rsqrt(x.pow(2).mean(-1, keep_dims=True) + self.eps)

    def forward(self, x: Tensor) -> Tensor:
        """
        Forward pass through the OpenELMRMSNorm layer.

        Args:
            x (mindspore.Tensor): The input tensor.

        Returns:
            mindspore.Tensor: The output tensor after applying OpenELMRMSNorm.

        """
        output = self._norm(x.float()).type_as(x)
        return output * self.weight

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRMSNorm.__init__(num_features, eps=1e-06)

Initialize the OpenELMRMSNorm normalization layer.

PARAMETER	DESCRIPTION
dim	The dimension of the input tensor. TYPE: int
eps	A small value added to the denominator for numerical stability. Default is 1e-6. TYPE: float DEFAULT: 1e-06

ATTRIBUTE	DESCRIPTION
eps	A small value added to the denominator for numerical stability. TYPE: float
weight	Learnable scaling parameter. TYPE: Parameter

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(self, num_features: int, eps: float = 1e-6):
    """
    Initialize the OpenELMRMSNorm normalization layer.

    Args:
        dim (int): The dimension of the input tensor.
        eps (float, optional): A small value added to the denominator for numerical stability. Default is 1e-6.

    Attributes:
        eps (float): A small value added to the denominator for numerical stability.
        weight (nn.Parameter): Learnable scaling parameter.

    """
    super().__init__()
    self.eps = eps
    self.weight = Parameter(ops.ones(num_features))
    self.num_features = num_features

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRMSNorm.forward(x)

Forward pass through the OpenELMRMSNorm layer.

PARAMETER	DESCRIPTION
x	The input tensor. TYPE: Tensor

RETURNS	DESCRIPTION
Tensor	mindspore.Tensor: The output tensor after applying OpenELMRMSNorm.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(self, x: Tensor) -> Tensor:
    """
    Forward pass through the OpenELMRMSNorm layer.

    Args:
        x (mindspore.Tensor): The input tensor.

    Returns:
        mindspore.Tensor: The output tensor after applying OpenELMRMSNorm.

    """
    output = self._norm(x.float()).type_as(x)
    return output * self.weight

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRotaryEmbedding

Bases: Module

The rotary position embeddings (aka RoPE) from RoFormer <https://arxiv.org/abs/2104.09864>_.

RoPE encodes the position information of tokens using a rotation matrix, and is able to capture explicit relative positional dependencies.

PARAMETER	DESCRIPTION
model_dim	The dimensionality of the model's hidden state. TYPE: int
max_seq_length	Maximum sequence length. TYPE: int
freq_constant	A constant used for computing frequencies. TYPE: int DEFAULT: 10000

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

class OpenELMRotaryEmbedding(nn.Module):
    """
    The rotary position embeddings (aka RoPE) from `RoFormer <https://arxiv.org/abs/2104.09864>`_.

    RoPE encodes the position information of tokens using a rotation matrix, and is able to capture
    explicit relative positional dependencies.

    Args:
        model_dim: The dimensionality of the model's hidden state.
        max_seq_length: Maximum sequence length.
        freq_constant: A constant used for computing frequencies.
    """
    def __init__(
        self, model_dim: int, max_seq_length: int, freq_constant: int = 10000
    ) -> None:
        """
        Initializes the OpenELMRotaryEmbedding instance with the specified parameters.

        Args:
            self: The object itself.
            model_dim (int): The dimension of the model.
            max_seq_length (int): The maximum sequence length.
            freq_constant (int, optional): The frequency constant used in the calculation. Defaults to 10000.

        Returns:
            None.

        Raises:
            None.
        """
        inv_freq = 1.0 / (
            freq_constant
            ** (ops.arange(0, model_dim, 2, dtype=mindspore.float32) / model_dim)
        )
        super().__init__()

        self.model_dim = model_dim
        self.freq_constant = freq_constant
        self.max_seq_length = max_seq_length

        self.inv_freq = inv_freq
        self._cached_cos = None
        self._cached_sin = None
        self._cached_seq_length = max_seq_length
        self._compute_sin_cos_embeddings(max_seq_length)

    def extra_repr(self) -> str:
        """
        This method generates a string representation that includes specific attributes of the OpenELMRotaryEmbedding
        class instance.

        Args:
            self: The instance of the OpenELMRotaryEmbedding class.

        Returns:
            str: A formatted string representing the model_dim, max_seq_length,
                and freq_constant attributes of the instance.

        Raises:
            None.
        """
        return f"\tmodel_dim={self.model_dim}, max_seq_length={self.max_seq_length}, freq_constant={self.freq_constant}"

    def _compute_sin_cos_embeddings(
        self,
        key_len: int,
        key_dtype = mindspore.float32,
    ) -> None:
        """
        Compute sine and cos embeddings.

        Args:
            key_len: Number of tokens in the key embeddings in the transformer model.
            key_dtype: Data type of the key embeddings.

        Returns:
            None

        Note:
            We recalculate the sine and cosine embeddings if any of the following conditions are met:

            1. The number of tokens in key embeddings are greater than the cached sequence length.
            2. Sine and cosine caches are empty.
        """
        if (
            key_len > self._cached_seq_length
            or self._cached_cos is None
            or (self._cached_cos is not None)
            or (self._cached_cos is not None and self._cached_cos.dtype != key_dtype)
            or self._cached_sin is None
            or (self._cached_sin is not None)
            or (self._cached_sin is not None and self._cached_sin.dtype != key_dtype)
        ):
            self._cached_seq_length = max(key_len, self._cached_seq_length)

            # The shape of 'pos_index' is [number of key tokens]
            pos_index = ops.arange(
                self._cached_seq_length,
                dtype=mindspore.float32,
            )
            # The shape of 'pos_index_theta' is [number of key tokens, model dimension]
            pos_index_theta = ops.einsum("i,j->ij", pos_index, self.inv_freq)
            # The shape of 'emb' is [number of key tokens, model dimension]
            emb = ops.cat((pos_index_theta, pos_index_theta), axis=-1)

            # the shape of cos and sin embeddings is [number of key tokens, model_dim]
            cos_emb = emb.cos().to(dtype=key_dtype)
            sin_emb = emb.sin().to(dtype=key_dtype)

            # the shape of cached cos and sin embeddings is [1, 1, number of key tokens, model_dim]
            self._cached_cos = cos_emb[None, None, :, :]
            self._cached_sin = sin_emb[None, None, :, :]

    def forward(
        self,
        query: mindspore.Tensor,
        key: mindspore.Tensor,
    ) -> Tuple[mindspore.Tensor, mindspore.Tensor]:
        """
        The forward function of RoPE embeddings.

        Args:
            query: Query embeddings in the transformer model. The shape of query embeddings is
                [Batch, number of query heads, number of query tokens, model dimension].
            key: Key embeddings in the transformer model. The shape of key embeddings is
                [Batch, number of key heads, number of key tokens, model dimension].

        Returns:
            tuple:
                A tuple containing the query and key embeddings with positional information.
                The shape of the returned query and key embeddings is the same as the input query and key embeddings
                respectively.

        Note:
            The RoPE embedding computation is done in full-precision. After the computation, input query and key tensors
            are casted to original input datatype.
        """
        dim = key.shape[-1]
        key_len = key.shape[2]
        query_len = query.shape[2]

        assert dim == self.model_dim
        assert key.dtype == query.dtype

        # In the context of self-attention, the lengths of keys and queries are equal.
        # However, in generation tasks, such as predicting the next token in a sequence, the lengths of keys and queries
        # can differ. For instance, when employing key-value (KV) caching for sequence prediction, the keys
        # represent embeddings of previous tokens and the current token, while the query corresponds
        # to the embedding of the current token only.
        assert (
            key_len >= query_len
        ), "Number of keys has to be greater than or equal to number of queries."

        query_float = query.float()
        key_float = key.float()

        self._compute_sin_cos_embeddings(
            key_len, key_dtype=key_float.dtype
        )
        query_float = _apply_rotary_pos_emb(
            x=query_float,
            pos_sin=self._cached_sin[..., key_len - query_len : key_len, :],
            pos_cos=self._cached_cos[..., key_len - query_len : key_len, :],
        )
        key_float = _apply_rotary_pos_emb(
            x=key_float,
            pos_sin=self._cached_sin[..., :key_len, :],
            pos_cos=self._cached_cos[..., :key_len, :],
        )

        return query_float.type_as(query), key_float.type_as(key)

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRotaryEmbedding.__init__(model_dim, max_seq_length, freq_constant=10000)

Initializes the OpenELMRotaryEmbedding instance with the specified parameters.

PARAMETER	DESCRIPTION
self	The object itself.
model_dim	The dimension of the model. TYPE: int
max_seq_length	The maximum sequence length. TYPE: int
freq_constant	The frequency constant used in the calculation. Defaults to 10000. TYPE: int DEFAULT: 10000

RETURNS	DESCRIPTION
None	None.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def __init__(
    self, model_dim: int, max_seq_length: int, freq_constant: int = 10000
) -> None:
    """
    Initializes the OpenELMRotaryEmbedding instance with the specified parameters.

    Args:
        self: The object itself.
        model_dim (int): The dimension of the model.
        max_seq_length (int): The maximum sequence length.
        freq_constant (int, optional): The frequency constant used in the calculation. Defaults to 10000.

    Returns:
        None.

    Raises:
        None.
    """
    inv_freq = 1.0 / (
        freq_constant
        ** (ops.arange(0, model_dim, 2, dtype=mindspore.float32) / model_dim)
    )
    super().__init__()

    self.model_dim = model_dim
    self.freq_constant = freq_constant
    self.max_seq_length = max_seq_length

    self.inv_freq = inv_freq
    self._cached_cos = None
    self._cached_sin = None
    self._cached_seq_length = max_seq_length
    self._compute_sin_cos_embeddings(max_seq_length)

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRotaryEmbedding.extra_repr()

This method generates a string representation that includes specific attributes of the OpenELMRotaryEmbedding class instance.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMRotaryEmbedding class.

RETURNS	DESCRIPTION
str	A formatted string representing the model_dim, max_seq_length, and freq_constant attributes of the instance. TYPE: str

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def extra_repr(self) -> str:
    """
    This method generates a string representation that includes specific attributes of the OpenELMRotaryEmbedding
    class instance.

    Args:
        self: The instance of the OpenELMRotaryEmbedding class.

    Returns:
        str: A formatted string representing the model_dim, max_seq_length,
            and freq_constant attributes of the instance.

    Raises:
        None.
    """
    return f"\tmodel_dim={self.model_dim}, max_seq_length={self.max_seq_length}, freq_constant={self.freq_constant}"

mindnlp.transformers.models.openelm.modeling_openelm.OpenELMRotaryEmbedding.forward(query, key)

The forward function of RoPE embeddings.

PARAMETER	DESCRIPTION
query	Query embeddings in the transformer model. The shape of query embeddings is [Batch, number of query heads, number of query tokens, model dimension]. TYPE: Tensor
key	Key embeddings in the transformer model. The shape of key embeddings is [Batch, number of key heads, number of key tokens, model dimension]. TYPE: Tensor

RETURNS	DESCRIPTION
tuple	A tuple containing the query and key embeddings with positional information. The shape of the returned query and key embeddings is the same as the input query and key embeddings respectively. TYPE: Tuple[Tensor, Tensor]

Note

The RoPE embedding computation is done in full-precision. After the computation, input query and key tensors are casted to original input datatype.

Source code in mindnlp/transformers/models/openelm/modeling_openelm.py

def forward(
    self,
    query: mindspore.Tensor,
    key: mindspore.Tensor,
) -> Tuple[mindspore.Tensor, mindspore.Tensor]:
    """
    The forward function of RoPE embeddings.

    Args:
        query: Query embeddings in the transformer model. The shape of query embeddings is
            [Batch, number of query heads, number of query tokens, model dimension].
        key: Key embeddings in the transformer model. The shape of key embeddings is
            [Batch, number of key heads, number of key tokens, model dimension].

    Returns:
        tuple:
            A tuple containing the query and key embeddings with positional information.
            The shape of the returned query and key embeddings is the same as the input query and key embeddings
            respectively.

    Note:
        The RoPE embedding computation is done in full-precision. After the computation, input query and key tensors
        are casted to original input datatype.
    """
    dim = key.shape[-1]
    key_len = key.shape[2]
    query_len = query.shape[2]

    assert dim == self.model_dim
    assert key.dtype == query.dtype

    # In the context of self-attention, the lengths of keys and queries are equal.
    # However, in generation tasks, such as predicting the next token in a sequence, the lengths of keys and queries
    # can differ. For instance, when employing key-value (KV) caching for sequence prediction, the keys
    # represent embeddings of previous tokens and the current token, while the query corresponds
    # to the embedding of the current token only.
    assert (
        key_len >= query_len
    ), "Number of keys has to be greater than or equal to number of queries."

    query_float = query.float()
    key_float = key.float()

    self._compute_sin_cos_embeddings(
        key_len, key_dtype=key_float.dtype
    )
    query_float = _apply_rotary_pos_emb(
        x=query_float,
        pos_sin=self._cached_sin[..., key_len - query_len : key_len, :],
        pos_cos=self._cached_cos[..., key_len - query_len : key_len, :],
    )
    key_float = _apply_rotary_pos_emb(
        x=key_float,
        pos_sin=self._cached_sin[..., :key_len, :],
        pos_cos=self._cached_cos[..., :key_len, :],
    )

    return query_float.type_as(query), key_float.type_as(key)

mindnlp.transformers.models.openelm.configuration_openelm

Implements HF OpenELMConfig based on PretrainedConfig

mindnlp.transformers.models.openelm.configuration_openelm.OpenELMConfig

Bases: PretrainedConfig

This is the configuration class to store the configuration of a [OpenELMModel]. It is used to instantiate an OpenELM model according to the specified arguments, defining the model architecture.

Configuration objects inherit from [PretrainedConfig] and can be used to control the model outputs. Read the documentation from [PretrainedConfig] for more information.

PARAMETER	DESCRIPTION
vocab_size	Vocabulary size of the OpenELM model. TYPE: `int`, *optional*, defaults to 32000 DEFAULT: 32000
max_context_length	Maximum number of input tokens. TYPE: `int`, *optional*, defaults to 2048 DEFAULT: 2048
num_transformer_layers	Number of hidden layers in the Transformer decoder. TYPE: `int`, *optional*, defaults to 12 DEFAULT: 12
model_dim	Dimension of the hidden representations. TYPE: `int`, *optional*, defaults to 2048 DEFAULT: 2048
head_dim	The attention head dimension. TYPE: `int`, *optional*, defaults to 128 DEFAULT: 128
qkv_multipliers	If the qkv_multipliers is a Number, then all attention layers have the same latent dimensions, resulting in uniform allocation of parameters. If the qkv_multipliers is a List of Number, then each attention layer have different latent dimensions assuming qkv_multipliers[0] != qkv_multipliers[1]. This results in variable allocation of parameters in attention layer. This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623 TYPE: `Union[Number, List[Number]]`, *optional*, defaults to 1.0 DEFAULT: 1.0
num_query_heads	The number of query heads, computed from compute_heads(model_dim=model_dim, head_dim=head_dim). TYPE: `Union[int, None]`, *optional*, defaults to None DEFAULT: None
num_gqa_groups	This variable allows to switch between multi-head attention, group query attention, and multi-query attention. When num_gqa_groups == 1, then it is multi-head attention. When 1 < num_gqa_groups < num_heads and num_heads is divisible by num_gqa_groups, then it is group query attention When num_gqa_groups == num_heads, then it is multi-query attention TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1
ffn_multipliers	Feed-forward network (FFN) multipliers. If the ffn_multipliers is a Number, then all FFN layers have the same latent dimensions, resulting in uniform allocation of parameters. If the ffn_multipliers is a List of Number, then each FFN layer have different latent dimensions assuming ffn_multipliers[0] != ffn_multipliers[1]. This results in variable allocation of parameters in FFN layer. This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623 TYPE: `Union[Number, List[Number]]`, *optional*, defaults to 4.0 DEFAULT: 4.0
ffn_with_glu	Whether to use FFN with Gated Linear Unit (GLU) TYPE: `bool`, *optional*, defaults to True DEFAULT: True
ffn_dim_divisor	The ffn layer dimension divisor. TYPE: `int`, *optional*, defaults to 256 DEFAULT: 256
activation_fn_name	The non-linear activation function (function or string) in the decoder. TYPE: `str` or `function`, *optional*, defaults to `"swish"` DEFAULT: 'swish'
normalization_layer_name	Type of normalization layer. TYPE: `str` or `function`, *optional*, defaults to `"rms_norm"` DEFAULT: 'rms_norm'
normalize_qk_projections	Whether to normalize queries and keys after projections TYPE: `bool`, *optional*, defaults to False DEFAULT: False
share_input_output_layers	Whether to share the embedding between input and output linear layer TYPE: `bool`, *optional*, defaults to False DEFAULT: False
rope_freq_constant	The base period of the RoPE embeddings. TYPE: `int`, *optional*, defaults to 10000 DEFAULT: 10000
rope_max_length	That rope_max_length is set to twice of max_context_length. This allows flexibility in token lengths during training or fine-tuning. TYPE: `int`, *optional*, defaults to 4096 DEFAULT: 4096
initializer_range	The standard deviation of the truncated_normal_initializer for initializing all weight matrices. TYPE: `float`, *optional*, defaults to 0.02 DEFAULT: 0.02
use_cache	Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if config.is_decoder=True. TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True
bos_token_id	Beginning of stream token id. TYPE: `int`, *optional*, defaults to 2 DEFAULT: 1
eos_token_id	End of stream token id. TYPE: `int`, *optional*, defaults to 1 DEFAULT: 2

Source code in mindnlp/transformers/models/openelm/configuration_openelm.py

class OpenELMConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`OpenELMModel`].
    It is used to instantiate an OpenELM model according to the specified arguments, defining the model architecture.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    Args:
        vocab_size (`int`, *optional*, defaults to 32000):
            Vocabulary size of the OpenELM model.
        max_context_length (`int`, *optional*, defaults to 2048):
            Maximum number of input tokens.
        num_transformer_layers (`int`, *optional*, defaults to 12):
            Number of hidden layers in the Transformer decoder.
        model_dim (`int`, *optional*, defaults to 2048):
            Dimension of the hidden representations.
        head_dim (`int`, *optional*, defaults to 128):
            The attention head dimension.
        qkv_multipliers (`Union[Number, List[Number]]`, *optional*, defaults to 1.0):
            If the qkv_multipliers is a Number, then all attention layers have the same latent dimensions,
            resulting in uniform allocation of parameters.
            If the qkv_multipliers is a List of Number, then each attention layer have different latent dimensions
            assuming qkv_multipliers[0] != qkv_multipliers[1]. This results in variable allocation of parameters
            in attention layer.
            This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
        num_query_heads (`Union[int, None]`, *optional*, defaults to None):
            The number of query heads, computed from `compute_heads(model_dim=model_dim, head_dim=head_dim)`.
        num_gqa_groups (`int`, *optional*, defaults to 1):
            This variable allows to switch between multi-head attention, group query attention, and multi-query attention.
            When num_gqa_groups == 1, then it is multi-head attention.
            When 1 < num_gqa_groups < num_heads and num_heads is divisible by num_gqa_groups,
            then it is group query attention
            When num_gqa_groups == num_heads, then it is multi-query attention
        ffn_multipliers (`Union[Number, List[Number]]`, *optional*, defaults to 4.0):
            Feed-forward network (FFN) multipliers.
            If the ffn_multipliers is a Number, then all FFN layers have the same latent dimensions,
            resulting in uniform allocation of parameters.
            If the ffn_multipliers is a List of Number, then each FFN layer have different latent dimensions
            assuming ffn_multipliers[0] != ffn_multipliers[1]. This results in variable allocation of parameters
            in FFN layer.
            This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
        ffn_with_glu (`bool`, *optional*, defaults to True):
            Whether to use FFN with Gated Linear Unit (GLU)
        ffn_dim_divisor (`int`, *optional*, defaults to 256):
            The ffn layer dimension divisor.
        activation_fn_name (`str` or `function`, *optional*, defaults to `"swish"`):
            The non-linear activation function (function or string) in the decoder.
        normalization_layer_name (`str` or `function`, *optional*, defaults to `"rms_norm"`):
            Type of normalization layer.
        normalize_qk_projections (`bool`, *optional*, defaults to False):
            Whether to normalize queries and keys after projections
        share_input_output_layers (`bool`, *optional*, defaults to False):
            Whether to share the embedding between input and output linear layer
        rope_freq_constant (`int`, *optional*, defaults to 10000):
            The base period of the RoPE embeddings.
        rope_max_length (`int`, *optional*, defaults to 4096):
            That rope_max_length is set to twice of max_context_length.
            This allows flexibility in token lengths during training or fine-tuning.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if `config.is_decoder=True`.
        bos_token_id (`int`, *optional*, defaults to 2):
            Beginning of stream token id.
        eos_token_id (`int`, *optional*, defaults to 1):
            End of stream token id.
    """
    model_type = "openelm"

    def __init__(
        self,
        vocab_size: int = 32000,
        max_context_length: int = 2048,
        num_transformer_layers: int = 12,
        model_dim: int = 2048,
        head_dim: int = 128,
        qkv_multipliers: Union[Number, List[Number]] = 1.0,
        num_query_heads: Union[int, None] = None,
        num_gqa_groups: int = 1,
        ffn_multipliers: Union[Number, List[Number]] = 4.0,
        ffn_with_glu: bool = True,
        ffn_dim_divisor: int = 256,
        activation_fn_name: str = "swish",
        normalization_layer_name: str = "rms_norm",
        normalize_qk_projections: bool = False,
        share_input_output_layers: bool = False,
        rope_freq_constant: int = 10000,
        rope_max_length: int = 4096,
        initializer_range: float = 0.02,
        use_cache: bool = True,
        bos_token_id: int = 1,
        eos_token_id: int = 2,
        **kwargs,
    ) -> None:
        """
        This method initializes an instance of the OpenELMConfig class with the provided parameters.

        Args:
            self: The instance of the class.
            vocab_size (int): The size of the vocabulary.
            max_context_length (int): The maximum length of the context.
            num_transformer_layers (int): The number of transformer layers.
            model_dim (int): The dimension of the model.
            head_dim (int): The dimension of the head.
            qkv_multipliers (Union[Number, List[Number]]): The multiplier(s) for query, key, and value vectors.
            num_query_heads (Union[int, None]): The number of query heads.
                If None, it will be computed based on model_dim and head_dim.
            num_gqa_groups (int): The number of groups for generalized query attention.
            ffn_multipliers (Union[Number, List[Number]]): The multiplier(s) for feed-forward network.
            ffn_with_glu (bool): A boolean indicating whether to use gated linear units in the feed-forward network.
            ffn_dim_divisor (int): The divisor for the feed-forward network dimension.
            activation_fn_name (str): The name of the activation function.
            normalization_layer_name (str): The name of the normalization layer.
            normalize_qk_projections (bool): A boolean indicating whether to normalize query and key projections.
            share_input_output_layers (bool): A boolean indicating whether to share input and output layers.
            rope_freq_constant (int): The frequency constant for the relative positional encoding.
            rope_max_length (int): The maximum length for the relative positional encoding.
            initializer_range (float): The range for random weight initialization.
            use_cache (bool): A boolean indicating whether to use cache.
            bos_token_id (int): The token ID for the beginning of sentence.
            eos_token_id (int): The token ID for the end of sentence.

        Returns:
            None.

        Raises:
            None.
        """
        self.vocab_size = vocab_size
        self.max_context_length = max_context_length
        self.num_transformer_layers = num_transformer_layers
        self.model_dim = model_dim
        self.head_dim = head_dim
        self.qkv_multipliers = qkv_multipliers
        self.num_query_heads = num_query_heads
        self.num_gqa_groups = num_gqa_groups
        self.ffn_multipliers = ffn_multipliers
        self.ffn_with_glu = ffn_with_glu
        self.ffn_dim_divisor = ffn_dim_divisor
        self.activation_fn_name = activation_fn_name
        self.normalization_layer_name = normalization_layer_name
        self.normalize_qk_projections = normalize_qk_projections
        self.share_input_output_layers = share_input_output_layers
        self.rope_freq_constant = rope_freq_constant
        self.rope_max_length = rope_max_length
        self.num_query_heads = (
            compute_heads(model_dim=model_dim, head_dim=head_dim)
            if num_query_heads is None
            else num_query_heads
        )
        self.initializer_range = initializer_range

        self.__post_init__()
        super().__init__(
            use_cache=use_cache,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            **kwargs,
        )

    def __post_init__(self) -> None:
        """
        This method initializes the configuration parameters for the OpenELM model.

        Args:
            self (OpenELMConfig): The instance of the OpenELMConfig class.

        Returns:
            None.

        Raises:
            NotImplementedError: If the QKV multipliers are not a single number or a list containing exactly two numbers,
                or if the FFN multipliers are not a single number or a list containing exactly two numbers.
            AssertionError: If the length of the FFN multipliers does not match the number of transformer layers, or if
                the number of query heads is not divisible by the number of key-value heads for any layer.
        """
        if self.num_gqa_groups is not None:
            head_multiple_of = self.num_gqa_groups
        else:
            head_multiple_of = 2

        if isinstance(self.qkv_multipliers, Number):
            # All attention layers have the same latent dimensions, resulting in uniform allocation of parameters.
            qkv_dim = make_divisible(
                self.model_dim * self.qkv_multipliers,
                divisor=self.head_dim * head_multiple_of,
            )
            query_dims = [int(qkv_dim)] * self.num_transformer_layers

        elif (
            isinstance(self.qkv_multipliers, (tuple, list))
            and len(self.qkv_multipliers) == 2
        ):
            # Each attention layer have different latent dimensions assuming qkv_multipliers[0] != qkv_multipliers[1].
            # This results in variable allocation of parameters in attention layer.
            # This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
            qkv_multipliers = [
                round(v, 2)
                for v in np.linspace(
                    self.qkv_multipliers[0],
                    self.qkv_multipliers[1],
                    num=self.num_transformer_layers,
                    dtype=float,
                )
            ]
            # Make sure that scaled model dimension is divisible by scaled head dimension.
            query_dims = [
                int(
                    make_divisible(
                        self.model_dim * m, divisor=self.head_dim * head_multiple_of
                    )
                )
                for m in qkv_multipliers
            ]
        else:
            raise NotImplementedError(
                f"QKV multipliers should be a single number or a list containing exactly two numbers. Got: {qkv_multipliers}."
            )

        # compute the number of query, key, and value heads
        # For multi-head and multi-query attention, the number of heads for query, key, and value are the same.
        # For group query attention, the number of key and value heads are the same.
        self.num_query_heads = [
            int(compute_heads(q_dim, self.head_dim)) for q_dim in query_dims
        ]
        self.num_kv_heads = [
            q_heads // self.num_gqa_groups for q_heads in self.num_query_heads
        ]

        # Feed-forward network (FFN) multipliers
        if isinstance(self.ffn_multipliers, Number):
            # All FFN layers have the same latent dimensions, resulting in uniform allocation of parameters.
            self.ffn_multipliers = [self.ffn_multipliers] * self.num_transformer_layers
        elif isinstance(self.ffn_multipliers, (tuple, list)):
            # Each FFN layer have different latent dimensions assuming ffn_multipliers[0] != ffn_multipliers[1].
            # This results in variable allocation of parameters in FFN layer.
            # This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
            if len(self.ffn_multipliers) == 2:
                self.ffn_multipliers = [
                    round(v, 2)
                    for v in np.linspace(
                        self.ffn_multipliers[0],
                        self.ffn_multipliers[1],
                        num=self.num_transformer_layers,
                        dtype=float,
                    )
                ]
            else:
                assert (
                    len(self.ffn_multipliers) == self.num_transformer_layers
                ), f"{len(self.ffn_multipliers)}!={self.num_transformer_layers}"
        else:
            raise NotImplementedError(
                f"FFN multipliers should be a single number or a list containing exactly two numbers. Got: {qkv_multipliers}."
            )

        # check num_query_heads divisible by num_kv_heads for every layer
        for layer_idx in range(len(query_dims)):
            assert self.num_query_heads[layer_idx] % self.num_kv_heads[layer_idx] == 0

mindnlp.transformers.models.openelm.configuration_openelm.OpenELMConfig.__init__(vocab_size=32000, max_context_length=2048, num_transformer_layers=12, model_dim=2048, head_dim=128, qkv_multipliers=1.0, num_query_heads=None, num_gqa_groups=1, ffn_multipliers=4.0, ffn_with_glu=True, ffn_dim_divisor=256, activation_fn_name='swish', normalization_layer_name='rms_norm', normalize_qk_projections=False, share_input_output_layers=False, rope_freq_constant=10000, rope_max_length=4096, initializer_range=0.02, use_cache=True, bos_token_id=1, eos_token_id=2, **kwargs)

This method initializes an instance of the OpenELMConfig class with the provided parameters.

PARAMETER	DESCRIPTION
self	The instance of the class.
vocab_size	The size of the vocabulary. TYPE: int DEFAULT: 32000
max_context_length	The maximum length of the context. TYPE: int DEFAULT: 2048
num_transformer_layers	The number of transformer layers. TYPE: int DEFAULT: 12
model_dim	The dimension of the model. TYPE: int DEFAULT: 2048
head_dim	The dimension of the head. TYPE: int DEFAULT: 128
qkv_multipliers	The multiplier(s) for query, key, and value vectors. TYPE: Union[Number, List[Number]] DEFAULT: 1.0
num_query_heads	The number of query heads. If None, it will be computed based on model_dim and head_dim. TYPE: Union[int, None] DEFAULT: None
num_gqa_groups	The number of groups for generalized query attention. TYPE: int DEFAULT: 1
ffn_multipliers	The multiplier(s) for feed-forward network. TYPE: Union[Number, List[Number]] DEFAULT: 4.0
ffn_with_glu	A boolean indicating whether to use gated linear units in the feed-forward network. TYPE: bool DEFAULT: True
ffn_dim_divisor	The divisor for the feed-forward network dimension. TYPE: int DEFAULT: 256
activation_fn_name	The name of the activation function. TYPE: str DEFAULT: 'swish'
normalization_layer_name	The name of the normalization layer. TYPE: str DEFAULT: 'rms_norm'
normalize_qk_projections	A boolean indicating whether to normalize query and key projections. TYPE: bool DEFAULT: False
share_input_output_layers	A boolean indicating whether to share input and output layers. TYPE: bool DEFAULT: False
rope_freq_constant	The frequency constant for the relative positional encoding. TYPE: int DEFAULT: 10000
rope_max_length	The maximum length for the relative positional encoding. TYPE: int DEFAULT: 4096
initializer_range	The range for random weight initialization. TYPE: float DEFAULT: 0.02
use_cache	A boolean indicating whether to use cache. TYPE: bool DEFAULT: True
bos_token_id	The token ID for the beginning of sentence. TYPE: int DEFAULT: 1
eos_token_id	The token ID for the end of sentence. TYPE: int DEFAULT: 2

RETURNS	DESCRIPTION
None	None.

Source code in mindnlp/transformers/models/openelm/configuration_openelm.py

def __init__(
    self,
    vocab_size: int = 32000,
    max_context_length: int = 2048,
    num_transformer_layers: int = 12,
    model_dim: int = 2048,
    head_dim: int = 128,
    qkv_multipliers: Union[Number, List[Number]] = 1.0,
    num_query_heads: Union[int, None] = None,
    num_gqa_groups: int = 1,
    ffn_multipliers: Union[Number, List[Number]] = 4.0,
    ffn_with_glu: bool = True,
    ffn_dim_divisor: int = 256,
    activation_fn_name: str = "swish",
    normalization_layer_name: str = "rms_norm",
    normalize_qk_projections: bool = False,
    share_input_output_layers: bool = False,
    rope_freq_constant: int = 10000,
    rope_max_length: int = 4096,
    initializer_range: float = 0.02,
    use_cache: bool = True,
    bos_token_id: int = 1,
    eos_token_id: int = 2,
    **kwargs,
) -> None:
    """
    This method initializes an instance of the OpenELMConfig class with the provided parameters.

    Args:
        self: The instance of the class.
        vocab_size (int): The size of the vocabulary.
        max_context_length (int): The maximum length of the context.
        num_transformer_layers (int): The number of transformer layers.
        model_dim (int): The dimension of the model.
        head_dim (int): The dimension of the head.
        qkv_multipliers (Union[Number, List[Number]]): The multiplier(s) for query, key, and value vectors.
        num_query_heads (Union[int, None]): The number of query heads.
            If None, it will be computed based on model_dim and head_dim.
        num_gqa_groups (int): The number of groups for generalized query attention.
        ffn_multipliers (Union[Number, List[Number]]): The multiplier(s) for feed-forward network.
        ffn_with_glu (bool): A boolean indicating whether to use gated linear units in the feed-forward network.
        ffn_dim_divisor (int): The divisor for the feed-forward network dimension.
        activation_fn_name (str): The name of the activation function.
        normalization_layer_name (str): The name of the normalization layer.
        normalize_qk_projections (bool): A boolean indicating whether to normalize query and key projections.
        share_input_output_layers (bool): A boolean indicating whether to share input and output layers.
        rope_freq_constant (int): The frequency constant for the relative positional encoding.
        rope_max_length (int): The maximum length for the relative positional encoding.
        initializer_range (float): The range for random weight initialization.
        use_cache (bool): A boolean indicating whether to use cache.
        bos_token_id (int): The token ID for the beginning of sentence.
        eos_token_id (int): The token ID for the end of sentence.

    Returns:
        None.

    Raises:
        None.
    """
    self.vocab_size = vocab_size
    self.max_context_length = max_context_length
    self.num_transformer_layers = num_transformer_layers
    self.model_dim = model_dim
    self.head_dim = head_dim
    self.qkv_multipliers = qkv_multipliers
    self.num_query_heads = num_query_heads
    self.num_gqa_groups = num_gqa_groups
    self.ffn_multipliers = ffn_multipliers
    self.ffn_with_glu = ffn_with_glu
    self.ffn_dim_divisor = ffn_dim_divisor
    self.activation_fn_name = activation_fn_name
    self.normalization_layer_name = normalization_layer_name
    self.normalize_qk_projections = normalize_qk_projections
    self.share_input_output_layers = share_input_output_layers
    self.rope_freq_constant = rope_freq_constant
    self.rope_max_length = rope_max_length
    self.num_query_heads = (
        compute_heads(model_dim=model_dim, head_dim=head_dim)
        if num_query_heads is None
        else num_query_heads
    )
    self.initializer_range = initializer_range

    self.__post_init__()
    super().__init__(
        use_cache=use_cache,
        bos_token_id=bos_token_id,
        eos_token_id=eos_token_id,
        **kwargs,
    )

mindnlp.transformers.models.openelm.configuration_openelm.OpenELMConfig.__post_init__()

This method initializes the configuration parameters for the OpenELM model.

PARAMETER	DESCRIPTION
self	The instance of the OpenELMConfig class. TYPE: OpenELMConfig

RETURNS	DESCRIPTION
None	None.

RAISES	DESCRIPTION
NotImplementedError	If the QKV multipliers are not a single number or a list containing exactly two numbers, or if the FFN multipliers are not a single number or a list containing exactly two numbers.
AssertionError	If the length of the FFN multipliers does not match the number of transformer layers, or if the number of query heads is not divisible by the number of key-value heads for any layer.

Source code in mindnlp/transformers/models/openelm/configuration_openelm.py

def __post_init__(self) -> None:
    """
    This method initializes the configuration parameters for the OpenELM model.

    Args:
        self (OpenELMConfig): The instance of the OpenELMConfig class.

    Returns:
        None.

    Raises:
        NotImplementedError: If the QKV multipliers are not a single number or a list containing exactly two numbers,
            or if the FFN multipliers are not a single number or a list containing exactly two numbers.
        AssertionError: If the length of the FFN multipliers does not match the number of transformer layers, or if
            the number of query heads is not divisible by the number of key-value heads for any layer.
    """
    if self.num_gqa_groups is not None:
        head_multiple_of = self.num_gqa_groups
    else:
        head_multiple_of = 2

    if isinstance(self.qkv_multipliers, Number):
        # All attention layers have the same latent dimensions, resulting in uniform allocation of parameters.
        qkv_dim = make_divisible(
            self.model_dim * self.qkv_multipliers,
            divisor=self.head_dim * head_multiple_of,
        )
        query_dims = [int(qkv_dim)] * self.num_transformer_layers

    elif (
        isinstance(self.qkv_multipliers, (tuple, list))
        and len(self.qkv_multipliers) == 2
    ):
        # Each attention layer have different latent dimensions assuming qkv_multipliers[0] != qkv_multipliers[1].
        # This results in variable allocation of parameters in attention layer.
        # This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
        qkv_multipliers = [
            round(v, 2)
            for v in np.linspace(
                self.qkv_multipliers[0],
                self.qkv_multipliers[1],
                num=self.num_transformer_layers,
                dtype=float,
            )
        ]
        # Make sure that scaled model dimension is divisible by scaled head dimension.
        query_dims = [
            int(
                make_divisible(
                    self.model_dim * m, divisor=self.head_dim * head_multiple_of
                )
            )
            for m in qkv_multipliers
        ]
    else:
        raise NotImplementedError(
            f"QKV multipliers should be a single number or a list containing exactly two numbers. Got: {qkv_multipliers}."
        )

    # compute the number of query, key, and value heads
    # For multi-head and multi-query attention, the number of heads for query, key, and value are the same.
    # For group query attention, the number of key and value heads are the same.
    self.num_query_heads = [
        int(compute_heads(q_dim, self.head_dim)) for q_dim in query_dims
    ]
    self.num_kv_heads = [
        q_heads // self.num_gqa_groups for q_heads in self.num_query_heads
    ]

    # Feed-forward network (FFN) multipliers
    if isinstance(self.ffn_multipliers, Number):
        # All FFN layers have the same latent dimensions, resulting in uniform allocation of parameters.
        self.ffn_multipliers = [self.ffn_multipliers] * self.num_transformer_layers
    elif isinstance(self.ffn_multipliers, (tuple, list)):
        # Each FFN layer have different latent dimensions assuming ffn_multipliers[0] != ffn_multipliers[1].
        # This results in variable allocation of parameters in FFN layer.
        # This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
        if len(self.ffn_multipliers) == 2:
            self.ffn_multipliers = [
                round(v, 2)
                for v in np.linspace(
                    self.ffn_multipliers[0],
                    self.ffn_multipliers[1],
                    num=self.num_transformer_layers,
                    dtype=float,
                )
            ]
        else:
            assert (
                len(self.ffn_multipliers) == self.num_transformer_layers
            ), f"{len(self.ffn_multipliers)}!={self.num_transformer_layers}"
    else:
        raise NotImplementedError(
            f"FFN multipliers should be a single number or a list containing exactly two numbers. Got: {qkv_multipliers}."
        )

    # check num_query_heads divisible by num_kv_heads for every layer
    for layer_idx in range(len(query_dims)):
        assert self.num_query_heads[layer_idx] % self.num_kv_heads[layer_idx] == 0

mindnlp.transformers.models.openelm.configuration_openelm.compute_heads(model_dim, head_dim)

Compute the number of heads.

PARAMETER	DESCRIPTION
model_dim	Model dimension. TYPE: int
head_dim	Head dimension. TYPE: int

RETURNS	DESCRIPTION
int	An integer denoting number of heads in multi-head attention is returned.

RAISES	DESCRIPTION
ValueError	if model dimension is not divisible by head dimension.

Source code in mindnlp/transformers/models/openelm/configuration_openelm.py

def compute_heads(model_dim: int, head_dim: int) -> int:
    """
    Compute the number of heads.

    Args:
        model_dim: Model dimension.
        head_dim: Head dimension.

    Returns:
        An integer denoting number of heads in multi-head attention is returned.

    Raises:
        ValueError: if model dimension is not divisible by head dimension.
    """
    if model_dim % head_dim == 0:
        return model_dim // head_dim
    else:
        raise ValueError(
            f"Model dimension should be divisible by head dimension. Got: {model_dim} and {head_dim}."
        )

mindnlp.transformers.models.openelm.configuration_openelm.make_divisible(v, divisor=8, min_value=None)

This function is taken from the original tf repo. It ensures that all layers have a channel number that is divisible by the divisor It can be seen at: https://github.com/tensorflow/models/blob/2cfc99eff5e5eb729c6793d2f3d03aa1c9be2b15/research/slim/nets/mobilenet/mobilenet.py#L62

PARAMETER	DESCRIPTION
v	input value TYPE: Union[float, int]
divisor	default to 8 TYPE: Optional[int] DEFAULT: 8
min_value	minimum divisor value TYPE: Optional[Union[float, int]] DEFAULT: None

RETURNS	DESCRIPTION
new_v	new divisible value TYPE: Union[float, int]

Source code in mindnlp/transformers/models/openelm/configuration_openelm.py

def make_divisible(
    v: Union[float, int],
    divisor: Optional[int] = 8,
    min_value: Optional[Union[float, int]] = None,
) -> Union[float, int]:
    """
    This function is taken from the original tf repo.
    It ensures that all layers have a channel number that is divisible by the divisor
    It can be seen at:
    https://github.com/tensorflow/models/blob/2cfc99eff5e5eb729c6793d2f3d03aa1c9be2b15/research/slim/nets/mobilenet/mobilenet.py#L62

    Args:
        v: input value
        divisor: default to 8
        min_value: minimum divisor value

    Returns:
        new_v: new divisible value
    """
    if min_value is None:
        min_value = divisor
    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
    # Make sure that round down does not go down by more than 10%.
    if new_v < 0.9 * v:
        new_v += divisor
    return new_v