Hugging Face's logo

Transformers documentation

AFMoE

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v4.57.3).
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 {release_date} and added to Hugging Face Transformers on 2025-11-18.

PyTorch FlashAttention SDPA

AFMoE

AFMoE (Arcee Foundational Mixture of Experts) is a decoder-only transformer model that extends the Llama architecture with a sparse Mixture of Experts (MoE) approach. The model combines token-choice routing with shared experts and employs several architectural innovations for efficient inference and improved performance.

Key Architecture Features

AFMoE introduces several key modifications to the standard transformer architecture:

  • Mixture of Experts with Shared Experts: Combines routed experts (activated per-token via learned routing) with always-active shared experts for stable base computation
  • Token-Choice Routing: Uses sigmoid or softmax-based routing with normalization and scaling for expert selection
  • Q/K Normalization and Gating: Applies RMSNorm to query and key projections and uses sigmoid gating on attention outputs for improved stability
  • Hybrid Attention Patterns: Alternates between sliding window attention and full attention across layers for efficiency with long contexts
  • Dual Normalization: Uses pre- and post-normalization around both attention and MLP blocks for training stability
  • Configurable Dense Layers: Allows initial layers to use dense MLPs before transitioning to sparse MoE layers

The model supports extended context lengths with RoPE embeddings and includes all standard Transformers features including Flash Attention 2, SDPA, gradient checkpointing, and quantization support.

AFMoE is particularly well-suited for scenarios requiring efficient scaling through sparsity while maintaining strong performance. The shared experts provide a stable computation baseline while routed experts enable model capacity scaling.

The example below demonstrates how to generate text with AFMoE using Pipeline or the AutoModel.

Pipeline
AutoModel 
import torch
from transformers import pipeline

pipeline = pipeline(
    task="text-generation",
    model="arcee-ai/Trinity-Mini",
    torch_dtype=torch.bfloat16,
    device=0
)

output = pipeline("The key innovation in mixture of experts is")
print(output[0]["generated_text"])

Model Architecture Details

Expert Routing

AFMoE uses token-choice routing where each token independently selects top-k experts based on router logits. The routing mechanism includes:

  • Configurable scoring function (sigmoid or softmax)
  • Optional route normalization for balanced expert utilization
  • Route scaling to control expert contribution strength
  • Bias correction for expert selection

Shared Experts

Unlike standard MoE models, AFMoE includes shared experts that are always activated for every token, providing:

  • A stable computation baseline across all tokens
  • Reduced variance in model outputs
  • Better handling of out-of-distribution inputs

Attention Mechanism

The hybrid attention pattern alternates between:

  • Sliding Window Attention: For efficiency on long sequences, with configurable window size
  • Full Attention: Applied every N layers (configurable via global_attn_every_n_layers) for global context

All attention layers include Q/K normalization and output gating for improved training dynamics.

AfmoeConfig

class transformers.AfmoeConfig

< >

( vocab_size: typing.Optional[int] = 200192 hidden_size: typing.Optional[int] = 2048 intermediate_size: typing.Optional[int] = 6144 moe_intermediate_size: typing.Optional[int] = 1408 num_hidden_layers: typing.Optional[int] = 32 num_dense_layers: typing.Optional[int] = 1 num_attention_heads: typing.Optional[int] = 16 num_key_value_heads: typing.Optional[int] = None head_dim: typing.Optional[int] = 128 hidden_act: typing.Optional[str] = 'silu' max_position_embeddings: typing.Optional[int] = 16384 initializer_range: typing.Optional[float] = 0.02 rms_norm_eps: typing.Optional[float] = 1e-05 use_cache: typing.Optional[bool] = True tie_word_embeddings: typing.Optional[bool] = False rope_theta: typing.Optional[float] = 10000.0 rope_parameters: typing.Union[transformers.modeling_rope_utils.RopeParameters, dict[str, transformers.modeling_rope_utils.RopeParameters], NoneType] = None num_experts: typing.Optional[int] = 64 num_experts_per_tok: typing.Optional[int] = 6 num_shared_experts: typing.Optional[int] = 2 route_scale: typing.Optional[float] = 1.0 global_attn_every_n_layers: typing.Optional[int] = 4 sliding_window: typing.Optional[int] = 1024 layer_types: typing.Optional[list] = None attention_dropout: typing.Optional[float] = 0.0 mup_enabled: typing.Optional[bool] = False **kwargs )

Parameters

  • vocab_size (int, optional, defaults to 200192) — Vocabulary size of the AFMoE model. Defines the number of different tokens that can be represented by the inputs_ids passed when calling AfmoeModel.
  • hidden_size (int, optional, defaults to 2048) — Dimension of the hidden representations.
  • intermediate_size (int, optional, defaults to 6144) — Dimension of the dense MLP representations.
  • moe_intermediate_size (int, optional, defaults to 1408) — Intermediate size of the routed expert MLPs.
  • num_hidden_layers (int, optional, defaults to 32) — Number of hidden layers in the Transformer decoder.
  • num_dense_layers (int, optional, defaults to 1) — Number of initial dense layers before MoE layers begin. Layers with index < num_dense_layers will use standard dense MLPs instead of MoE.
  • num_attention_heads (int, optional, defaults to 16) — Number of attention heads for each attention layer in the Transformer decoder.
  • num_key_value_heads (int, optional) — This is the number of key_value heads that should be used to implement Grouped Query Attention. If num_key_value_heads=num_attention_heads, the model will use Multi Head Attention (MHA), if num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details, check out this paper. If it is not specified, will default to num_attention_heads.
  • head_dim (int, optional, defaults to 128) — The dimension of each attention head.
  • hidden_act (str or function, optional, defaults to "silu") — The non-linear activation function (function or string) in the MLP blocks.
  • max_position_embeddings (int, optional, defaults to 16384) — The maximum sequence length that this model might ever be used with.
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
  • rms_norm_eps (float, optional, defaults to 1e-05) — The epsilon used by the RMS normalization layers.
  • 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.
  • tie_word_embeddings (bool, optional, defaults to False) — Whether the model’s input and output word embeddings should be tied.
  • rope_theta (float, optional, defaults to 10000.0) — The base period of the RoPE embeddings.
  • rope_parameters (RopeParameters, optional) — Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain a value for rope_theta and optionally parameters used for scaling in case you want to use RoPE with longer max_position_embeddings.
  • num_experts (int, optional, defaults to 64) — Number of routed experts in MoE layers.
  • num_experts_per_tok (int, optional, defaults to 6) — Number of experts to route each token to. This is the top-k value for the token-choice routing.
  • num_shared_experts (int, optional, defaults to 2) — Number of shared experts that are always activated for all tokens.
  • route_scale (float, optional, defaults to 1.0) — Scaling factor applied to routing weights.
  • global_attn_every_n_layers (int, optional, defaults to 4) — The frequency of full attention layers. Every Nth layer will use full attention, while others use sliding window attention.
  • sliding_window (int, optional, defaults to 1024) — Sliding window size for local attention layers.
  • layer_types (list[str], optional) — A list that explicitly maps each layer index with its attention type. Each element should be either “sliding_attention” or “full_attention”. If not provided, it will be automatically generated based on global_attn_every_n_layers.
  • attention_dropout (float, optional, defaults to 0.0) — The dropout ratio for the attention probabilities.
  • mup_enabled (bool, optional, defaults to False) — Whether to enable muP (Maximal Update Parametrization) input scaling. When enabled, input embeddings are scaled by sqrt(hidden_size).

This is the configuration class to store the configuration of a AfmoeModel. It is used to instantiate an AFMoE model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of arcee-ai/Trinity-Mini.

AFMoE is an Adaptive Feedforward MoE (Mixture of Experts) model with token-choice routing, shared experts, and a hybrid attention mechanism combining sliding window and full attention patterns.

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

Example:

>>> from transformers import AfmoeModel, AfmoeConfig

>>> # Initializing an AFMoE configuration
>>> configuration = AfmoeConfig()

>>> # Initializing a model from the afmoe-small-sft-v1 style configuration
>>> model = AfmoeModel(configuration)

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

AfmoeModel

class transformers.AfmoeModel

< >

( config: AfmoeConfig )

Parameters

  • config (AfmoeConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

The bare Afmoe Model outputting raw hidden-states without any specific head on top.

This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)

This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

forward

< >

( input_ids: typing.Optional[torch.LongTensor] = None attention_mask: typing.Optional[torch.Tensor] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None cache_position: typing.Optional[torch.LongTensor] = None use_cache: typing.Optional[bool] = None **kwargs: typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs] ) → transformers.modeling_outputs.MoeModelOutputWithPast 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?

  • 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.
  • 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 (~cache_utils.Cache, 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).

  • 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.
  • 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).

Returns

transformers.modeling_outputs.MoeModelOutputWithPast or tuple(torch.FloatTensor)

A transformers.modeling_outputs.MoeModelOutputWithPast 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 (None) and inputs.

  • last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.

  • 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 and optionally if config.is_encoder_decoder=True in the cross-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.

  • router_logits (tuple(torch.FloatTensor), optional, returned when output_router_probs=True and config.add_router_probs=True is passed or when config.output_router_probs=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, sequence_length, num_experts).

    Raw router logtis (post-softmax) that are computed by MoE routers, these terms are used to compute the auxiliary loss for Mixture of Experts models.

The AfmoeModel 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.

AfmoeForCausalLM

class transformers.AfmoeForCausalLM

< >

( config )

Parameters

  • config (AfmoeForCausalLM) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.

The Afmoe Model for causal language modeling.

This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)

This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.

forward

< >

( input_ids: typing.Optional[torch.LongTensor] = None attention_mask: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None labels: typing.Optional[torch.LongTensor] = None use_cache: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = None logits_to_keep: typing.Union[int, torch.Tensor] = 0 **kwargs: typing_extensions.Unpack[transformers.utils.generic.TransformersKwargs] ) → transformers.modeling_outputs.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 (~cache_utils.Cache, 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], 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

transformers.modeling_outputs.CausalLMOutputWithPast or tuple(torch.FloatTensor)

A transformers.modeling_outputs.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 (AfmoeConfig) and inputs.

  • 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.

The AfmoeForCausalLM 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.

Example:

>>> from transformers import AutoTokenizer, AfmoeForCausalLM

>>> model = AfmoeForCausalLM.from_pretrained("meta-afmoe/Afmoe-2-7b-hf")
>>> tokenizer = AutoTokenizer.from_pretrained("meta-afmoe/Afmoe-2-7b-hf")

>>> 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."
Update on GitHub

←Kernels ALBERT→