Hugging Face's logo

Transformers documentation

NemotronHConfig

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v5.2.0).
Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

This model was released on 2025-12-15 and added to Hugging Face Transformers on 2026-03-02.

NemotronHConfig

class transformers.NemotronHConfig

< >

( vocab_size = 131072 hidden_size = 4096 layers_block_type = None num_hidden_layers = None tie_word_embeddings = False use_cache = True num_logits_to_keep = 1 pad_token_id = 0 bos_token_id = 1 eos_token_id = 2 num_attention_heads = 32 num_key_value_heads = 8 head_dim = 128 max_position_embeddings = 4096 attention_bias = False attention_dropout = 0.0 sliding_window = None intermediate_size = 21504 mlp_hidden_act = 'relu2' mlp_bias = False use_mamba_kernels = True ssm_state_size = 128 mamba_num_heads = 128 mamba_n_groups = 8 mamba_head_dim = 64 mamba_d_conv = 4 mamba_expand = 2 mamba_hidden_act = 'silu' mamba_dt_min = 0.001 mamba_dt_max = 0.1 mamba_dt_limit = (0.0, inf) mamba_dt_init_floor = 0.0001 mamba_conv_bias = True mamba_proj_bias = False mamba_chunk_size = 128 mamba_ssm_cache_dtype = 'float32' n_routed_experts = 8 n_shared_experts = 1 moe_intermediate_size = 7688 moe_shared_expert_intermediate_size = 7688 moe_latent_size = None moe_shared_expert_overlap = True num_experts_per_tok = 2 routed_scaling_factor = 1.0 n_group = 1 topk_group = 1 norm_topk_prob = True num_nextn_predict_layers = 0 mtp_layers_block_type = ['attention', 'moe'] use_bias = False initializer_range = 0.02 layer_norm_epsilon = 1e-05 residual_in_fp32 = False hidden_dropout = 0.0 rescale_prenorm_residual = True **kwargs )

Parameters

  • vocab_size (int, optional, defaults to 131072) — Vocabulary size of the NemotronH model. Defines the number of different tokens that can be represented by the inputs_ids passed when calling NemotronHModel.
  • hidden_size (int, optional, defaults to 4096) — Dimension of the hidden representations.
  • layers_block_type (list, optional) — Explicit list of layer types for each layer. Each element must be one of: “mamba”, “attention”, or “moe”. The number of layers is determined by the length of this list.
  • num_hidden_layers (int, optional) — Number of hidden layers in the Transformer encoder. This parameter is deprecated and only kept for backward compatibility. The number of layers is now determined by the length of layers_block_type.
  • tie_word_embeddings (bool, optional, defaults to False) — Whether the model’s input and output word embeddings should be tied.
  • use_cache (bool, optional, defaults to True) — Whether or not the model should return the last key/values attentions.
  • num_logits_to_keep (int, optional, defaults to 1) — Number of prompt logits to calculate during generation. If None, all logits will be calculated.
  • pad_token_id (int, optional, defaults to 0) — The id of the padding token.
  • bos_token_id (int, optional, defaults to 1) — The id of the “beginning-of-sequence” token.
  • eos_token_id (int, optional, defaults to 2) — The id of the “end-of-sequence” token.
  • num_attention_heads (int, optional, defaults to 32) — Number of attention heads for each attention layer in the Transformer encoder.
  • num_key_value_heads (int, optional, defaults to 8) — This is the number of key_value heads that should be used to implement Grouped Query Attention.
  • head_dim (int, optional, defaults to 128) — Dimension of each attention head.
  • max_position_embeddings (int, optional, defaults to 4096) — The maximum sequence length that this model might ever be used with.
  • attention_bias (bool, optional, defaults to False) — Whether to use bias in attention layers.
  • attention_dropout (float, optional, defaults to 0.0) — The dropout ratio for the attention probabilities.
  • sliding_window (int, optional) — Sliding window attention window size.
  • intermediate_size (int, optional, defaults to 21504) — Dimension of the MLP representations.
  • mlp_hidden_act (str, optional, defaults to "relu2") — The non-linear activation function in the MLP layers.
  • mlp_bias (bool, optional, defaults to False) — Whether to use bias in MLP layers.
  • use_mamba_kernels (bool, optional, defaults to True) — Flag indicating whether or not to use the fast mamba kernels.
  • ssm_state_size (int, optional, defaults to 128) — The dimension of the mamba state space latents.
  • mamba_num_heads (int, optional, defaults to 128) — Number of heads in Mamba layers.
  • mamba_n_groups (int, optional, defaults to 8) — Number of groups in Mamba layers.
  • mamba_head_dim (int, optional, defaults to 64) — Dimension of each Mamba head.
  • mamba_d_conv (int, optional, defaults to 4) — The size of the mamba convolution kernel.
  • mamba_expand (int, optional, defaults to 2) — Expanding factor used to determine the mamba intermediate size.
  • mamba_hidden_act (str, optional, defaults to "silu") — The non-linear activation function in the Mamba layers.
  • mamba_dt_min (float, optional, defaults to 0.001) — Minimum value for the time step in Mamba.
  • mamba_dt_max (float, optional, defaults to 0.1) — Maximum value for the time step in Mamba.
  • mamba_dt_limit (tuple, optional, defaults to (0.0, inf)) — Limits for the time step in Mamba.
  • mamba_dt_init_floor (float, optional, defaults to 0.0001) — Floor value for time step initialization in Mamba.
  • mamba_conv_bias (bool, optional, defaults to True) — Whether to use bias in the convolution layer of the mamba mixer block.
  • mamba_proj_bias (bool, optional, defaults to False) — Whether to use bias in the input and output projections of the mamba mixer block.
  • mamba_chunk_size (int, optional, defaults to 128) — Size of chunks for Mamba processing.
  • mamba_ssm_cache_dtype (str, optional, defaults to "float32") — Data type for Mamba SSM cache states.
  • n_routed_experts (int, optional, defaults to 8) — Number of routed experts in MoE layers.
  • n_shared_experts (int, optional, defaults to 1) — Number of shared experts that are always activated in MoE layers.
  • moe_intermediate_size (int, optional, defaults to 7688) — Dimension of the MLP representations in routed experts.
  • moe_shared_expert_intermediate_size (int, optional, defaults to 7688) — Dimension of the MLP representations in shared experts.
  • moe_latent_size (int, optional) — Latent size for MoE expert projections. If None, uses hidden_size.
  • moe_shared_expert_overlap (bool, optional, defaults to True) — Whether shared experts overlap with routed experts.
  • num_experts_per_tok (int, optional, defaults to 2) — The number of experts to route per token (top-k routing parameter).
  • routed_scaling_factor (float, optional, defaults to 1.0) — Scaling factor applied to routed expert outputs.
  • n_group (int, optional, defaults to 1) — Number of groups for expert routing.
  • topk_group (int, optional, defaults to 1) — Top-k group parameter for expert selection.
  • norm_topk_prob (bool, optional, defaults to True) — Whether to normalize top-k probabilities in expert routing.
  • num_nextn_predict_layers (int, optional, defaults to 0) — Number of additional layers for multi-token prediction. If 0, multi-token prediction is disabled.
  • mtp_layers_block_type (list, optional, defaults to ['attention', 'moe']) — Explicit list of layer types for multi-token prediction layers when num_nextn_predict_layers > 0.
  • use_bias (bool, optional, defaults to False) — Whether to use bias in the model.
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
  • layer_norm_epsilon (float, optional, defaults to 1e-05) — The epsilon used by the layer normalization layers.
  • residual_in_fp32 (bool, optional, defaults to False) — Whether or not residuals should be in float32.
  • hidden_dropout (float, optional, defaults to 0.0) — The dropout ratio for the hidden states.
  • rescale_prenorm_residual (bool, optional, defaults to True) — Whether to rescale the pre-normalization residual connections.

This is the configuration class to store the configuration of a NemotronHModel. It is used to instantiate a NemotronH model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16.

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

>>> from transformers import NemotronHModel, NemotronHConfig

>>> # Initializing a NemotronH configuration
>>> configuration = NemotronHConfig()

>>> # Initializing a model (with random weights) from the configuration
>>> model = NemotronHModel(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

NemotronHForCausalLM

class transformers.NemotronHForCausalLM

< >

( config )

forward

< >

( input_ids: torch.LongTensor | None = None attention_mask: torch.Tensor | None = None position_ids: torch.LongTensor | None = None past_key_values: transformers.models.nemotron_h.modeling_nemotron_h.NemotronHHybridDynamicCache | None = None inputs_embeds: torch.FloatTensor | None = None labels: torch.LongTensor | None = None use_cache: bool | None = None cache_position: torch.LongTensor | None = None logits_to_keep: int | torch.Tensor = 0 **kwargs ) → CausalLMOutputWithPast or tuple(torch.FloatTensor)

Parameters

  • input_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default.

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    What are input IDs?

  • attention_mask (torch.Tensor of shape (batch_size, sequence_length), optional) — Mask to avoid performing attention on padding token indices. Mask values selected in [0, 1]:

    • 1 for tokens that are not masked,
    • 0 for tokens that are masked.

    What are attention masks?

  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1].

    What are position IDs?

  • past_key_values (~models.nemotron_h.modeling_nemotron_h.NemotronHHybridDynamicCache, optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the past_key_values returned by the model at a previous stage of decoding, when use_cache=True or config.use_cache=True.

    Only Cache instance is allowed as input, see our kv cache guide. If no past_key_values are passed, DynamicCache will be initialized by default.

    The model will output the same cache format that is fed as input.

    If past_key_values are used, the user is expected to input only unprocessed input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, unprocessed_length) instead of all input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • labels (torch.LongTensor of shape (batch_size, sequence_length), optional) — Labels for computing the masked language modeling loss. Indices should either be in [0, ..., config.vocab_size] or -100 (see input_ids docstring). Tokens with indices set to -100 are ignored (masked), the loss is only computed for the tokens with labels in [0, ..., config.vocab_size].
  • 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).
  • cache_position (torch.LongTensor of shape (sequence_length), optional) — Indices depicting the position of the input sequence tokens in the sequence. Contrarily to position_ids, this tensor is not affected by padding. It is used to update the cache in the correct position and to infer the complete sequence length.
  • logits_to_keep (Union[int, torch.Tensor], optional, defaults to 0) — If an int, compute logits for the last logits_to_keep tokens. If 0, calculate logits for all input_ids (special case). Only last token logits are needed for generation, and calculating them only for that token can save memory, which becomes pretty significant for long sequences or large vocabulary size. If a torch.Tensor, must be 1D corresponding to the indices to keep in the sequence length dimension. This is useful when using packed tensor format (single dimension for batch and sequence length).

Returns

CausalLMOutputWithPast or tuple(torch.FloatTensor)

A CausalLMOutputWithPast or a tuple of torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various elements depending on the configuration (NemotronHConfig) and inputs.

The NemotronHForCausalLM forward method, overrides the __call__ special method.

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the pre and post processing steps while the latter silently ignores them.

  • loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction).

  • logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).

  • past_key_values (Cache, optional, returned when use_cache=True is passed or when config.use_cache=True) — It is a Cache instance. For more details, see our kv cache guide.

    Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.

  • hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.FloatTensor (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape (batch_size, sequence_length, hidden_size).

    Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.

  • attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

    Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.

Example:

>>> from transformers import AutoTokenizer, NemotronHForCausalLM

>>> model = NemotronHForCausalLM.from_pretrained("Zyphra/NemotronH-7B-v1")
>>> tokenizer = AutoTokenizer.from_pretrained("Zyphra/NemotronH-7B-v1")

>>> prompt = "Hey, are you conscious? Can you talk to me?"
>>> inputs = tokenizer(prompt, return_tensors="pt")

>>> # Generate
>>> generate_ids = model.generate(inputs.input_ids, max_length=30)
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
"Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."

NemotronHModel

class transformers.NemotronHModel

< >

( config )

forward

< >

( input_ids: torch.LongTensor | None = None inputs_embeds: torch.LongTensor | None = None position_ids: torch.LongTensor | None = None past_key_values: transformers.models.nemotron_h.modeling_nemotron_h.NemotronHHybridDynamicCache | None = None use_cache: bool | None = None cache_position: torch.LongTensor | None = None attention_mask: torch.Tensor | None = None **kwargs: typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs] )

Update on GitHub

←Nemotron NLLB→