GitLab Repo

amachine.am_transformers.am_control_model_exp 

1from .am_control_model_exp import ControlGmhForCausalLM, ControlGmhConfig
2
3__all__ = ["ControlGmhConfig", "ControlGmhForCausalLM"]
class ControlGmhConfig(transformers.models.granitemoehybrid.configuration_granitemoehybrid.GraniteMoeHybridConfig): 
38class ControlGmhConfig(GraniteMoeHybridConfig):
39    
40    model_type = my_model_type
41
42    def __init__(
43        self, 
44        control_params = None,
45        **kwargs 
46    ) :
47        super().__init__(**kwargs)
48
49        if control_params is None :
50            control_params = {}
51
52        control_params[ "add_control" ]                   = control_params.get( "add_control",                         True )
53        control_params[ "override_control_residual" ]     = control_params.get( "override_control_residual",           True )
54        control_params[ "bypass_control_output" ]         = control_params.get( "bypass_control_output",              False )
55        control_params[ "add_control_loss" ]              = control_params.get( "add_control_loss",                    True )
56        control_params[ "control_loss_skip_tokens" ]      = control_params.get( "control_loss_skip_tokens",               0 )
57        control_params[ "control_loss_schedule_tokens" ]  = control_params.get( "control_loss_schedule_tokens", 183_500_800 )
58        control_params[ "alpha_c" ]                       = control_params.get( "alpha_c",                              0.2 )
59        control_params[ "alpha_w" ]                       = control_params.get( "alpha_w",                              2.0 )
60        control_params[ "beta_div" ]                      = control_params.get( "beta_div",                             0.2 )
61        control_params[ "covariance_weight" ]             = control_params.get( "covariance_weight",                    1.0 )
62
63        assert isinstance( self.layer_types , list )
64
65        cl = control_params.get( "control_layer", len( self.layer_types ) // 2 + 1 )
66        control_params[ "control_layer" ]  = cl
67
68        assert isinstance( cl, int )
69        assert cl > 0 and cl < len( self.layer_types ) - 1
70
71        self.control_params = control_params
ControlGmhConfig(control_params=None, **kwargs) 
42    def __init__(
43        self, 
44        control_params = None,
45        **kwargs 
46    ) :
47        super().__init__(**kwargs)
48
49        if control_params is None :
50            control_params = {}
51
52        control_params[ "add_control" ]                   = control_params.get( "add_control",                         True )
53        control_params[ "override_control_residual" ]     = control_params.get( "override_control_residual",           True )
54        control_params[ "bypass_control_output" ]         = control_params.get( "bypass_control_output",              False )
55        control_params[ "add_control_loss" ]              = control_params.get( "add_control_loss",                    True )
56        control_params[ "control_loss_skip_tokens" ]      = control_params.get( "control_loss_skip_tokens",               0 )
57        control_params[ "control_loss_schedule_tokens" ]  = control_params.get( "control_loss_schedule_tokens", 183_500_800 )
58        control_params[ "alpha_c" ]                       = control_params.get( "alpha_c",                              0.2 )
59        control_params[ "alpha_w" ]                       = control_params.get( "alpha_w",                              2.0 )
60        control_params[ "beta_div" ]                      = control_params.get( "beta_div",                             0.2 )
61        control_params[ "covariance_weight" ]             = control_params.get( "covariance_weight",                    1.0 )
62
63        assert isinstance( self.layer_types , list )
64
65        cl = control_params.get( "control_layer", len( self.layer_types ) // 2 + 1 )
66        control_params[ "control_layer" ]  = cl
67
68        assert isinstance( cl, int )
69        assert cl > 0 and cl < len( self.layer_types ) - 1
70
71        self.control_params = control_params
Arguments:
  • vocab_size (int, optional, defaults to 32000): Vocabulary size of the model. Defines the number of different tokens that can be represented by the input_ids.
  • hidden_size (int, optional, defaults to 4096): Dimension of the hidden representations.
  • intermediate_size (int, optional, defaults to 11008): Dimension of the MLP representations.
  • num_hidden_layers (int, optional, defaults to 32): Number of hidden layers in the Transformer decoder.
  • num_attention_heads (int, optional, defaults to 32): 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.
  • hidden_act (str, optional, defaults to silu): The non-linear activation function (function or string) in the decoder. For example, "gelu", "relu", "silu", etc.
  • max_position_embeddings (int, optional, defaults to 2048): 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-06): 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 or when the model is a decoder-only generative model.
  • pad_token_id (int, optional): Token id used for padding in the vocabulary.
  • bos_token_id (int, optional, defaults to 1): Token id used for beginning-of-stream in the vocabulary.
  • eos_token_id (Union[int, list[int]], optional, defaults to 2): Token id used for end-of-stream in the vocabulary.
  • tie_word_embeddings (bool, optional, defaults to False): Whether to tie weight embeddings according to model's tied_weights_keys mapping.
  • rope_parameters (Union[~modeling_rope_utils.RopeParameters, dict], 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.
  • attention_bias (bool, optional, defaults to False): Whether to use a bias in the query, key, value and output projection layers during self-attention.
  • attention_dropout (Union[float, int], optional, defaults to 0.0): The dropout ratio for the attention probabilities.
  • embedding_multiplier (float, optional, defaults to 1.0): embedding multiplier.
  • logits_scaling (float, optional, defaults to 1.0): divisor for output logits.
  • residual_multiplier (float, optional, defaults to 1.0): residual multiplier.
  • attention_multiplier (float, optional, defaults to 1.0): attention multiplier.
  • num_local_experts (int, optional, defaults to 8): Number of local experts on each device. num_experts should be divisible by num_local_experts.
  • num_experts_per_tok (int, optional, defaults to 2): Number of experts to route each token to. This is the top-k value for the token-choice routing.
  • output_router_logits (bool, optional, defaults to False): Whether or not the router logits should be returned by the model. Enabling this will also allow the model to output the auxiliary loss, including load balancing loss and router z-loss.
  • router_aux_loss_coef (float, optional, defaults to 0.001): Auxiliary load balancing loss coefficient. Used to penalize uneven expert routing in MoE models.
  • shared_intermediate_size (int, optional, defaults to 1024): intermediate size for shared experts.
  • position_embedding_type (str, optional): Positional embedding type to be used; defaults to None. Allowed options: [None, "rope"]
  • layer_types (list[str], optional): A list that explicitly maps each layer index with its layer type. If not provided, it will be automatically generated based on config values.
  • mamba_n_heads (int, optional, defaults to 128): The number of mamba heads used in the v2 implementation.
  • mamba_n_groups (int, optional, defaults to 1): The number of the mamba groups used in the v2 implementation.
  • mamba_d_state (int, optional, defaults to 256): Size of the SSM state (latent state dimension) in the Mamba layers.
  • mamba_d_head (Union[int, str], optional, defaults to auto): Head embedding dimension size
  • mamba_d_conv (int, optional, defaults to 4): The size of the mamba convolution kernel
  • mamba_expand (int, optional, defaults to 2): Expanding factor (relative to hidden_size) used to determine the mamba intermediate size
  • mamba_chunk_size (int, optional, defaults to 256): The chunks in which to break the sequence when doing prefill/training
  • mamba_conv_bias (bool, optional, defaults to True): Flag indicating whether or not to use bias in the convolution layer of the mamba mixer block.
  • mamba_proj_bias (bool, optional, defaults to False): Flag indicating whether or not to use bias in the input and output projections (["in_proj", "out_proj"]) of the mamba mixer block
  • time_step_min (float, optional, defaults to 0.001): Minimum time_step used to bound dt_proj.bias.
  • time_step_max (float, optional, defaults to 0.1): Maximum time_step used to bound dt_proj.bias.
  • time_step_limit (Union[list[float, float], tuple[float, float]], optional, defaults to (0.0, inf)): Accepted range of time step values for clamping.

Example:

>>> from transformers import GraniteMoeHybridModel, GraniteMoeHybridConfig

>>> # Initializing a GraniteMoeHybrid config
>>> configuration = GraniteMoeHybridConfig()

>>> # Accessing the model configuration
>>> configuration = model.config
model_type = 'control_gmh'
control_params
class ControlGmhForCausalLM(transformers.models.granitemoehybrid.modeling_granitemoehybrid.GraniteMoeHybridForCausalLM): 
303class ControlGmhForCausalLM(GraniteMoeHybridForCausalLM):
304    
305    config_class = ControlGmhConfig          
306    _no_split_modules = ["ControlGmhDecoderLayer"]
307
308    def __init__(self, config):
309        
310        super().__init__(config)
311        
312        self.model = ControlGmhModel(config)
313        self.register_buffer("global_token_counter", torch.tensor(0, dtype=torch.long))
314        self.post_init()
315
316    @can_return_tuple
317    def forward(
318        self,
319        input_ids: torch.LongTensor | None = None,
320        attention_mask: torch.Tensor | None = None,
321        position_ids: torch.LongTensor | None = None,
322        past_key_values: Cache | None = None,
323        inputs_embeds: torch.FloatTensor | None = None,
324        labels: torch.LongTensor | None = None,
325        output_router_logits: bool | None = None,
326        logits_to_keep: int | torch.Tensor = 0,
327        **kwargs,
328    ) -> tuple | MoeCausalLMOutputWithPast:
329
330        output_router_logits = (
331            output_router_logits if output_router_logits is not None else self.config.output_router_logits
332        )
333        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
334        outputs = self.model(
335            input_ids=input_ids,
336            attention_mask=attention_mask,
337            position_ids=position_ids,
338            past_key_values=past_key_values,
339            inputs_embeds=inputs_embeds,
340            **kwargs,
341        )
342
343        # Only compute necessary logits
344        hidden_states = outputs.last_hidden_state
345        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
346        logits = self.lm_head(hidden_states[:, slice_indices, :])
347        logits = logits / self.config.logits_scaling
348
349        loss = None
350        if labels is not None:
351            # Flatten the tokens
352            loss = self.loss_function(
353                logits,
354                labels,
355                vocab_size=self.config.vocab_size,
356                **kwargs,
357            )
358
359        aux_loss = None
360        if output_router_logits:
361            aux_loss = load_balancing_loss_func(
362                outputs.router_logits,
363                self.num_experts,
364                self.num_experts_per_tok,
365                attention_mask,
366            )
367            if labels is not None:
368                loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
369
370        #############################################################
371        # Custom Information-Theoretic Synergy Loss
372        #############################################################
373        
374        main_loss = loss
375        metrics = {}
376        
377        # Only execute if we have a valid main loss (training mode) and the required control outputs
378        if main_loss is not None and hasattr(outputs, "control_output") and hasattr(outputs, "without_control"):
379
380            entropy_chunk_size = kwargs.get( "entropy_chunk_size", 512 )
381
382            X_c = outputs.control_output[:, slice_indices, :]
383            X_w = outputs.without_control[:, slice_indices, :]
384
385            tokens_seen = self.global_token_counter
386            self.global_token_counter += ( X_c.shape[ 0 ]*X_c.shape[ 1 ] )
387
388            if X_c.ndim != 3 or X_w.ndim != 3:
389                raise ValueError("Expected X_c and X_w to have shape (B, S, D)")
390            if X_c.shape != X_w.shape:
391                raise ValueError(f"X_c and X_w must have matching shapes, got {X_c.shape} and {X_w.shape}")
392                
393            X_c = X_c.reshape(-1, X_c.shape[-1])
394            X_w = X_w.reshape(-1, X_w.shape[-1])
395            
396            if X_c.shape[0] > 0:
397
398                #-----------------------------------------------------------------------
399
400                add_control_loss = self.config.control_params.get( "add_control_loss" )
401                control_loss_schedule_tokens = self.config.control_params.get( "control_loss_schedule_tokens" )
402                control_loss_skip_tokens = self.config.control_params.get( "control_loss_skip_tokens" )
403
404                assert isinstance( add_control_loss, bool )
405                assert isinstance( control_loss_schedule_tokens, int )
406                assert isinstance( control_loss_skip_tokens, int )
407
408                alpha_c  = self.config.control_params.get( "alpha_c"  )
409                alpha_w  = self.config.control_params.get( "alpha_w"  )
410                beta_div = self.config.control_params.get( "beta_div" )
411
412                assert isinstance( alpha_c, float )
413                assert isinstance( alpha_w, float )
414                assert isinstance( beta_div, float )
415
416                covariance_weight = self.config.control_params.get( "covariance_weight" )
417
418                assert isinstance( covariance_weight, float )
419
420                #----------------------------------------------------------------------------
421
422                logits_scaling = float(self.config.logits_scaling)
423                
424                if logits_scaling <= 0:
425                    raise ValueError("logits_scaling must be positive")
426                
427                # ------------------------------------------------------------------
428                # Predictive-information proxy
429                # ------------------------------------------------------------------
430
431                loss_info_c = normalized_lm_head_entropy_deficit(
432                    hidden=X_c, 
433                    lm_head=self.lm_head, 
434                    logits_scaling=logits_scaling, 
435                    use_checkpoint=True,
436                    chunk_size=128,
437                    include_bias=False,
438                    non_uniformity_weight=1.0
439                )
440                
441                loss_info_w = normalized_lm_head_entropy_deficit(
442                    hidden=X_w, 
443                    lm_head=self.lm_head, 
444                    logits_scaling=logits_scaling, 
445                    use_checkpoint=True,
446                    chunk_size=128,
447                    include_bias=False,
448                    non_uniformity_weight=1.0
449                )
450
451                # ------------------------------------------------------------------
452                # VICReg-style non-collapse and feature decorrelation
453                # ------------------------------------------------------------------
454
455                loss_covariance_c = covariance_loss(X_c) * ( X_c.shape[-1] / 4.0 )
456                loss_div_c = covariance_weight * loss_covariance_c
457
458                # ------------------------------------------------------------------
459                # Schedule and aggregation
460                # ------------------------------------------------------------------
461                
462                schedule_position = max( 0, tokens_seen - control_loss_skip_tokens ) / control_loss_schedule_tokens
463
464                schedule_c = get_sigmoid_weight(
465                    r=schedule_position,
466                    midpoint=0.5,
467                    steepness=12.0,
468                )
469                
470                weighted_ctrl = alpha_c * loss_info_c + alpha_w * loss_info_w + beta_div * loss_div_c
471                loss_ctrl_scheduled = schedule_c * weighted_ctrl
472
473                if add_control_loss :
474                    loss = main_loss + schedule_c * weighted_ctrl
475
476                metrics = {
477                    "main_loss" : main_loss.detach(),
478                    "loss_info_c": loss_info_c.detach(),
479                    "loss_info_w": loss_info_w.detach(),
480                    "loss_covariance_c": loss_covariance_c.detach(),
481                    "loss_diversity_c": loss_div_c.detach(),
482                    "loss_ctrl_weighted": weighted_ctrl.detach(),
483                    "loss_ctrl_scheduled": loss_ctrl_scheduled.detach(),
484                    "train_synergy_schedule": schedule_c,
485                }
486
487        ############################################################
488
489        return ControlCausalLMOutput(
490            loss=loss,
491            aux_loss=aux_loss,
492            logits=logits,
493            past_key_values=outputs.past_key_values,
494            hidden_states=outputs.hidden_states,
495            attentions=outputs.attentions,
496            router_logits=outputs.router_logits,
497            **metrics
498        )

The Granitemoehybrid 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.

Arguments:
  • config ([GraniteMoeHybridConfig]): 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 [~PreTrainedModel.from_pretrained] method to load the model weights.
ControlGmhForCausalLM(config) 
308    def __init__(self, config):
309        
310        super().__init__(config)
311        
312        self.model = ControlGmhModel(config)
313        self.register_buffer("global_token_counter", torch.tensor(0, dtype=torch.long))
314        self.post_init()

Args: config ([GraniteMoeHybridConfig]): 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 [~PreTrainedModel.from_pretrained] method to load the model weights.

config_class = <class 'ControlGmhConfig'>
model
@can_return_tuple
def forward( self, input_ids: torch.LongTensor | None = None, attention_mask: torch.Tensor | None = None, position_ids: torch.LongTensor | None = None, past_key_values: transformers.cache_utils.Cache | None = None, inputs_embeds: torch.FloatTensor | None = None, labels: torch.LongTensor | None = None, output_router_logits: bool | None = None, logits_to_keep: int | torch.Tensor = 0, **kwargs) -> tuple | transformers.modeling_outputs.MoeCausalLMOutputWithPast: 
316    @can_return_tuple
317    def forward(
318        self,
319        input_ids: torch.LongTensor | None = None,
320        attention_mask: torch.Tensor | None = None,
321        position_ids: torch.LongTensor | None = None,
322        past_key_values: Cache | None = None,
323        inputs_embeds: torch.FloatTensor | None = None,
324        labels: torch.LongTensor | None = None,
325        output_router_logits: bool | None = None,
326        logits_to_keep: int | torch.Tensor = 0,
327        **kwargs,
328    ) -> tuple | MoeCausalLMOutputWithPast:
329
330        output_router_logits = (
331            output_router_logits if output_router_logits is not None else self.config.output_router_logits
332        )
333        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
334        outputs = self.model(
335            input_ids=input_ids,
336            attention_mask=attention_mask,
337            position_ids=position_ids,
338            past_key_values=past_key_values,
339            inputs_embeds=inputs_embeds,
340            **kwargs,
341        )
342
343        # Only compute necessary logits
344        hidden_states = outputs.last_hidden_state
345        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
346        logits = self.lm_head(hidden_states[:, slice_indices, :])
347        logits = logits / self.config.logits_scaling
348
349        loss = None
350        if labels is not None:
351            # Flatten the tokens
352            loss = self.loss_function(
353                logits,
354                labels,
355                vocab_size=self.config.vocab_size,
356                **kwargs,
357            )
358
359        aux_loss = None
360        if output_router_logits:
361            aux_loss = load_balancing_loss_func(
362                outputs.router_logits,
363                self.num_experts,
364                self.num_experts_per_tok,
365                attention_mask,
366            )
367            if labels is not None:
368                loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
369
370        #############################################################
371        # Custom Information-Theoretic Synergy Loss
372        #############################################################
373        
374        main_loss = loss
375        metrics = {}
376        
377        # Only execute if we have a valid main loss (training mode) and the required control outputs
378        if main_loss is not None and hasattr(outputs, "control_output") and hasattr(outputs, "without_control"):
379
380            entropy_chunk_size = kwargs.get( "entropy_chunk_size", 512 )
381
382            X_c = outputs.control_output[:, slice_indices, :]
383            X_w = outputs.without_control[:, slice_indices, :]
384
385            tokens_seen = self.global_token_counter
386            self.global_token_counter += ( X_c.shape[ 0 ]*X_c.shape[ 1 ] )
387
388            if X_c.ndim != 3 or X_w.ndim != 3:
389                raise ValueError("Expected X_c and X_w to have shape (B, S, D)")
390            if X_c.shape != X_w.shape:
391                raise ValueError(f"X_c and X_w must have matching shapes, got {X_c.shape} and {X_w.shape}")
392                
393            X_c = X_c.reshape(-1, X_c.shape[-1])
394            X_w = X_w.reshape(-1, X_w.shape[-1])
395            
396            if X_c.shape[0] > 0:
397
398                #-----------------------------------------------------------------------
399
400                add_control_loss = self.config.control_params.get( "add_control_loss" )
401                control_loss_schedule_tokens = self.config.control_params.get( "control_loss_schedule_tokens" )
402                control_loss_skip_tokens = self.config.control_params.get( "control_loss_skip_tokens" )
403
404                assert isinstance( add_control_loss, bool )
405                assert isinstance( control_loss_schedule_tokens, int )
406                assert isinstance( control_loss_skip_tokens, int )
407
408                alpha_c  = self.config.control_params.get( "alpha_c"  )
409                alpha_w  = self.config.control_params.get( "alpha_w"  )
410                beta_div = self.config.control_params.get( "beta_div" )
411
412                assert isinstance( alpha_c, float )
413                assert isinstance( alpha_w, float )
414                assert isinstance( beta_div, float )
415
416                covariance_weight = self.config.control_params.get( "covariance_weight" )
417
418                assert isinstance( covariance_weight, float )
419
420                #----------------------------------------------------------------------------
421
422                logits_scaling = float(self.config.logits_scaling)
423                
424                if logits_scaling <= 0:
425                    raise ValueError("logits_scaling must be positive")
426                
427                # ------------------------------------------------------------------
428                # Predictive-information proxy
429                # ------------------------------------------------------------------
430
431                loss_info_c = normalized_lm_head_entropy_deficit(
432                    hidden=X_c, 
433                    lm_head=self.lm_head, 
434                    logits_scaling=logits_scaling, 
435                    use_checkpoint=True,
436                    chunk_size=128,
437                    include_bias=False,
438                    non_uniformity_weight=1.0
439                )
440                
441                loss_info_w = normalized_lm_head_entropy_deficit(
442                    hidden=X_w, 
443                    lm_head=self.lm_head, 
444                    logits_scaling=logits_scaling, 
445                    use_checkpoint=True,
446                    chunk_size=128,
447                    include_bias=False,
448                    non_uniformity_weight=1.0
449                )
450
451                # ------------------------------------------------------------------
452                # VICReg-style non-collapse and feature decorrelation
453                # ------------------------------------------------------------------
454
455                loss_covariance_c = covariance_loss(X_c) * ( X_c.shape[-1] / 4.0 )
456                loss_div_c = covariance_weight * loss_covariance_c
457
458                # ------------------------------------------------------------------
459                # Schedule and aggregation
460                # ------------------------------------------------------------------
461                
462                schedule_position = max( 0, tokens_seen - control_loss_skip_tokens ) / control_loss_schedule_tokens
463
464                schedule_c = get_sigmoid_weight(
465                    r=schedule_position,
466                    midpoint=0.5,
467                    steepness=12.0,
468                )
469                
470                weighted_ctrl = alpha_c * loss_info_c + alpha_w * loss_info_w + beta_div * loss_div_c
471                loss_ctrl_scheduled = schedule_c * weighted_ctrl
472
473                if add_control_loss :
474                    loss = main_loss + schedule_c * weighted_ctrl
475
476                metrics = {
477                    "main_loss" : main_loss.detach(),
478                    "loss_info_c": loss_info_c.detach(),
479                    "loss_info_w": loss_info_w.detach(),
480                    "loss_covariance_c": loss_covariance_c.detach(),
481                    "loss_diversity_c": loss_div_c.detach(),
482                    "loss_ctrl_weighted": weighted_ctrl.detach(),
483                    "loss_ctrl_scheduled": loss_ctrl_scheduled.detach(),
484                    "train_synergy_schedule": schedule_c,
485                }
486
487        ############################################################
488
489        return ControlCausalLMOutput(
490            loss=loss,
491            aux_loss=aux_loss,
492            logits=logits,
493            past_key_values=outputs.past_key_values,
494            hidden_states=outputs.hidden_states,
495            attentions=outputs.attentions,
496            router_logits=outputs.router_logits,
497            **metrics
498        )

The [GraniteMoeHybridForCausalLM] 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.

Arguments:

  • 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_utils.Cache] instance is allowed as input, see our kv cache guide. If no past_key_values are passed, [~cache_utils.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].
  • output_router_logits (bool, optional): Whether or not to return the logits of all the routers. They are useful for computing the router loss, and should not be returned during inference.
  • 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:

[~modeling_outputs.MoeCausalLMOutputWithPast] or tuple(torch.FloatTensor): A [~modeling_outputs.MoeCausalLMOutputWithPast] 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.

  • 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).
  • aux_loss (torch.FloatTensor, optional, returned when labels is provided) -- aux_loss for the sparse modules.

  • 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.

  • past_key_values (Cache, optional, returned when use_cache=True is passed or when config.use_cache=True) -- It is a [~cache_utils.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, GraniteMoeHybridForCausalLM

>>> model = GraniteMoeHybridForCausalLM.from_pretrained("ibm-granite/granite-4.0-h-tiny")
>>> tokenizer = AutoTokenizer.from_pretrained("ibm-granite/granite-4.0-h-tiny")

>>> 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."