diff --git a/docs/source/en/_toctree.yml b/docs/source/en/_toctree.yml index 14cf7340ea7..7d505721f05 100644 --- a/docs/source/en/_toctree.yml +++ b/docs/source/en/_toctree.yml @@ -495,6 +495,8 @@ title: Granite - local: model_doc/granitemoe title: GraniteMoe + - local: model_doc/granitemoehybrid + title: GraniteMoeHybrid - local: model_doc/granitemoeshared title: GraniteMoeShared - local: model_doc/helium diff --git a/docs/source/en/model_doc/granitemoehybrid.md b/docs/source/en/model_doc/granitemoehybrid.md new file mode 100644 index 00000000000..49fe0b04ac2 --- /dev/null +++ b/docs/source/en/model_doc/granitemoehybrid.md @@ -0,0 +1,64 @@ + + +# GraniteMoeHybrid + +## Overview + + +The `GraniteMoeHybrid` model builds on top of `GraniteMoeSharedModel` and `Bamba`. Its decoding layers consist of state space layers or MoE attention layers with shared experts. By default, the attention layers do not use positional encoding. + + +```python +from transformers import AutoModelForCausalLM, AutoTokenizer + +model_path = "ibm-granite/granite-4.0-tiny-preview" +tokenizer = AutoTokenizer.from_pretrained(model_path) + +# drop device_map if running on CPU +model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto") +model.eval() + +# change input text as desired +prompt = "Write a code to find the maximum value in a list of numbers." + +# tokenize the text +input_tokens = tokenizer(prompt, return_tensors="pt") +# generate output tokens +output = model.generate(**input_tokens, max_new_tokens=100) +# decode output tokens into text +output = tokenizer.batch_decode(output) +# loop over the batch to print, in this example the batch size is 1 +for i in output: + print(i) +``` + +This HF implementation is contributed by [Sukriti Sharma](https://huggingface.co/SukritiSharma) and [Alexander Brooks](https://huggingface.co/abrooks9944). + + +## GraniteMoeHybridConfig + +[[autodoc]] GraniteMoeHybridConfig + +## GraniteMoeHybridModel + +[[autodoc]] GraniteMoeHybridModel + - forward + +## GraniteMoeHybridForCausalLM + +[[autodoc]] GraniteMoeHybridForCausalLM + - forward \ No newline at end of file diff --git a/src/transformers/models/__init__.py b/src/transformers/models/__init__.py index d3513db4310..1be149faf68 100644 --- a/src/transformers/models/__init__.py +++ b/src/transformers/models/__init__.py @@ -129,6 +129,7 @@ if TYPE_CHECKING: from .granite import * from .granite_speech import * from .granitemoe import * + from .granitemoehybrid import * from .granitemoeshared import * from .grounding_dino import * from .groupvit import * diff --git a/src/transformers/models/auto/configuration_auto.py b/src/transformers/models/auto/configuration_auto.py index f4dd37d64c7..66ed95acad3 100644 --- a/src/transformers/models/auto/configuration_auto.py +++ b/src/transformers/models/auto/configuration_auto.py @@ -146,6 +146,7 @@ CONFIG_MAPPING_NAMES = OrderedDict( ("granite", "GraniteConfig"), ("granite_speech", "GraniteSpeechConfig"), ("granitemoe", "GraniteMoeConfig"), + ("granitemoehybrid", "GraniteMoeHybridConfig"), ("granitemoeshared", "GraniteMoeSharedConfig"), ("granitevision", "LlavaNextConfig"), ("graphormer", "GraphormerConfig"), @@ -509,6 +510,7 @@ MODEL_NAMES_MAPPING = OrderedDict( ("granite", "Granite"), ("granite_speech", "GraniteSpeech"), ("granitemoe", "GraniteMoeMoe"), + ("granitemoehybrid", "GraniteMoeHybrid"), ("granitemoeshared", "GraniteMoeSharedMoe"), ("granitevision", "LLaVA-NeXT"), ("graphormer", "Graphormer"), diff --git a/src/transformers/models/auto/modeling_auto.py b/src/transformers/models/auto/modeling_auto.py index 1dd4a251021..b8ec5c6ccb8 100644 --- a/src/transformers/models/auto/modeling_auto.py +++ b/src/transformers/models/auto/modeling_auto.py @@ -138,6 +138,7 @@ MODEL_MAPPING_NAMES = OrderedDict( ("gptsan-japanese", "GPTSanJapaneseForConditionalGeneration"), ("granite", "GraniteModel"), ("granitemoe", "GraniteMoeModel"), + ("granitemoehybrid", "GraniteMoeHybridModel"), ("granitemoeshared", "GraniteMoeSharedModel"), ("graphormer", "GraphormerModel"), ("grounding-dino", "GroundingDinoModel"), @@ -558,6 +559,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict( ("gptj", "GPTJForCausalLM"), ("granite", "GraniteForCausalLM"), ("granitemoe", "GraniteMoeForCausalLM"), + ("granitemoehybrid", "GraniteMoeHybridForCausalLM"), ("granitemoeshared", "GraniteMoeSharedForCausalLM"), ("helium", "HeliumForCausalLM"), ("jamba", "JambaForCausalLM"), diff --git a/src/transformers/models/bamba/modeling_bamba.py b/src/transformers/models/bamba/modeling_bamba.py index a6220fbf926..7d172b55cc4 100644 --- a/src/transformers/models/bamba/modeling_bamba.py +++ b/src/transformers/models/bamba/modeling_bamba.py @@ -854,6 +854,7 @@ class BambaMixer(nn.Module): # Init cache if ssm_state is not None and cache_params is not None: cache_params.ssm_states[self.layer_idx].copy_(ssm_state) + cache_params.has_previous_state = True scan_output = self.norm(y, gate) diff --git a/src/transformers/models/bamba/modular_bamba.py b/src/transformers/models/bamba/modular_bamba.py index 3ee6e72711e..d6530747aea 100644 --- a/src/transformers/models/bamba/modular_bamba.py +++ b/src/transformers/models/bamba/modular_bamba.py @@ -651,6 +651,7 @@ class BambaMixer(nn.Module): # Init cache if ssm_state is not None and cache_params is not None: cache_params.ssm_states[self.layer_idx].copy_(ssm_state) + cache_params.has_previous_state = True scan_output = self.norm(y, gate) diff --git a/src/transformers/models/granitemoe/configuration_granitemoe.py b/src/transformers/models/granitemoe/configuration_granitemoe.py index 79f3d72cc85..e17860a41d6 100644 --- a/src/transformers/models/granitemoe/configuration_granitemoe.py +++ b/src/transformers/models/granitemoe/configuration_granitemoe.py @@ -166,6 +166,8 @@ class GraniteMoeConfig(PretrainedConfig): self.use_cache = use_cache self.rope_theta = rope_theta self.rope_scaling = rope_scaling + # this model has rope embedding type, hardcoded for BC + self.position_embedding_type = "rope" self.attention_bias = attention_bias self.attention_dropout = attention_dropout diff --git a/src/transformers/models/granitemoe/modeling_granitemoe.py b/src/transformers/models/granitemoe/modeling_granitemoe.py index c7c48c7470d..58d8093f4fa 100644 --- a/src/transformers/models/granitemoe/modeling_granitemoe.py +++ b/src/transformers/models/granitemoe/modeling_granitemoe.py @@ -13,25 +13,24 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -from typing import List, Optional, Tuple, Union +from typing import Callable, List, Optional, Tuple, Union import torch import torch.nn.functional as F -import torch.utils.checkpoint from torch import nn from ...activations import ACT2FN from ...cache_utils import Cache, DynamicCache from ...generation import GenerationMixin from ...modeling_attn_mask_utils import AttentionMaskConverter -from ...modeling_flash_attention_utils import _flash_attention_forward, flash_attn_supports_top_left_mask +from ...modeling_layers import GradientCheckpointingLayer from ...modeling_outputs import ( BaseModelOutputWithPast, MoeCausalLMOutputWithPast, MoeModelOutputWithPast, ) from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update -from ...modeling_utils import PreTrainedModel +from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel from ...pytorch_utils import ALL_LAYERNORM_LAYERS from ...utils import ( add_start_docstrings, @@ -439,10 +438,9 @@ class GraniteMoeAttention(nn.Module): attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_value: Optional[Cache] = None, - output_attentions: bool = False, use_cache: bool = False, cache_position: Optional[torch.LongTensor] = None, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # None or rope embeddings **kwargs, ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: bsz, q_len, _ = hidden_states.size() @@ -455,260 +453,75 @@ class GraniteMoeAttention(nn.Module): key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - cos, sin = position_embeddings - query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) + cos, sin = position_embeddings if position_embeddings is not None else (None, None) + if position_embeddings is not None: + query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) if past_key_value is not None: # sin and cos are specific to RoPE models; cache_position needed for the static cache cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - key_states = repeat_kv(key_states, self.num_key_value_groups) - value_states = repeat_kv(value_states, self.num_key_value_groups) - - attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) * self.scaling - - if attention_mask is not None: # no matter the length, we just slice it - causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] - attn_weights = attn_weights + causal_mask - - # upcast attention to fp32 - attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype) - attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training) - attn_output = torch.matmul(attn_weights, value_states) - - if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim): - raise ValueError( - f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is" - f" {attn_output.size()}" - ) - - attn_output = attn_output.transpose(1, 2).contiguous() - - attn_output = attn_output.view(bsz, q_len, -1) - attn_output = self.o_proj(attn_output) - - if not output_attentions: - attn_weights = None - - return attn_output, attn_weights, past_key_value - - -# NO LONGER EXIST Copied from transformers.models.granite.modeling_granite.GraniteFlashAttention2 with Granite->GraniteMoe -# TODO cyril: modular -class GraniteMoeFlashAttention2(GraniteMoeAttention): - """ - GraniteMoe flash attention module. This module inherits from `GraniteMoeAttention` as the weights of the module stays - untouched. The only required change would be on the forward pass where it needs to correctly call the public API of - flash attention and deal with padding tokens in case the input contains any of them. - """ - - def __init__(self, *args, **kwargs): - super().__init__(*args, **kwargs) - - # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1. - # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignment, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0. - # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left). - self._flash_attn_uses_top_left_mask = flash_attn_supports_top_left_mask() - - def forward( - self, - hidden_states: torch.Tensor, - attention_mask: Optional[torch.LongTensor] = None, - position_ids: Optional[torch.LongTensor] = None, - past_key_value: Optional[Cache] = None, - output_attentions: bool = False, - use_cache: bool = False, - cache_position: Optional[torch.LongTensor] = None, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC - ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: - output_attentions = False - - bsz, q_len, _ = hidden_states.size() - - query_states = self.q_proj(hidden_states) - key_states = self.k_proj(hidden_states) - value_states = self.v_proj(hidden_states) - - # Flash attention requires the input to have the shape - # batch_size x seq_length x head_dim x hidden_dim - # therefore we just need to keep the original shape - query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) - key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - - cos, sin = position_embeddings - query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) - - if past_key_value is not None: - # sin and cos are specific to RoPE models; cache_position needed for the static cache - cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} - key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - - # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache - # to be able to avoid many of these transpose/reshape/view. - query_states = query_states.transpose(1, 2) - key_states = key_states.transpose(1, 2) - value_states = value_states.transpose(1, 2) - - dropout_rate = self.attention_dropout if self.training else 0.0 - - # In PEFT, usually we cast the layer norms in float32 for training stability reasons - # therefore the input hidden states gets silently casted in float32. Hence, we need - # cast them back in the correct dtype just to be sure everything works as expected. - # This might slowdown training & inference so it is recommended to not cast the LayerNorms - # in fp32. (GraniteMoeRMSNorm handles it correctly) - - input_dtype = query_states.dtype - if input_dtype == torch.float32: - if torch.is_autocast_enabled(): - target_dtype = torch.get_autocast_gpu_dtype() - # Handle the case where the model is quantized - elif hasattr(self.config, "_pre_quantization_dtype"): - target_dtype = self.config._pre_quantization_dtype + attention_interface: Callable = eager_attention_forward + if self.config._attn_implementation != "eager": + if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False): + logger.warning_once( + "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to " + 'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' + ) else: - target_dtype = self.q_proj.weight.dtype + attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation] - logger.warning_once( - f"The input hidden states seems to be silently casted in float32, this might be related to" - f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in" - f" {target_dtype}." - ) - - query_states = query_states.to(target_dtype) - key_states = key_states.to(target_dtype) - value_states = value_states.to(target_dtype) - - attn_output = _flash_attention_forward( + attn_output, attn_weights = attention_interface( + self, query_states, key_states, value_states, attention_mask, - q_len, - position_ids=position_ids, - dropout=dropout_rate, - softmax_scale=self.scaling, - sliding_window=getattr(self, "sliding_window", None), - use_top_left_mask=self._flash_attn_uses_top_left_mask, - is_causal=self.is_causal, + dropout=0.0 if not self.training else self.attention_dropout, + scaling=self.scaling, + **kwargs, ) - attn_output = attn_output.reshape(bsz, q_len, -1).contiguous() + attn_output = attn_output.view(bsz, q_len, -1) attn_output = self.o_proj(attn_output) - if not output_attentions: - attn_weights = None - return attn_output, attn_weights, past_key_value -# NO LONGER EXIST Copied from transformers.models.granite.modeling_granite.GraniteSdpaAttention with Granite->GraniteMoe -# TODO cyril: modular -class GraniteMoeSdpaAttention(GraniteMoeAttention): - """ - GraniteMoe attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from - `GraniteMoeAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to - SDPA API. - """ +def eager_attention_forward( + module: nn.Module, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + attention_mask: Optional[torch.Tensor], + scaling: float, + dropout: float = 0.0, + **kwargs, +): + key_states = repeat_kv(key, module.num_key_value_groups) + value_states = repeat_kv(value, module.num_key_value_groups) - # Adapted from GraniteMoeAttention.forward - def forward( - self, - hidden_states: torch.Tensor, - attention_mask: Optional[torch.Tensor] = None, - position_ids: Optional[torch.LongTensor] = None, - past_key_value: Optional[Cache] = None, - output_attentions: bool = False, - use_cache: bool = False, - cache_position: Optional[torch.LongTensor] = None, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC - **kwargs, - ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: - if output_attentions: - # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented. - logger.warning_once( - "GraniteMoeModel is using GraniteMoeSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, " - 'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' - ) - return super().forward( - hidden_states=hidden_states, - attention_mask=attention_mask, - position_ids=position_ids, - past_key_value=past_key_value, - output_attentions=output_attentions, - use_cache=use_cache, - cache_position=cache_position, - position_embeddings=position_embeddings, - ) + attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling + if attention_mask is not None: + causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] + attn_weights = attn_weights + causal_mask - bsz, q_len, _ = hidden_states.size() + # upcast attention to fp32 + attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype) + attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training) + attn_output = torch.matmul(attn_weights, value_states) + attn_output = attn_output.transpose(1, 2).contiguous() - query_states = self.q_proj(hidden_states) - key_states = self.k_proj(hidden_states) - value_states = self.v_proj(hidden_states) - - query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) - key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - - cos, sin = position_embeddings - query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) - - if past_key_value is not None: - # sin and cos are specific to RoPE models; cache_position needed for the static cache - cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} - key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - - key_states = repeat_kv(key_states, self.num_key_value_groups) - value_states = repeat_kv(value_states, self.num_key_value_groups) - - causal_mask = attention_mask - if attention_mask is not None: - causal_mask = causal_mask[:, :, :, : key_states.shape[-2]] - - # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask, - # Reference: https://github.com/pytorch/pytorch/issues/112577. - if query_states.device.type == "cuda" and causal_mask is not None: - query_states = query_states.contiguous() - key_states = key_states.contiguous() - value_states = value_states.contiguous() - - # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment - # in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling. - is_causal = True if causal_mask is None and q_len > 1 else False - - attn_output = torch.nn.functional.scaled_dot_product_attention( - query_states, - key_states, - value_states, - attn_mask=causal_mask, - dropout_p=self.attention_dropout if self.training else 0.0, - is_causal=is_causal, - scale=self.scaling, - ) - - attn_output = attn_output.transpose(1, 2).contiguous() - attn_output = attn_output.view(bsz, q_len, -1) - - attn_output = self.o_proj(attn_output) - - return attn_output, None, past_key_value + return attn_output, attn_weights -GRANITEMOE_ATTENTION_CLASSES = { - "eager": GraniteMoeAttention, - "flash_attention_2": GraniteMoeFlashAttention2, - "sdpa": GraniteMoeSdpaAttention, -} - - -class GraniteMoeDecoderLayer(nn.Module): +class GraniteMoeDecoderLayer(GradientCheckpointingLayer): def __init__(self, config: GraniteMoeConfig, layer_idx: int): super().__init__() self.hidden_size = config.hidden_size - self.self_attn = GRANITEMOE_ATTENTION_CLASSES[config._attn_implementation](config=config, layer_idx=layer_idx) - + self.self_attn = GraniteMoeAttention(config=config, layer_idx=layer_idx) self.block_sparse_moe = GraniteMoeMoE(config) self.input_layernorm = GraniteMoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.post_attention_layernorm = GraniteMoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps) @@ -827,13 +640,12 @@ class GraniteMoePreTrainedModel(PreTrainedModel): _supports_static_cache = False # MoE models don't work with torch.compile (`torch.where(condition)` not supported) def _init_weights(self, module): - std = self.config.initializer_range if isinstance(module, nn.Linear): - module.weight.data.normal_(mean=0.0, std=std) + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) if module.bias is not None: module.bias.data.zero_() elif isinstance(module, nn.Embedding): - module.weight.data.normal_(mean=0.0, std=std) + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) if module.padding_idx is not None: module.weight.data[module.padding_idx].zero_() elif isinstance(module, GraniteMoeRMSNorm): @@ -947,8 +759,8 @@ class GraniteMoeModel(GraniteMoePreTrainedModel): self.max_position_embeddings = config.max_position_embeddings self.rope_theta = config.rope_theta - # rope - self.rotary_emb = GraniteMoeRotaryEmbedding(config) + self.position_embedding_type = config.position_embedding_type + self.rotary_emb = GraniteMoeRotaryEmbedding(config) if self.position_embedding_type == "rope" else None # Initialize weights and apply final processing self.post_init() @@ -1019,8 +831,10 @@ class GraniteMoeModel(GraniteMoePreTrainedModel): # embed positions hidden_states = inputs_embeds + position_embeddings = None # create position embeddings to be shared across the decoder layers - position_embeddings = self.rotary_emb(hidden_states, position_ids) + if self.rotary_emb is not None: + position_embeddings = self.rotary_emb(hidden_states, position_ids) # decoder layers all_hidden_states = () if output_hidden_states else None @@ -1032,31 +846,17 @@ class GraniteMoeModel(GraniteMoePreTrainedModel): 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, - output_router_logits, - position_embeddings, - ) - 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, - output_router_logits=output_router_logits, - position_embeddings=position_embeddings, - ) + 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, + output_router_logits=output_router_logits, + position_embeddings=position_embeddings, + ) hidden_states = layer_outputs[0] @@ -1265,6 +1065,7 @@ class GraniteMoeForCausalLM(GraniteMoePreTrainedModel, GenerationMixin): output_router_logits: Optional[bool] = None, return_dict: Optional[bool] = None, cache_position: Optional[torch.LongTensor] = None, + logits_to_keep: Union[int, torch.Tensor] = 0, **kwargs, ) -> Union[Tuple, MoeCausalLMOutputWithPast]: r""" @@ -1273,6 +1074,13 @@ class GraniteMoeForCausalLM(GraniteMoePreTrainedModel, GenerationMixin): 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]`. + logits_to_keep (`int` or `torch.Tensor`, *optional*): + 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: Example: @@ -1315,8 +1123,10 @@ class GraniteMoeForCausalLM(GraniteMoePreTrainedModel, GenerationMixin): cache_position=cache_position, ) + # Only compute necessary logits hidden_states = outputs[0] - logits = self.lm_head(hidden_states) + slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep + logits = self.lm_head(hidden_states[:, slice_indices, :]) logits = logits / self.config.logits_scaling loss = None diff --git a/src/transformers/models/granitemoehybrid/__init__.py b/src/transformers/models/granitemoehybrid/__init__.py new file mode 100644 index 00000000000..7b4db962865 --- /dev/null +++ b/src/transformers/models/granitemoehybrid/__init__.py @@ -0,0 +1,29 @@ +# coding=utf-8 +# Copyright 2025 IBM and the HuggingFace Inc. team. All rights reserved. +# +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from typing import TYPE_CHECKING + +from ...utils import _LazyModule +from ...utils.import_utils import define_import_structure + + +if TYPE_CHECKING: + from .configuration_granitemoehybrid import * + from .modeling_granitemoehybrid import * +else: + import sys + + _file = globals()["__file__"] + sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__) diff --git a/src/transformers/models/granitemoehybrid/configuration_granitemoehybrid.py b/src/transformers/models/granitemoehybrid/configuration_granitemoehybrid.py new file mode 100644 index 00000000000..4a3b1da88a8 --- /dev/null +++ b/src/transformers/models/granitemoehybrid/configuration_granitemoehybrid.py @@ -0,0 +1,256 @@ +# coding=utf-8 +# Copyright 2025 IBM and the HuggingFace Inc. team. All rights reserved. +# +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +"""GraniteMoeHybrid model configuration""" + +from ...configuration_utils import PretrainedConfig +from ...modeling_rope_utils import rope_config_validation +from ...utils import logging + + +logger = logging.get_logger(__name__) + + +class GraniteMoeHybridConfig(PretrainedConfig): + r""" + This is the configuration class to store the configuration of a [`GraniteMoeHybridConfig`]. It is used to + instantiate an GraniteMoeHybrid 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 GraniteMoeHybrid model. Defines the number of different tokens that + can be represented by the `inputs_ids` passed when calling [`GraniteMoeHybridModel`] + 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 checkout [this + paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to + `num_attention_heads`. + hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): + The non-linear activation function (function or string) in the decoder. + 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`. + pad_token_id (`int`, *optional*): + Padding token id. + bos_token_id (`int`, *optional*, defaults to 1): + Beginning of stream token id. + eos_token_id (`int`, *optional*, defaults to 2): + End of stream token id. + tie_word_embeddings (`bool`, *optional*, defaults to `False`): + Whether to tie weight embeddings + rope_theta (`float`, *optional*, defaults to 10000.0): + The base period of the RoPE embeddings. + rope_scaling (`Dict`, *optional*): + Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling + strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is + `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update + `max_position_embeddings` to the expected new maximum. See the following thread for more information on how + these scaling strategies behave: + https://www.reddit.com/r/LocalLLaMA/comments/14mrgpr/dynamically_scaled_rope_further_increases/. This is an + experimental feature, subject to breaking API changes in future versions. + 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 (`float`, *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): total number of experts. + num_experts_per_tok (`int`, *optional*, defaults to 2): number of experts per token. + 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. + router_aux_loss_coef (`float`, *optional*, defaults to 0.001): router auxialiary loss coefficient + 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`, *optional*): list of strings to be used as layer types. + Allowed choices: "mamba", "attention". + mamba_n_heads (`int`, *optional*, defaults to 128): + The number of mamba heads used. + mamba_n_groups (`int`, *optional*, defaults to 1): + The number of the mamba groups used. + mamba_d_state (`int`, *optional*, defaults to 256): + The dimension the mamba latent state space. + mamba_d_head (`int`, *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. + ```python + >>> from transformers import GraniteMoeHybridModel, GraniteMoeHybridConfig + + >>> # Initializing a GraniteMoeHybrid config + >>> configuration = GraniteMoeHybridConfig() + + + >>> # Accessing the model configuration + >>> configuration = model.config + ```""" + + model_type = "granitemoehybrid" + attribute_map = { + "layers_block_type": "layer_types", + } + keys_to_ignore_at_inference = ["past_key_values"] + + def __init__( + self, + vocab_size=32000, + hidden_size=4096, + intermediate_size=11008, + num_hidden_layers=32, + num_attention_heads=32, + num_key_value_heads=None, + hidden_act="silu", + max_position_embeddings=2048, + initializer_range=0.02, + rms_norm_eps=1e-6, + use_cache=True, + pad_token_id=None, + bos_token_id=1, + eos_token_id=2, + tie_word_embeddings=False, + rope_theta=10000.0, + rope_scaling=None, + attention_bias=False, + attention_dropout=0.0, + embedding_multiplier=1.0, + logits_scaling=1.0, + residual_multiplier=1.0, + attention_multiplier=1.0, + num_local_experts=8, + num_experts_per_tok=2, + output_router_logits=False, + router_aux_loss_coef=0.001, + shared_intermediate_size=1024, + position_embedding_type=None, + layer_types=None, + mamba_n_heads=128, + mamba_n_groups=1, + mamba_d_state=256, + mamba_d_head="auto", + mamba_d_conv=4, + mamba_expand=2, + mamba_chunk_size=256, + mamba_conv_bias=True, + mamba_proj_bias=False, + **kwargs, + ): + self.vocab_size = vocab_size + self.max_position_embeddings = max_position_embeddings + self.hidden_size = hidden_size + self.intermediate_size = intermediate_size + self.num_hidden_layers = num_hidden_layers + self.num_attention_heads = num_attention_heads + + # for backward compatibility + if num_key_value_heads is None: + num_key_value_heads = num_attention_heads + + self.num_key_value_heads = num_key_value_heads + self.hidden_act = hidden_act + self.initializer_range = initializer_range + self.rms_norm_eps = rms_norm_eps + self.use_cache = use_cache + self.rope_theta = rope_theta + self.rope_scaling = rope_scaling + self.attention_bias = attention_bias + self.embedding_multiplier = embedding_multiplier + self.logits_scaling = logits_scaling + self.residual_multiplier = residual_multiplier + self.attention_multiplier = attention_multiplier + self.attention_dropout = attention_dropout + self.num_local_experts = num_local_experts + self.num_experts_per_tok = num_experts_per_tok + self.output_router_logits = output_router_logits + self.router_aux_loss_coef = router_aux_loss_coef + self.shared_intermediate_size = shared_intermediate_size + self.position_embedding_type = position_embedding_type + + mamba_intermediate = mamba_expand * hidden_size + + if layer_types is not None and any(layer_type not in ["mamba", "attention"] for layer_type in layer_types): + raise ValueError("layer_types must be a list strings in [`mamba` `attention`]") + + if mamba_intermediate % mamba_n_heads != 0: + raise ValueError("mamba_n_heads must divide mamba_expand * hidden_size") + + # for the mamba_v2, must satisfy the following + if mamba_d_head == "auto": + mamba_d_head = mamba_intermediate // mamba_n_heads + + if mamba_d_head * mamba_n_heads != mamba_intermediate: + raise ValueError("The dimensions for the Mamba head state do not match the model intermediate_size") + + self.mamba_n_heads = mamba_n_heads + self.mamba_d_head = mamba_d_head + self.mamba_n_groups = mamba_n_groups + self.mamba_d_state = mamba_d_state + self.mamba_d_conv = mamba_d_conv + self.mamba_chunk_size = mamba_chunk_size + self.mamba_conv_bias = mamba_conv_bias + self.mamba_proj_bias = mamba_proj_bias + self.mamba_expand = mamba_expand + self.layer_types = layer_types + + super().__init__( + pad_token_id=pad_token_id, + bos_token_id=bos_token_id, + eos_token_id=eos_token_id, + tie_word_embeddings=tie_word_embeddings, + **kwargs, + ) + + if self.position_embedding_type == "rope": + rope_config_validation(self) + + # overwrite the function to use in `HybridMambaAttentionDynamicCache` + @property + def layers_block_type(self): + return self.layer_types if self.layer_types else ["mamba"] * self.num_hidden_layers + + +__all__ = ["GraniteMoeHybridConfig"] diff --git a/src/transformers/models/granitemoehybrid/modeling_granitemoehybrid.py b/src/transformers/models/granitemoehybrid/modeling_granitemoehybrid.py new file mode 100644 index 00000000000..016b2af2fa0 --- /dev/null +++ b/src/transformers/models/granitemoehybrid/modeling_granitemoehybrid.py @@ -0,0 +1,1939 @@ +# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 +# This file was automatically generated from src/transformers/models/granitemoehybrid/modular_granitemoehybrid.py. +# Do NOT edit this file manually as any edits will be overwritten by the generation of +# the file from the modular. If any change should be done, please apply the change to the +# modular_granitemoehybrid.py file directly. One of our CI enforces this. +# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 +# coding=utf-8 +# Copyright 2025 IBM and the HuggingFace Inc. team. All rights reserved. +# +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from typing import Callable, List, Optional, Tuple, Union + +import torch +import torch.nn.functional as F +from torch import nn + +import transformers.models.jamba.modeling_jamba as modeling_jamba +from transformers.activations import ACT2FN + +from ...cache_utils import Cache, StaticCache +from ...generation import GenerationMixin +from ...modeling_attn_mask_utils import AttentionMaskConverter +from ...modeling_layers import GradientCheckpointingLayer +from ...modeling_outputs import BaseModelOutputWithPast, MoeCausalLMOutputWithPast, MoeModelOutputWithPast +from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update +from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel +from ...utils import ( + add_start_docstrings, + add_start_docstrings_to_model_forward, + can_return_tuple, + is_torch_flex_attn_available, + logging, + replace_return_docstrings, +) +from ...utils.import_utils import is_causal_conv1d_available, is_mamba_2_ssm_available +from .configuration_granitemoehybrid import GraniteMoeHybridConfig + + +if is_mamba_2_ssm_available(): + from mamba_ssm.ops.triton.selective_state_update import selective_state_update + from mamba_ssm.ops.triton.ssd_combined import mamba_chunk_scan_combined, mamba_split_conv1d_scan_combined +else: + selective_state_update = None + +if is_causal_conv1d_available(): + from causal_conv1d import causal_conv1d_fn, causal_conv1d_update +else: + causal_conv1d_update, causal_conv1d_fn = None, None + + +if is_torch_flex_attn_available(): + from torch.nn.attention.flex_attention import BlockMask + + from ...integrations.flex_attention import make_flex_block_causal_mask + + +logger = logging.get_logger(__name__) + + +_CONFIG_FOR_DOC = "GraniteMoeHybridConfig" + + +def rotate_half(x): + """Rotates half the hidden dims of the input.""" + x1 = x[..., : x.shape[-1] // 2] + x2 = x[..., x.shape[-1] // 2 :] + return torch.cat((-x2, x1), dim=-1) + + +def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1): + """Applies Rotary Position Embedding to the query and key tensors. + + Args: + q (`torch.Tensor`): The query tensor. + k (`torch.Tensor`): The key tensor. + cos (`torch.Tensor`): The cosine part of the rotary embedding. + sin (`torch.Tensor`): The sine part of the rotary embedding. + position_ids (`torch.Tensor`, *optional*): + Deprecated and unused. + unsqueeze_dim (`int`, *optional*, defaults to 1): + The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and + sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note + that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and + k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes + cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have + the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. + Returns: + `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. + """ + cos = cos.unsqueeze(unsqueeze_dim) + sin = sin.unsqueeze(unsqueeze_dim) + q_embed = (q * cos) + (rotate_half(q) * sin) + k_embed = (k * cos) + (rotate_half(k) * sin) + return q_embed, k_embed + + +def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor: + """ + This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, + num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim) + """ + batch, num_key_value_heads, slen, head_dim = hidden_states.shape + if n_rep == 1: + return hidden_states + hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim) + return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim) + + +def eager_attention_forward( + module: nn.Module, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + attention_mask: Optional[torch.Tensor], + scaling: float, + dropout: float = 0.0, + **kwargs, +): + key_states = repeat_kv(key, module.num_key_value_groups) + value_states = repeat_kv(value, module.num_key_value_groups) + + attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling + if attention_mask is not None: + causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] + attn_weights = attn_weights + causal_mask + + # upcast attention to fp32 + attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype) + attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training) + attn_output = torch.matmul(attn_weights, value_states) + attn_output = attn_output.transpose(1, 2).contiguous() + + return attn_output, attn_weights + + +# copied from transformers.models.granite.modeling_granite.GraniteAttention with Granite->GraniteMoeHybrid +# no longer copied after attention refactors +class GraniteMoeHybridAttention(nn.Module): + """Multi-headed attention from 'Attention Is All You Need' paper""" + + def __init__(self, config: GraniteMoeHybridConfig, layer_idx: int): + super().__init__() + self.config = config + self.layer_idx = layer_idx + if layer_idx is None: + logger.warning_once( + f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will " + "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` " + "when creating this class." + ) + + self.attention_dropout = config.attention_dropout + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.num_key_value_heads = config.num_key_value_heads + self.num_key_value_groups = self.num_heads // self.num_key_value_heads + self.is_causal = True + + self.scaling = config.attention_multiplier + + if (self.head_dim * self.num_heads) != self.hidden_size: + raise ValueError( + f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}" + f" and `num_heads`: {self.num_heads})." + ) + + self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.attention_bias) + self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias) + self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=config.attention_bias) + + def forward( + self, + hidden_states: torch.Tensor, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_value: Optional[Cache] = None, + use_cache: bool = False, + cache_position: Optional[torch.LongTensor] = None, + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # None or rope embeddings + **kwargs, + ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: + bsz, q_len, _ = hidden_states.size() + + query_states = self.q_proj(hidden_states) + key_states = self.k_proj(hidden_states) + value_states = self.v_proj(hidden_states) + + query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) + key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) + value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) + + cos, sin = position_embeddings if position_embeddings is not None else (None, None) + if position_embeddings is not None: + query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) + + if past_key_value is not None: + # sin and cos are specific to RoPE models; cache_position needed for the static cache + cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} + key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) + + attention_interface: Callable = eager_attention_forward + if self.config._attn_implementation != "eager": + if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False): + logger.warning_once( + "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to " + 'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' + ) + else: + attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation] + + attn_output, attn_weights = attention_interface( + self, + query_states, + key_states, + value_states, + attention_mask, + dropout=0.0 if not self.training else self.attention_dropout, + scaling=self.scaling, + **kwargs, + ) + + attn_output = attn_output.view(bsz, q_len, -1) + attn_output = self.o_proj(attn_output) + + return attn_output, attn_weights, past_key_value + + +# Adapted from transformers.models.jamba.modeling_jamba.HybridMambaAttentionDynamicCache for the v2 mixer +class HybridMambaAttentionDynamicCache(modeling_jamba.HybridMambaAttentionDynamicCache): + """ + A dynamic cache that can handle both the attention cache (which has a seq_len dimension) and the mamba cache + (which has a constant shape regardless of seq_len). + + This cache has two sets of lists of tensors: `key_cache` and `value_cache` for attention cache and `conv_states` + and `ssm_states` for mamba cache. Each of these lists has `num_layers` tensors. The expected shape for each tensor + For attention layers, `key_cache` and `value_cache` have a shape of `(batch_size, num_heads, seq_len, head_dim)`, + while `conv_states` and `ssm_states` have a shape of `(batch_size, 0)` (empty tensors). + For mamba layers, `key_cache` and `value_cache` have a shape of `(batch_size, 0)` (empty tensors), + while `conv_states` represents the convolution state and has a shape of `(batch_size, d_inner, d_conv)`, + and `ssm_states` represents the ssm state and has a shape of `(batch_size, d_inner, d_state)`. + """ + + def __init__(self, config: GraniteMoeHybridConfig, batch_size, dtype=torch.float16, device=None): + super().__init__(config, batch_size, dtype, device) + self.layers_block_type = config.layers_block_type + self.has_previous_state = False # only used by mamba + conv_kernel_size = config.mamba_d_conv + ssm_state_size = config.mamba_d_state + + self.conv_states = [] + self.ssm_states = [] + self.transformer_layers = [] + for i in range(config.num_hidden_layers): + if self.layers_block_type[i] == "mamba": + self.conv_states += [ + torch.zeros( + batch_size, + (config.mamba_expand * config.hidden_size + 2 * config.mamba_n_groups * ssm_state_size), + conv_kernel_size, + device=device, + dtype=dtype, + ) + ] + self.ssm_states += [ + torch.zeros( + batch_size, + config.mamba_n_heads, + config.mamba_d_head, + ssm_state_size, + device=device, + dtype=dtype, + ) + ] + else: + self.conv_states += [torch.tensor([[]] * batch_size, device=device)] + self.ssm_states += [torch.tensor([[]] * batch_size, device=device)] + self.transformer_layers.append(i) + + self.key_cache = [torch.tensor([[]] * batch_size, device=device) for _ in range(config.num_hidden_layers)] + self.value_cache = [torch.tensor([[]] * batch_size, device=device) for _ in range(config.num_hidden_layers)] + + +# Helper methods for segment sum computation + + +def pad_tensor_by_size(input_tensor: torch.Tensor, pad_size: int): + """ + Padding x tensor with `pad_size` on the seq_len dim (dim=1) + + Assumes that we only have tensors of either size 4 or 3 + """ + pad_shape = (0, 0, 0, 0, 0, pad_size, 0, 0) if len(input_tensor.shape) == 4 else (0, 0, 0, pad_size, 0, 0) + + return torch.nn.functional.pad(input_tensor, pad_shape, mode="constant", value=0) + + +def reshape_into_chunks(input_tensor, pad_size, chunk_size): + """ + Padding input_tensor with `pad_size` on the seq_len dim (dim=1) and + simultaneously splitting it into chunk sequences. + + Assumes that we only have tensors of either size 4 or 3 + """ + # [bsz, seq_len, ...] -> [bsz, seq_len multiple of chunk_size, ...] + input_tensor = pad_tensor_by_size(input_tensor, pad_size) + + if len(input_tensor.shape) == 3: + # [bsz, seq_len multiple of chunk_size, num_heads] -> [bsz, -1, chunk_size, num_heads] + return input_tensor.reshape(input_tensor.shape[0], -1, chunk_size, input_tensor.shape[2]) + else: + # [bsz, seq_len multiple of chunk_size, num_heads, head_dim or state_size] -> [bsz, -1, chunk_size, num_heads, head_dim or state_size] + return input_tensor.reshape( + input_tensor.shape[0], -1, chunk_size, input_tensor.shape[2], input_tensor.shape[3] + ) + + +def segment_sum(input_tensor): + """ + More stable segment sum calculation. Uses cumulative sums and masking instead of direct subtractions. + """ + chunk_size = input_tensor.size(-1) + # 1. expand input tensor to have an additional dimension and repeat along that dimension + # [..., chunk_size] -> [..., chunk_size, chunk_size] + input_tensor = input_tensor[..., None].expand(*input_tensor.size(), chunk_size) + # 2. create a lower triangular mask with the diagonal set to 0 to 0 out elements above diag + mask = torch.tril(torch.ones(chunk_size, chunk_size, device=input_tensor.device, dtype=torch.bool), diagonal=-1) + input_tensor = input_tensor.masked_fill(~mask, 0) + # 3. compute actual cumsum + tensor_segsum = torch.cumsum(input_tensor, dim=-2) + + # 4. apply mask to keep only the lower triangular part of the cumulative sum result (incl diagonal this time) + mask = torch.tril(torch.ones(chunk_size, chunk_size, device=input_tensor.device, dtype=torch.bool), diagonal=0) + tensor_segsum = tensor_segsum.masked_fill(~mask, -torch.inf) + return tensor_segsum + + +is_fast_path_available = all((selective_state_update, causal_conv1d_fn, causal_conv1d_update)) + + +def apply_mask_to_padding_states(hidden_states, attention_mask): + """ + Tunes out the hidden states for padding tokens, see https://github.com/state-spaces/mamba/issues/66 + """ + if attention_mask is not None and attention_mask.shape[1] > 1 and attention_mask.shape[0] > 1: + dtype = hidden_states.dtype + hidden_states = (hidden_states * attention_mask[:, :, None]).to(dtype) + + return hidden_states + + +# Adapted from transformers.models.mamba2.modeling_mamba2.Mamba2Mixer +class GraniteMoeHybridMambaLayer(nn.Module): + """ + Compute ∆, A, B, C, and D the state space parameters and compute the `contextualized_states`. + A, D are input independent (see Mamba paper [1] Section 3.5.2 "Interpretation of A" for why A isn't selective) + ∆, B, C are input-dependent (this is a key difference between Mamba and the linear time invariant S4, + and is why Mamba is called **selective** state spaces) + + The are a few differences between this and Mamba2Mixer: + - The variable use_precomputed_states is slightly different due to the HybridCache structure + - There's a few non-obvious bugs fixed with batching in the slow path that exist in main + - Some extra variables that our layer doesn't need have been removed + - We ported most of the refactors in https://github.com/huggingface/transformers/pull/35154, which is (as of Dec 18, 2024) unmerged + """ + + def __init__(self, config: GraniteMoeHybridConfig, layer_idx: int): + super().__init__() + self.num_heads = config.mamba_n_heads + self.hidden_size = config.hidden_size + self.ssm_state_size = config.mamba_d_state + self.conv_kernel_size = config.mamba_d_conv + self.intermediate_size = int(config.mamba_expand * self.hidden_size) + self.layer_idx = layer_idx + self.use_conv_bias = config.mamba_conv_bias + self.activation = config.hidden_act + self.act = ACT2FN[config.hidden_act] + self.use_bias = config.mamba_proj_bias + + self.layer_norm_epsilon = config.rms_norm_eps + + self.n_groups = config.mamba_n_groups + self.head_dim = config.mamba_d_head + self.chunk_size = config.mamba_chunk_size + + # FIXME: + self.time_step_limit = (0.0, float("inf")) + self.time_step_min = 0.001 + self.time_step_max = 0.1 + + self.conv_dim = self.intermediate_size + 2 * self.n_groups * self.ssm_state_size + self.conv1d = nn.Conv1d( + in_channels=self.conv_dim, + out_channels=self.conv_dim, + bias=config.mamba_conv_bias, + kernel_size=self.conv_kernel_size, + groups=self.conv_dim, + padding=self.conv_kernel_size - 1, + ) + + # projection of the input hidden states + projection_size = self.intermediate_size + self.conv_dim + self.num_heads + self.in_proj = nn.Linear( + self.hidden_size, + projection_size, + bias=self.use_bias, + ) + # selective projection used to make dt, B and C input dependent + + # time step projection (discretization) + # instantiate once and copy inv_dt in init_weights of PretrainedModel + self.dt_bias = nn.Parameter(torch.ones(self.num_heads)) + + # S4D real initialization. These are not discretized! + # The core is to load them, compute the discrete states, then write the updated state. Keeps the memory bounded + A = torch.arange(1, self.num_heads + 1) + self.A_log = nn.Parameter(torch.log(A)) + self.A_log._no_weight_decay = True + self.norm = GraniteMoeHybridRMSNormGated(self.intermediate_size, eps=self.layer_norm_epsilon) + self.D = nn.Parameter(torch.ones(self.num_heads)) + self.D._no_weight_decay = True + + self.out_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=self.use_bias) + + if not is_fast_path_available: + logger.warning_once( + "The fast path is not available because on of `(selective_state_update, causal_conv1d_fn, causal_conv1d_update)`" + " is None. Falling back to the naive implementation. To install follow https://github.com/state-spaces/mamba/#installation and" + " https://github.com/Dao-AILab/causal-conv1d" + ) + else: + logger.warning_once("The fast path for GraniteMoeHybrid will be used when running the model on a GPU") + + def cuda_kernels_forward( + self, + hidden_states: torch.Tensor, + cache_params: Optional[HybridMambaAttentionDynamicCache] = None, + cache_position: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + ): + # 1. Gated MLP's linear projection + hidden_states = apply_mask_to_padding_states(hidden_states, attention_mask) + projected_states = self.in_proj(hidden_states) + + # Set up dimensions for reshapes later + batch_size, seq_len, _ = hidden_states.shape + groups_time_state_size = self.n_groups * self.ssm_state_size + + use_precomputed_states = ( + cache_params is not None + and cache_params.has_previous_state + and seq_len == 1 + and cache_params.conv_states[self.layer_idx].shape[0] + == cache_params.ssm_states[self.layer_idx].shape[0] + == batch_size + and cache_position is not None + and cache_position[0] > 0 + ) + + # getting projected states from cache if it exists + if use_precomputed_states: + gate, hidden_states_B_C, dt = projected_states.squeeze(1).split( + [self.intermediate_size, self.conv_dim, self.num_heads], dim=-1 + ) + + # 2. Convolution sequence transformation + hidden_states_B_C = causal_conv1d_update( + hidden_states_B_C, + cache_params.conv_states[self.layer_idx], + self.conv1d.weight.squeeze(1), + self.conv1d.bias, + self.activation, + ) + + hidden_states, B, C = torch.split( + hidden_states_B_C, + [self.intermediate_size, groups_time_state_size, groups_time_state_size], + dim=-1, + ) + + # 3. SSM transformation + A = -torch.exp(self.A_log.float()) # (nheads,) + A = A[:, None, ...][:, :, None].expand(-1, self.head_dim, self.ssm_state_size).to(dtype=torch.float32) + dt = dt[:, :, None].expand(-1, -1, self.head_dim) + dt_bias = self.dt_bias[:, None, ...].expand(-1, self.head_dim) + D = self.D[:, None, ...].expand(-1, self.head_dim) + B = B.view(batch_size, self.n_groups, B.shape[1] // self.n_groups) + C = C.view(batch_size, self.n_groups, C.shape[1] // self.n_groups) + hidden_states_reshaped = hidden_states.view(batch_size, self.num_heads, self.head_dim) + hidden_states = selective_state_update( + cache_params.ssm_states[self.layer_idx], + hidden_states_reshaped, + dt, + A, + B, + C, + D, + z=None, + dt_bias=dt_bias, + dt_softplus=True, + ) + hidden_states = hidden_states.view(batch_size, self.num_heads * self.head_dim) + hidden_states = self.norm(hidden_states, gate) + + # 4. Final linear projection + out = self.out_proj(hidden_states)[:, None, ...] + # Fused calculations or step by step if no initialized cache is found + else: + A = -torch.exp(self.A_log.float()) # (num_heads) or (intermediate_size, state_size) + dt_limit_kwargs = {} if self.time_step_limit == (0.0, float("inf")) else {"dt_limit": self.time_step_limit} + + # 2-4. Fused kernel for conv1d, SSM, and the final projection + if self.training and cache_params is None: + out = mamba_split_conv1d_scan_combined( + projected_states, + self.conv1d.weight.squeeze(1), + self.conv1d.bias, + self.dt_bias, + A, + D=self.D, + chunk_size=self.chunk_size, + seq_idx=None, # was seq_idx + activation=self.activation, + rmsnorm_weight=self.norm.weight, + rmsnorm_eps=self.norm.variance_epsilon, + outproj_weight=self.out_proj.weight, + outproj_bias=self.out_proj.bias, + headdim=self.head_dim, + ngroups=self.n_groups, + norm_before_gate=False, + return_final_states=False, + **dt_limit_kwargs, + ) + + else: + gate, hidden_states_B_C, dt = projected_states.split( + [self.intermediate_size, self.conv_dim, self.num_heads], dim=-1 + ) + + # 2. Convolution sequence transformation + # Init cache + if cache_params is not None: + # storing the states + # If we just take xBC[:, :, -self.d_conv :], it will error if seqlen < self.d_conv + # Instead F.pad will pad with zeros if seqlen < self.d_conv, and truncate otherwise. + hidden_states_B_C_transposed = hidden_states_B_C.transpose(1, 2) + conv_states = nn.functional.pad( + hidden_states_B_C_transposed, + (self.conv_kernel_size - hidden_states_B_C_transposed.shape[-1], 0), + ) + cache_params.conv_states[self.layer_idx].copy_(conv_states) + + if self.activation not in ["silu", "swish"]: + hidden_states_B_C = self.act( + self.conv1d(hidden_states_B_C.transpose(1, 2))[..., :seq_len].transpose(1, 2) + ) + else: + hidden_states_B_C = causal_conv1d_fn( + x=hidden_states_B_C.transpose(1, 2), + weight=self.conv1d.weight.squeeze(1), + bias=self.conv1d.bias, + activation=self.activation, + ).transpose(1, 2) + + hidden_states_B_C = apply_mask_to_padding_states(hidden_states_B_C, attention_mask) + hidden_states, B, C = torch.split( + hidden_states_B_C, + [self.intermediate_size, groups_time_state_size, groups_time_state_size], + dim=-1, + ) + + # 3. SSM transformation + scan_output, ssm_state = mamba_chunk_scan_combined( + hidden_states.view(batch_size, seq_len, -1, self.head_dim), + dt, + A, + B.view(batch_size, seq_len, self.n_groups, -1), + C.view(batch_size, seq_len, self.n_groups, -1), + chunk_size=self.chunk_size, + D=self.D, + z=None, + seq_idx=None, + return_final_states=True, + dt_bias=self.dt_bias, + dt_softplus=True, + **dt_limit_kwargs, + ) + + # Init cache + if ssm_state is not None and cache_params is not None: + cache_params.ssm_states[self.layer_idx].copy_(ssm_state) + + scan_output = scan_output.view(batch_size, seq_len, -1) + # Multiply "gate" branch and apply extra normalization layer + scan_output = self.norm(scan_output, gate) + + # 4. Final linear projection + out = self.out_proj(scan_output) + return out + + # fmt: off + def torch_forward( + self, + input_states, + cache_params: Optional[HybridMambaAttentionDynamicCache] = None, + cache_position: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + ): + batch_size, seq_len, _ = input_states.shape + dtype = input_states.dtype + + # 1. Gated MLP's linear projection + input_states = apply_mask_to_padding_states(input_states, attention_mask) + projected_states = self.in_proj(input_states) + gate, hidden_states_B_C, dt = projected_states.split( + [self.intermediate_size, self.conv_dim, self.num_heads], dim=-1 + ) + + use_precomputed_states = ( + cache_params is not None + and cache_params.has_previous_state + and seq_len == 1 + and cache_params.conv_states[self.layer_idx].shape[0] + == cache_params.ssm_states[self.layer_idx].shape[0] + == batch_size + and cache_position is not None + and cache_position[0] > 0 + ) + + # 2. Convolution sequence transformation + if use_precomputed_states: + cache_params.conv_states[self.layer_idx] = cache_params.conv_states[self.layer_idx].roll(shifts=-1, dims=-1) + cache_params.conv_states[self.layer_idx][:, :, -1] = hidden_states_B_C[:, 0, :].to(cache_params.conv_states[self.layer_idx].device) + + # We need to guarantee that anything regarding the cache is on the same device + conv_states = cache_params.conv_states[self.layer_idx].to(device=self.conv1d.weight.device) + + hidden_states_B_C = torch.sum( + conv_states * self.conv1d.weight.squeeze(1), dim=-1 + ) + if self.use_conv_bias: + hidden_states_B_C = hidden_states_B_C + self.conv1d.bias + hidden_states_B_C = self.act(hidden_states_B_C) + else: + # Init cache + if cache_params is not None: + hidden_states_B_C_transposed = hidden_states_B_C.transpose(1, 2) + conv_states = nn.functional.pad( + hidden_states_B_C_transposed, (self.conv_kernel_size - hidden_states_B_C_transposed.shape[-1], 0) + ) + cache_params.conv_states[self.layer_idx].copy_(conv_states) + + hidden_states_B_C = self.act(self.conv1d(hidden_states_B_C.transpose(1, 2))[..., :seq_len].transpose(1, 2)) + + hidden_states_B_C = apply_mask_to_padding_states(hidden_states_B_C, attention_mask) + hidden_states, B, C = torch.split( + hidden_states_B_C, + [self.intermediate_size, self.n_groups * self.ssm_state_size, self.n_groups * self.ssm_state_size], + dim=-1 + ) + + # 3. SSM transformation + A = -torch.exp(self.A_log.float()) # [num_heads] + if use_precomputed_states: + # We need to guarantee that anything regarding the cache is on the same device + cache_device = cache_params.ssm_states[self.layer_idx].device + + # Note: there is no need to pad parameter matrices here, as there is just one new token + # for batched generation + dt = dt[:, 0, :][:, None, ...] + dt = dt.transpose(1, 2).expand(batch_size, dt.shape[-1], self.head_dim) + # [num_heads] -> [num_heads, head_dim] + dt_bias = self.dt_bias[..., None].expand(self.dt_bias.shape[0], self.head_dim) + + dt = torch.nn.functional.softplus(dt + dt_bias.to(dt.dtype)) + dt = torch.clamp(dt, self.time_step_limit[0], self.time_step_limit[1]) + A = A[..., None, None].expand(self.num_heads, self.head_dim, self.ssm_state_size).to(dtype=torch.float32) + # [bsz, num_heads, head_dim, state_size] + dA = (torch.exp(dt[..., None] * A)).to(device=cache_device) + + # Discretize B + # [bsz, n_groups * state_size] -> [bsz, n_groups, 1, state_size] -> + # -> [bsz, n_groups, group to head repetition factor, state_size] -> [bsz, num_heads, state_size] + B = B.reshape(batch_size, self.n_groups, -1)[..., None, :] + B = B.expand(batch_size, self.n_groups, self.num_heads // self.n_groups, B.shape[-1]).contiguous() + B = B.reshape(batch_size, -1, B.shape[-1]) + # [bsz, num_heads, head_dim, state_size] + dB = dt[..., None] * B[..., None, :] + + # Discretize x into dB + # [bsz, intermediate_size] -> [bsz, num_heads, head_dim] + hidden_states = hidden_states.reshape(batch_size, -1, self.head_dim) + dBx = (dB * hidden_states[..., None]).to(device=cache_device) + + # State calculation + cache_params.ssm_states[self.layer_idx].copy_( + cache_params.ssm_states[self.layer_idx] * dA + dBx + ) + + # Subsequent output + # [bsz, n_groups * state_size] -> [bsz, num_heads, state_size] + C = C.reshape(batch_size, self.n_groups, -1)[..., None, :] + C = C.expand(batch_size, self.n_groups, self.num_heads // self.n_groups, C.shape[-1]).contiguous() + C = C.reshape(batch_size, -1, C.shape[-1]) + # [bsz, num_heads, head_dim] + + ssm_states = cache_params.ssm_states[self.layer_idx].to(device=C.device, dtype=C.dtype) # Shape: [b, h, d, n] + # Reshape ssm_states to merge the first two dimensions + ssm_states_reshaped = ssm_states.view(batch_size * self.num_heads, self.head_dim, self.ssm_state_size) # Shape: [b*h, d, n] + C_reshaped = C.view(batch_size * self.num_heads, self.ssm_state_size, 1) # Shape: [b*h, n, 1] + y = torch.bmm(ssm_states_reshaped, C_reshaped) + y = y.view(batch_size, self.num_heads, self.head_dim) + + # D skip connection + # [num_heads] -> [num_heads, head_dim] + D = self.D[..., None].expand(self.D.shape[0], self.head_dim) + y = (y + hidden_states * D).to(y.dtype) + + # [bsz, num_heads, head_dim] -> [bsz, 1, intermediate_size] + y = y.reshape(batch_size, -1)[:, None, ...] + else: + # begin ssd naive implementation without einsums + dt = nn.functional.softplus(dt + self.dt_bias) + dt = torch.clamp(dt, self.time_step_limit[0], self.time_step_limit[1]) + hidden_states = hidden_states.reshape(batch_size, seq_len, -1, self.head_dim).float() + B = B.reshape(batch_size, seq_len, -1, self.ssm_state_size).float() + C = C.reshape(batch_size, seq_len, -1, self.ssm_state_size).float() + B = B.repeat_interleave(self.num_heads // self.n_groups, dim=2, output_size=self.num_heads) + C = C.repeat_interleave(self.num_heads // self.n_groups, dim=2, output_size=self.num_heads) + pad_size = (self.chunk_size - seq_len % self.chunk_size) % self.chunk_size + + D_residual = self.D[..., None] * pad_tensor_by_size(hidden_states, pad_size) + + # Discretize x and A + hidden_states = hidden_states * dt[..., None] + A = A.to(hidden_states.dtype) * dt + + # Rearrange into blocks/chunks + hidden_states, A, B, C = [reshape_into_chunks(t, pad_size, self.chunk_size) for t in (hidden_states, A, B, C)] + + # [bsz, -1, chunk_size, num_heads] -> [bsz, num_heads, -1, chunk_size] + A = A.permute(0, 3, 1, 2) + A_cumsum = torch.cumsum(A, dim=-1) + + # 1. Compute the output for each intra-chunk (diagonal blocks) + # This is the analog of a causal mask + L = torch.exp(segment_sum(A)) + + # Contraction of C and B to get G (attention-weights like) + G_intermediate = C[:, :, :, None, :, :] * B[:, :, None, :, :, :] # shape: (b, c, l, s, h, n) + G = G_intermediate.sum(dim=-1) # shape: (b, c, l, s, h) + + # Compute M, equivalent to applying attention mask to weights + M_intermediate = G[..., None] * L.permute(0, 2, 3, 4, 1)[..., None] + M = M_intermediate.sum(dim=-1) + + # Compute Y_diag (apply to values) + Y_diag = (M[..., None] * hidden_states[:, :, None]).sum(dim=3) + + # 2. Compute the state for each intra-chunk + # (right term of low-rank factorization of off-diagonal blocks; B terms) + decay_states = torch.exp((A_cumsum[:, :, :, -1:] - A_cumsum)) + B_decay = B * decay_states.permute(0, -2, -1, 1)[..., None] + states = (B_decay[..., None, :] * hidden_states[..., None]).sum(dim=2) + + # 3. Compute the inter-chunk SSM recurrence; produces correct SSM states at chunk boundaries + # (middle term of factorization of off-diag blocks; A terms) + if use_precomputed_states: + previous_states = cache_params.ssm_states[self.layer_idx][:, None, ...].to(device=states.device) + else: + previous_states = torch.zeros_like(states[:, :1]) + states = torch.cat([previous_states, states], dim=1) + decay_chunk = torch.exp(segment_sum(nn.functional.pad(A_cumsum[:, :, :, -1], (1, 0)))) + decay_chunk = decay_chunk.transpose(1, 3) + new_states = (decay_chunk[..., None, None] * states[:, :, None, ...]).sum(dim=1) + states, ssm_state = new_states[:, :-1], new_states[:, -1] + + # 4. Compute state -> output conversion per chunk + # (left term of low-rank factorization of off-diagonal blocks; C terms) + state_decay_out = torch.exp(A_cumsum) + C_times_states = (C[..., None, :] * states[:, :, None, ...]) + state_decay_out_permuted = state_decay_out.permute(0, 2, 3, 1) + Y_off = (C_times_states.sum(-1) * state_decay_out_permuted[..., None]) + + # Add output of intra-chunk and inter-chunk terms (diagonal and off-diagonal blocks) + y = Y_diag + Y_off + # [bsz, -1, self.chunk_size, num_heads, head_dim] -> [bsz, (padded) seq_len, num_heads, head_dim] + y = y.reshape(batch_size, -1, self.num_heads, self.head_dim) + + y = y + D_residual + # Cutting off padded chunks + if pad_size > 0: + y = y[:, :seq_len, :, :] + y = y.reshape(batch_size, seq_len, -1) + + # Init cache + if ssm_state is not None and cache_params is not None: + cache_params.ssm_states[self.layer_idx].copy_(ssm_state) + cache_params.has_previous_state = True + + scan_output = self.norm(y, gate) + + # end ssd naive + + # 4. Final linear projection + contextualized_states = self.out_proj(scan_output.to(dtype)) # [batch, seq_len, hidden_size] + return contextualized_states + # fmt: on + + def forward( + self, + hidden_states, + cache_params: Optional[HybridMambaAttentionDynamicCache] = None, + cache_position: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + ): + if is_fast_path_available and "cuda" in self.in_proj.weight.device.type: + return self.cuda_kernels_forward(hidden_states, cache_params, cache_position, attention_mask) + dtype = hidden_states.dtype + if attention_mask is not None and attention_mask.shape[1] > 1 and attention_mask.shape[0] > 1: + # tune out hidden states for pad tokens, see https://github.com/state-spaces/mamba/issues/66 + hidden_states = (hidden_states * attention_mask[:, :, None]).to(dtype) + + return self.torch_forward(hidden_states, cache_params, cache_position, attention_mask) + + +class GraniteMoeHybridRMSNormGated(torch.nn.Module): + def __init__(self, hidden_size, eps=1e-6): + super().__init__() + self.weight = nn.Parameter(torch.ones(hidden_size)) + self.variance_epsilon = eps + + def forward(self, hidden_states, gate=None): + input_dtype = hidden_states.dtype + hidden_states = hidden_states.to(torch.float32) + + if gate is not None: + hidden_states = hidden_states * nn.functional.silu(gate.to(torch.float32)) + variance = hidden_states.pow(2).mean(-1, keepdim=True) + hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon) + + return self.weight * hidden_states.to(input_dtype) + + +class GraniteMoeHybridMLP(nn.Module): + """ + MLP layer for shared experts + + Args: + config: + Configuration object with model hyperparameters. + """ + + def __init__(self, config: GraniteMoeHybridConfig): + super(GraniteMoeHybridMLP, self).__init__() + + self.input_size = config.hidden_size + self.hidden_size = config.shared_intermediate_size + self.activation = ACT2FN[config.hidden_act] + self.input_linear = nn.Linear(self.input_size, self.hidden_size * 2, bias=False) + self.output_linear = nn.Linear(self.hidden_size, self.input_size, bias=False) + + def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: + hidden_states = self.input_linear(hidden_states) + chunked_hidden_states = hidden_states.chunk(2, dim=-1) + hidden_states = self.activation(chunked_hidden_states[0]) * chunked_hidden_states[1] + hidden_states = self.output_linear(hidden_states) + return hidden_states + + +class GraniteMoeHybridRMSNorm(nn.Module): + def __init__(self, hidden_size, eps=1e-6): + """ + GraniteMoeHybridRMSNorm is equivalent to T5LayerNorm + """ + super().__init__() + self.weight = nn.Parameter(torch.ones(hidden_size)) + self.variance_epsilon = eps + + def forward(self, hidden_states): + input_dtype = hidden_states.dtype + hidden_states = hidden_states.to(torch.float32) + variance = hidden_states.pow(2).mean(-1, keepdim=True) + hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon) + return self.weight * hidden_states.to(input_dtype) + + def extra_repr(self): + return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}" + + +class GraniteMoeHybridParallelExperts(nn.Module): + def __init__(self, num_experts: int, input_size: int, output_size: int) -> None: + """ + Initialize the GraniteMoeHybridParallelExperts module. + The experts weights are stored in [num_experts, output_size, input_size] format. Such that it's compatible with + many MoE libraries, such as [Megablock](https://github.com/databricks/megablocks) and + [ScatterMoE](https://github.com/shawntan/scattermoe), as well as the + [MoE kernel](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/fused_moe/fused_moe.py) + used in vllm. + + Args: + num_experts (int): + Number of experts. + input_size (int): + Size of the input. + output_size (int): + Size of the output. + """ + super().__init__() + self.weight = nn.Parameter(torch.empty(num_experts, output_size, input_size)) + self.num_experts = num_experts + self.input_size = input_size + self.output_size = output_size + + def forward(self, inputs, expert_size): + """ + Forward pass of the GraniteMoeHybridParallelExperts module. + + Args: + inputs (Tensor): + Input tensor. + expert_size: + Expert size information. + + Returns: + Tensor: Output tensor. + """ + input_list = inputs.split(expert_size, dim=0) + output_list = [] + for i in range(self.num_experts): + output_list.append(F.linear(input_list[i], self.weight[i])) + results = torch.cat(output_list, dim=0) + return results + + +class GraniteMoeHybridTopKGating(nn.Module): + def __init__(self, input_size: int, num_experts: int, top_k: int): + """ + Initialize the top-k gating mechanism. + Args: + input_size (`int`): + Size of the input. + num_experts (`int`): + Number of experts. + top_k (`int`): + Number of top experts to select. + """ + super().__init__() + + self.num_experts = num_experts + self.input_size = input_size + self.top_k = top_k + + self.layer = nn.Linear(input_size, num_experts, bias=False) + + def forward(self, hidden_states): + # compute the top_k routing decision + logits = self.layer(hidden_states).float() # [batch_size x seq_len, num_experts] + top_k_logits, top_k_indices = logits.topk(self.top_k, dim=1) # [num_tokens, top_k] + top_k_gates = torch.softmax(top_k_logits, dim=1).type_as(hidden_states) # [num_tokens, top_k] + + # compute number of input given to each expert + zeros = torch.zeros( + [top_k_gates.size(0), self.num_experts], dtype=top_k_gates.dtype, device=top_k_gates.device + ) # [num_tokens, num_experts] + gates = zeros.scatter(1, top_k_indices, 1) # [num_tokens, num_experts] + expert_size = gates.long().sum(0) # [num_experts,] + # (This cause torch.compile to fail with `torch._dynamo.exc.Unsupported: Backend compiler failed with a fake tensor exception at`) + # (and `DataDependentOutputException`) + expert_size = expert_size.tolist() + + # sort and group input tokens according to expert assignment + top_k_experts = top_k_indices.flatten() # [num_tokens * top_k] + _, index_sorted_experts = top_k_experts.sort(0) # [num_tokens * top_k] + batch_index = index_sorted_experts.div(self.top_k, rounding_mode="trunc") # [num_tokens * top_k] + + # gather the gate values for grouped input tokens + top_k_gates = top_k_gates.flatten() # [num_tokens * top_k] + batch_gates = top_k_gates[index_sorted_experts] # [num_tokens * top_k] + + return index_sorted_experts, batch_index, batch_gates, expert_size, logits + + +class GraniteMoeHybridMoE(nn.Module): + """ + A Sparsely gated mixture of experts layer with 1-layer Feed-Forward networks as experts. + + Args: + config: + Configuration object with model hyperparameters. + """ + + def __init__(self, config: GraniteMoeHybridConfig): + super(GraniteMoeHybridMoE, self).__init__() + + self.input_size = config.hidden_size + self.hidden_size = config.intermediate_size + self.activation = ACT2FN[config.hidden_act] + self.input_linear = GraniteMoeHybridParallelExperts( + config.num_local_experts, self.input_size, self.hidden_size * 2 + ) + self.output_linear = GraniteMoeHybridParallelExperts( + config.num_local_experts, self.hidden_size, self.input_size + ) + + self.router = GraniteMoeHybridTopKGating( + input_size=self.input_size, + num_experts=config.num_local_experts, + top_k=config.num_experts_per_tok, + ) + + def forward(self, layer_input): + """ + Forward pass of the mixture of experts layer. + + Args: + layer_input (Tensor): + Input tensor. + + Returns: + Tensor: + Output tensor. + Tensor: + Router logits. + """ + bsz, length, emb_size = layer_input.size() + layer_input = layer_input.reshape(-1, emb_size) + _, batch_index, batch_gates, expert_size, router_logits = self.router(layer_input) + + expert_inputs = layer_input[batch_index] + hidden_states = self.input_linear(expert_inputs, expert_size) + chunked_hidden_states = hidden_states.chunk(2, dim=-1) + hidden_states = self.activation(chunked_hidden_states[0]) * chunked_hidden_states[1] + expert_outputs = self.output_linear(hidden_states, expert_size) + + expert_outputs = expert_outputs * batch_gates[:, None] + + zeros = torch.zeros((bsz * length, self.input_size), dtype=expert_outputs.dtype, device=expert_outputs.device) + layer_output = zeros.index_add(0, batch_index, expert_outputs) + layer_output = layer_output.view(bsz, length, self.input_size) + return layer_output, router_logits + + +class GraniteMoeHybridDecoderLayer(GradientCheckpointingLayer): + def __init__(self, config: GraniteMoeHybridConfig, layer_idx: int): + super().__init__() + self.hidden_size = config.hidden_size + # Either attention or mamba will be initialized, depending on the layer type. + self.self_attn = None + self.block_sparse_moe = GraniteMoeHybridMoE(config) + self.input_layernorm = GraniteMoeHybridRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.post_attention_layernorm = GraniteMoeHybridRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + + self.residual_multiplier = config.residual_multiplier + self.shared_mlp = GraniteMoeHybridMLP(config) + self.mamba = None + + if config.layers_block_type[layer_idx] == "mamba": + self.mamba = GraniteMoeHybridMambaLayer(config, layer_idx) + else: + self.self_attn = GraniteMoeHybridAttention(config, layer_idx) + self.layer_type = config.layers_block_type[layer_idx] + + def forward( + self, + hidden_states: torch.Tensor, + attention_mask: Optional[torch.Tensor] = None, + past_key_value: Optional[Cache] = None, + output_attentions: Optional[bool] = False, + use_cache: Optional[bool] = False, + cache_position: Optional[torch.LongTensor] = None, + output_router_logits: Optional[bool] = False, + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, + **kwargs, + ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: + """ + Args: + hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)` + attention_mask (`torch.FloatTensor`, *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. + past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states + 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`). + cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*): + Indices depicting the position of the input sequence tokens in the sequence + 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. + position_embeddings (`Tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*): + Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`, + with `head_dim` being the embedding dimension of each attention head. + kwargs (`dict`, *optional*): + Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code + into the model + """ + residual = hidden_states + hidden_states = self.input_layernorm(hidden_states) + + if self.mamba is not None: + hidden_states = self.mamba( + hidden_states=hidden_states, + cache_position=cache_position, + cache_params=past_key_value, + attention_mask=attention_mask, + ) + # No attention weights for state space layers + self_attn_weights = None + else: + hidden_states, self_attn_weights, _ = self.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, + position_embeddings=position_embeddings, + **kwargs, + ) + + hidden_states = residual + hidden_states * self.residual_multiplier + + # Fully Connected + residual = hidden_states + hidden_states = self.post_attention_layernorm(hidden_states) + moe_hidden_states, router_logits = self.block_sparse_moe(hidden_states) + + hidden_states = moe_hidden_states + self.shared_mlp(hidden_states) + hidden_states = residual + hidden_states * self.residual_multiplier + + outputs = (hidden_states,) + + if output_attentions: + outputs += (self_attn_weights,) + + if use_cache: + outputs += (past_key_value,) + + if output_router_logits: + outputs += (router_logits,) + + return outputs + + +GRANITEMOEHYBRID_START_DOCSTRING = r""" + 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](https://pytorch.org/docs/stable/nn.html#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. + + Parameters: + 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. +""" + + +@add_start_docstrings( + "The bare GraniteMoeHybrid Model outputting raw hidden-states without any specific head on top.", + GRANITEMOEHYBRID_START_DOCSTRING, +) +class GraniteMoeHybridPreTrainedModel(PreTrainedModel): + config_class = GraniteMoeHybridConfig + base_model_prefix = "model" + supports_gradient_checkpointing = True + _no_split_modules = ["GraniteMoeHybridDecoderLayer"] + _skip_keys_device_placement = ["past_key_values"] + _supports_flash_attn_2 = True + _supports_sdpa = True + _supports_cache_class = True + _supports_quantized_cache = True + _supports_static_cache = False # MoE models don't work with torch.compile (`torch.where(condition)` not supported) + _is_stateful = True + + def _init_weights(self, module): + if isinstance(module, nn.Linear): + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) + if module.bias is not None: + module.bias.data.zero_() + elif isinstance(module, nn.Embedding): + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) + if module.padding_idx is not None: + module.weight.data[module.padding_idx].zero_() + elif isinstance(module, GraniteMoeHybridRMSNorm): + module.weight.data.fill_(1.0) + elif isinstance(module, GraniteMoeHybridParallelExperts): + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) + # Initialize Mamba modules + if isinstance(module, (nn.Conv1d)): + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) + if module.bias is not None: + module.bias.data.zero_() + elif isinstance(module, GraniteMoeHybridMambaLayer): + module.dt_bias.data.fill_(1.0) + module.A_log.data = torch.log(torch.arange(1, module.num_heads + 1)) + module.D.data.fill_(1.0) + elif isinstance(module, GraniteMoeHybridRMSNormGated): + module.weight.data.fill_(1.0) + + +class GraniteMoeHybridRotaryEmbedding(nn.Module): + def __init__(self, config: GraniteMoeHybridConfig, device=None): + super().__init__() + # BC: "rope_type" was originally "type" + if hasattr(config, "rope_scaling") and config.rope_scaling is not None: + self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type")) + else: + self.rope_type = "default" + self.max_seq_len_cached = config.max_position_embeddings + self.original_max_seq_len = config.max_position_embeddings + + self.config = config + self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type] + + inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self.original_inv_freq = self.inv_freq + + @torch.no_grad() + @dynamic_rope_update # power user: used with advanced RoPE types (e.g. dynamic rope) + def forward(self, x, position_ids): + inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device) + position_ids_expanded = position_ids[:, None, :].float() + + device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu" + with torch.autocast(device_type=device_type, enabled=False): # Force float32 + freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2) + emb = torch.cat((freqs, freqs), dim=-1) + cos = emb.cos() * self.attention_scaling + sin = emb.sin() * self.attention_scaling + + return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype) + + +GRANITEMOEHYBRID_INPUTS_DOCSTRING = r""" + Args: + input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`): + Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide + it. + + Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and + [`PreTrainedTokenizer.__call__`] for details. + + [What are input IDs?](../glossary#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?](../glossary#attention-mask) + + Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and + [`PreTrainedTokenizer.__call__`] for details. + + If `past_key_values` is used, optionally only the last `input_ids` have to be input (see + `past_key_values`). + + If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`] + and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more + information on the default strategy. + + - 1 indicates the head is **not masked**, + - 0 indicates the head is **masked**. + 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?](../glossary#position-ids) + past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *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`. + + Two formats are allowed: + - a [`~cache_utils.Cache`] instance; + - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of + shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy + cache format. + + The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the + legacy cache format will be returned. + + If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't + have their past key value states given to this model) of shape `(batch_size, 1)` 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. + 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`). + output_attentions (`bool`, *optional*): + Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned + tensors for more detail. + output_hidden_states (`bool`, *optional*): + Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for + more detail. + return_dict (`bool`, *optional*): + Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. + 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. +""" + + +@add_start_docstrings( + "The bare GraniteMoeHybrid Model outputting raw hidden-states without any specific head on top.", + GRANITEMOEHYBRID_START_DOCSTRING, +) +class GraniteMoeHybridModel(GraniteMoeHybridPreTrainedModel): + """ + Transformer decoder consisting of *config.num_hidden_layers* layers. + Each layer is a [`GraniteMoeHybridDecoderLayer`] + + Args: + config: GraniteMoeHybridConfig + """ + + def __init__(self, config: GraniteMoeHybridConfig): + super().__init__(config) + self.padding_idx = config.pad_token_id + self.vocab_size = config.vocab_size + + self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx) + self.layers = nn.ModuleList( + [GraniteMoeHybridDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + self.norm = GraniteMoeHybridRMSNorm(config.hidden_size, eps=config.rms_norm_eps) + self.gradient_checkpointing = False + + self.embedding_multiplier = config.embedding_multiplier + self.hidden_size = config.hidden_size + self.num_heads = config.num_attention_heads + self.head_dim = self.hidden_size // self.num_heads + self.max_position_embeddings = config.max_position_embeddings + self.rope_theta = config.rope_theta + + self.position_embedding_type = config.position_embedding_type + self.rotary_emb = GraniteMoeHybridRotaryEmbedding(config) if self.position_embedding_type == "rope" else None + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.embed_tokens + + def set_input_embeddings(self, value): + self.embed_tokens = value + + @can_return_tuple + @add_start_docstrings_to_model_forward(GRANITEMOEHYBRID_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + output_router_logits: Optional[bool] = None, + return_dict: Optional[bool] = None, + cache_position: Optional[torch.LongTensor] = None, + ) -> Union[Tuple, BaseModelOutputWithPast]: + 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 must specify exactly one of input_ids or inputs_embeds") + + 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.embed_tokens(input_ids) + + inputs_embeds = inputs_embeds * self.embedding_multiplier + + ## overwritten because `HybridMambaAttentionDynamicCache` is needed + if use_cache and past_key_values is None: + logger.warning_once( + "GraniteMoeHybrid requires an initialized `HybridMambaAttentionDynamicCache` to return a cache. " + "Because one was not provided, no cache will be returned." + ) + + if cache_position is None: + past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 + cache_position = torch.arange( + past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device + ) + if position_ids is None: + position_ids = cache_position.unsqueeze(0) + + causal_mask = self._update_causal_mask( + attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions + ) + mamba_mask = self._update_mamba_mask(attention_mask, cache_position) + + # embed positions + hidden_states = inputs_embeds + + position_embeddings = None + # create position embeddings to be shared across the decoder layers + if self.rotary_emb is not None: + position_embeddings = self.rotary_emb(hidden_states, position_ids) + + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + all_router_logits = () if output_router_logits else None + next_decoder_cache = None + + for decoder_layer in self.layers: + # Depending on the layer type we opt for 2D base attention mask (Mamba) or 4D causal mask (Attention) + layer_mask = mamba_mask if decoder_layer.layer_type == "mamba" else causal_mask + + if output_hidden_states: + all_hidden_states += (hidden_states,) + + layer_outputs = decoder_layer( + hidden_states, + attention_mask=layer_mask, + past_key_value=past_key_values, + output_attentions=output_attentions, + use_cache=use_cache, + cache_position=cache_position, + output_router_logits=output_router_logits, + position_embeddings=position_embeddings, + ) + + hidden_states = layer_outputs[0] + + if use_cache: + next_decoder_cache = layer_outputs[2 if output_attentions else 1] + + if output_attentions: + if layer_outputs[1] is not None: + # append attentions only of attention layers. Mamba layers return `None` as the attention weights + all_self_attns += (layer_outputs[1],) + + if output_router_logits: + if layer_outputs[-1] is not None: + # append router logits only of expert layers. Regular MLP layers return `None` as the router logits + all_router_logits += (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 = next_decoder_cache if use_cache else None + + return MoeModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=next_cache, + hidden_states=all_hidden_states, + attentions=all_self_attns, + router_logits=all_router_logits, + ) + + def _update_causal_mask( + self, + attention_mask: Union[torch.Tensor, "BlockMask"], + input_tensor: torch.Tensor, + cache_position: torch.Tensor, + past_key_values: Cache, + output_attentions: bool = False, + ): + if self.config._attn_implementation == "flash_attention_2": + if attention_mask is not None and (attention_mask == 0.0).any(): + return attention_mask + return None + if self.config._attn_implementation == "flex_attention": + if isinstance(attention_mask, torch.Tensor): + attention_mask = make_flex_block_causal_mask(attention_mask) + return attention_mask + + # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in + # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail + # to infer the attention mask. + past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 + using_static_cache = isinstance(past_key_values, StaticCache) + + # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward + if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions: + if AttentionMaskConverter._ignore_causal_mask_sdpa( + attention_mask, + inputs_embeds=input_tensor, + past_key_values_length=past_seen_tokens, + is_training=self.training, + ): + return None + + dtype = input_tensor.dtype + sequence_length = input_tensor.shape[1] + if using_static_cache: + target_length = past_key_values.get_max_cache_shape() + else: + target_length = ( + attention_mask.shape[-1] + if isinstance(attention_mask, torch.Tensor) + else past_seen_tokens + sequence_length + 1 + ) + + # In case the provided `attention` mask is 2D, we generate a causal mask here (4D). + causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position( + attention_mask, + sequence_length=sequence_length, + target_length=target_length, + dtype=dtype, + cache_position=cache_position, + batch_size=input_tensor.shape[0], + ) + + if ( + self.config._attn_implementation == "sdpa" + and attention_mask is not None + and attention_mask.device.type in ["cuda", "xpu", "npu"] + and not output_attentions + ): + # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when + # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path. + # Details: https://github.com/pytorch/pytorch/issues/110213 + min_dtype = torch.finfo(dtype).min + causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype) + + return causal_mask + + @staticmethod + def _prepare_4d_causal_attention_mask_with_cache_position( + attention_mask: torch.Tensor, + sequence_length: int, + target_length: int, + dtype: torch.dtype, + cache_position: torch.Tensor, + batch_size: int, + **kwargs, + ): + """ + Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape + `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing. + + Args: + attention_mask (`torch.Tensor`): + A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape + `(batch_size, 1, query_length, key_value_length)`. + sequence_length (`int`): + The sequence length being processed. + target_length (`int`): + The target length: when generating with static cache, the mask should be as long as the static cache, + to account for the 0 padding, the part of the cache that is not filled yet. + dtype (`torch.dtype`): + The dtype to use for the 4D attention mask. + cache_position (`torch.Tensor`): + Indices depicting the position of the input sequence tokens in the sequence. + batch_size (`torch.Tensor`): + Batch size. + """ + if attention_mask is not None and attention_mask.dim() == 4: + # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing. + causal_mask = attention_mask + else: + min_dtype = torch.finfo(dtype).min + causal_mask = torch.full( + (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=cache_position.device + ) + if sequence_length != 1: + causal_mask = torch.triu(causal_mask, diagonal=1) + causal_mask *= torch.arange(target_length, device=cache_position.device) > cache_position.reshape(-1, 1) + causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1) + if attention_mask is not None: + causal_mask = causal_mask.clone() # copy to contiguous memory for in-place edit + mask_length = attention_mask.shape[-1] + padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :].to( + causal_mask.device + ) + padding_mask = padding_mask == 0 + causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill( + padding_mask, min_dtype + ) + + return causal_mask + + def _update_mamba_mask(self, attention_mask, cache_position): + """ + No need for zeroing states when + 1. Cached forward + 2. Attending to all inputs + """ + mamba_mask = attention_mask + if cache_position[0] > 0 or (attention_mask is not None and torch.all(attention_mask == 1)): + mamba_mask = None + return mamba_mask + + +def load_balancing_loss_func( + gate_logits: Union[torch.Tensor, Tuple[torch.Tensor], None], + num_experts: Optional[int] = None, + top_k=2, + attention_mask: Optional[torch.Tensor] = None, +) -> Union[torch.Tensor, int]: + r""" + Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch. + + See Switch Transformer (https://arxiv.org/abs/2101.03961) for more details. This function implements the loss + function presented in equations (4) - (6) of the paper. It aims at penalizing cases where the routing between + experts is too unbalanced. + + Args: + gate_logits: + Logits from the `gate`, should be a tuple of model.config.num_hidden_layers tensors of + shape [batch_size X sequence_length, num_experts]. + num_experts: + Number of experts + top_k: + The number of experts to route per-token, can be also interpreted as the `top-k` routing + parameter. + attention_mask (`torch.Tensor`, *optional*): + The attention_mask used in forward function + shape [batch_size X sequence_length] if not None. + + Returns: + The auxiliary loss. + """ + if gate_logits is None or not isinstance(gate_logits, tuple): + return 0 + + if isinstance(gate_logits, tuple): + compute_device = gate_logits[0].device + concatenated_gate_logits = torch.cat([layer_gate.to(compute_device) for layer_gate in gate_logits], dim=0) + + routing_weights = torch.nn.functional.softmax(concatenated_gate_logits, dim=-1) + + _, selected_experts = torch.topk(routing_weights, top_k, dim=-1) + + expert_mask = torch.nn.functional.one_hot(selected_experts, num_experts) + + if attention_mask is None: + # Compute the percentage of tokens routed to each experts + tokens_per_expert = torch.mean(expert_mask.float(), dim=0) + + # Compute the average probability of routing to these experts + router_prob_per_expert = torch.mean(routing_weights, dim=0) + else: + batch_size, sequence_length = attention_mask.shape + num_hidden_layers = concatenated_gate_logits.shape[0] // (batch_size * sequence_length) + + # Compute the mask that masks all padding tokens as 0 with the same shape of expert_mask + expert_attention_mask = ( + attention_mask[None, :, :, None, None] + .expand((num_hidden_layers, batch_size, sequence_length, top_k, num_experts)) + .reshape(-1, top_k, num_experts) + .to(compute_device) + ) + + # Compute the percentage of tokens routed to each experts + tokens_per_expert = torch.sum(expert_mask.float() * expert_attention_mask, dim=0) / torch.sum( + expert_attention_mask, dim=0 + ) + + # Compute the mask that masks all padding tokens as 0 with the same shape of tokens_per_expert + router_per_expert_attention_mask = ( + attention_mask[None, :, :, None] + .expand((num_hidden_layers, batch_size, sequence_length, num_experts)) + .reshape(-1, num_experts) + .to(compute_device) + ) + + # Compute the average probability of routing to these experts + router_prob_per_expert = torch.sum(routing_weights * router_per_expert_attention_mask, dim=0) / torch.sum( + router_per_expert_attention_mask, dim=0 + ) + + overall_loss = torch.sum(tokens_per_expert * router_prob_per_expert.unsqueeze(0)) + return overall_loss * num_experts + + +class GraniteMoeHybridForCausalLM(GraniteMoeHybridPreTrainedModel, GenerationMixin): + _tied_weights_keys = ["lm_head.weight"] + + def __init__(self, config: GraniteMoeHybridConfig): + super().__init__(config) + self.model = GraniteMoeHybridModel(config) + self.vocab_size = config.vocab_size + self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False) + + self.router_aux_loss_coef = config.router_aux_loss_coef + self.num_experts = config.num_local_experts + self.num_experts_per_tok = config.num_experts_per_tok + + # Initialize weights and apply final processing + self.post_init() + + def get_input_embeddings(self): + return self.model.embed_tokens + + def set_input_embeddings(self, value): + self.model.embed_tokens = value + + def get_output_embeddings(self): + return self.lm_head + + def set_output_embeddings(self, new_embeddings): + self.lm_head = new_embeddings + + def set_decoder(self, decoder): + self.model = decoder + + def get_decoder(self): + return self.model + + @add_start_docstrings_to_model_forward(GRANITEMOEHYBRID_INPUTS_DOCSTRING) + @replace_return_docstrings(output_type=MoeCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC) + def forward( + self, + input_ids: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + labels: Optional[torch.LongTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + output_router_logits: Optional[bool] = None, + return_dict: Optional[bool] = None, + cache_position: Optional[torch.LongTensor] = None, + logits_to_keep: Union[int, torch.Tensor] = 0, + **kwargs, + ) -> Union[Tuple, MoeCausalLMOutputWithPast]: + r""" + 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]`. + + logits_to_keep (`int` or `torch.Tensor`, *optional*): + 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: + + Example: + + ```python + >>> from transformers import AutoTokenizer, GraniteMoeHybridForCausalLM + + >>> model = GraniteMoeHybridForCausalLM.from_pretrained("ibm/PowerMoE-3b") + >>> tokenizer = AutoTokenizer.from_pretrained("ibm/PowerMoE-3b") + + >>> 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." + ```""" + output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions + output_router_logits = ( + output_router_logits if output_router_logits is not None else self.config.output_router_logits + ) + 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.model( + 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, + output_router_logits=output_router_logits, + return_dict=return_dict, + cache_position=cache_position, + ) + + # Only compute necessary logits + hidden_states = outputs[0] + slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep + logits = self.lm_head(hidden_states[:, slice_indices, :]) + logits = logits / self.config.logits_scaling + + loss = None + if labels is not None: + # Upcast to float if we need to compute the loss to avoid potential precision issues + logits = logits.float() + # Flatten the tokens + loss = self.loss_function( + logits, + labels, + vocab_size=self.config.vocab_size, + **kwargs, + ) + + aux_loss = None + if output_router_logits: + aux_loss = load_balancing_loss_func( + outputs.router_logits if return_dict else outputs[-1], + self.num_experts, + self.num_experts_per_tok, + attention_mask, + ) + if labels is not None: + loss += self.router_aux_loss_coef * aux_loss.to(loss.device) # make sure to reside in the same device + + if not return_dict: + output = (logits,) + outputs[1:] + if output_router_logits: + output = (aux_loss,) + output + return (loss,) + output if loss is not None else output + + return MoeCausalLMOutputWithPast( + loss=loss, + aux_loss=aux_loss, + logits=logits, + past_key_values=outputs.past_key_values, + hidden_states=outputs.hidden_states, + attentions=outputs.attentions, + router_logits=outputs.router_logits, + ) + + @staticmethod + def _reorder_cache(past_key_values, beam_idx): + reordered_past = () + for layer_past in past_key_values: + reordered_past += ( + tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past), + ) + return reordered_past + + def prepare_inputs_for_generation( + self, + input_ids, + past_key_values=None, + attention_mask=None, + inputs_embeds=None, + cache_position=None, + position_ids=None, + use_cache=True, + **kwargs, + ): + # Overwritten -- has a unique cache type, `HybridMambaAttentionDynamicCache` + + empty_past_kv = past_key_values is None + + # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens + # Exception 1: when passing input_embeds, input_ids may be missing entries + # Exception 2: some generation methods do special slicing of input_ids, so we don't need to do it here + # Exception 3: with synced GPUs cache_position may go out of bounds, but we only want dummy token in that case. + # (we can't check exception 3 while compiling) + if not empty_past_kv: + if ( + inputs_embeds is not None # Exception 1 + or cache_position[-1] >= input_ids.shape[1] # Exception 3 + ): + input_ids = input_ids[:, -cache_position.shape[0] :] + elif input_ids.shape[1] != cache_position.shape[0]: # Default case (the "else", a no op, is Exception 2) + input_ids = input_ids[:, cache_position] + else: + past_key_values = HybridMambaAttentionDynamicCache( + self.config, input_ids.shape[0], self.dtype, device=self.device + ) + + 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.masked_fill_(attention_mask == 0, 1) + if not empty_past_kv: + position_ids = position_ids[:, -input_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 empty_past_kv: + model_inputs = {"inputs_embeds": inputs_embeds} + else: + model_inputs = {"input_ids": input_ids.contiguous()} # `contiguous()` needed for compilation use cases + + model_inputs.update( + { + "position_ids": position_ids, + "past_key_values": past_key_values, + "use_cache": use_cache, + "attention_mask": attention_mask, + "cache_position": cache_position, + } + ) + return model_inputs + + def _supports_default_dynamic_cache(self) -> bool: + """ + Function overwritten as this class uses its own `HybridMambaAttentionDynamicCache` + and do not need to initialize the Cache in advance in order to save memory + (because no back and forth `to_legacy_cache` and `from_legacy_cache` will be performed + for `HybridMambaAttentionDynamicCache`). + """ + return False + + +__all__ = ["GraniteMoeHybridForCausalLM", "GraniteMoeHybridModel", "GraniteMoeHybridPreTrainedModel"] diff --git a/src/transformers/models/granitemoehybrid/modular_granitemoehybrid.py b/src/transformers/models/granitemoehybrid/modular_granitemoehybrid.py new file mode 100644 index 00000000000..c9fe70d5b0f --- /dev/null +++ b/src/transformers/models/granitemoehybrid/modular_granitemoehybrid.py @@ -0,0 +1,510 @@ +# coding=utf-8 +# Copyright 2025 IBM and the HuggingFace Inc. team. All rights reserved. +# +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from typing import List, Optional, Tuple, Union + +import torch +from torch import nn + +from ...cache_utils import Cache +from ...modeling_outputs import BaseModelOutputWithPast, MoeModelOutputWithPast +from ...utils import ( + add_start_docstrings, + add_start_docstrings_to_model_forward, + can_return_tuple, + logging, +) +from ..bamba.configuration_bamba import BambaConfig +from ..bamba.modeling_bamba import ( + BambaMixer, + BambaRMSNormGated, + HybridMambaAttentionDynamicCache, +) +from ..granitemoeshared.modeling_granitemoeshared import ( + GraniteMoeSharedAttention, + GraniteMoeSharedDecoderLayer, + GraniteMoeSharedForCausalLM, + GraniteMoeSharedMLP, + GraniteMoeSharedModel, + GraniteMoeSharedPreTrainedModel, +) +from .configuration_granitemoehybrid import GraniteMoeHybridConfig + + +logger = logging.get_logger(__name__) + + +class GraniteMoeHybridAttention(GraniteMoeSharedAttention): + def __init__(self, config: GraniteMoeHybridConfig, layer_idx: int): + super().__init__(config, layer_idx) + + +class GraniteMoeHybridMambaLayer(BambaMixer): + def __init__(self, config: GraniteMoeHybridConfig, layer_idx: int): + super().__init__(BambaConfig(config), layer_idx) + + +class GraniteMoeHybridRMSNormGated(BambaRMSNormGated): + def __init__(self, hidden_size, eps=1e-6): + super().__init__(hidden_size, eps) + + +class GraniteMoeHybridMLP(GraniteMoeSharedMLP): + def __init__(self, config: GraniteMoeHybridConfig): + super().__init__(config) + + +class GraniteMoeHybridDecoderLayer(GraniteMoeSharedDecoderLayer): + def __init__(self, config: GraniteMoeHybridConfig, layer_idx: int): + super().__init__(config, layer_idx) + self.shared_mlp = GraniteMoeHybridMLP(config) + # Either attention or mamba will be initialized, depending on the layer type. + self.self_attn = None + self.mamba = None + + if config.layers_block_type[layer_idx] == "mamba": + self.mamba = GraniteMoeHybridMambaLayer(config, layer_idx) + else: + self.self_attn = GraniteMoeHybridAttention(config, layer_idx) + self.layer_type = config.layers_block_type[layer_idx] + + def forward( + self, + hidden_states: torch.Tensor, + attention_mask: Optional[torch.Tensor] = None, + past_key_value: Optional[Cache] = None, + output_attentions: Optional[bool] = False, + use_cache: Optional[bool] = False, + cache_position: Optional[torch.LongTensor] = None, + output_router_logits: Optional[bool] = False, + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, + **kwargs, + ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: + """ + Args: + hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)` + attention_mask (`torch.FloatTensor`, *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. + past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states + 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`). + cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*): + Indices depicting the position of the input sequence tokens in the sequence + 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. + position_embeddings (`Tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*): + Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`, + with `head_dim` being the embedding dimension of each attention head. + kwargs (`dict`, *optional*): + Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code + into the model + """ + residual = hidden_states + hidden_states = self.input_layernorm(hidden_states) + + if self.mamba is not None: + hidden_states = self.mamba( + hidden_states=hidden_states, + cache_position=cache_position, + cache_params=past_key_value, + attention_mask=attention_mask, + ) + # No attention weights for state space layers + self_attn_weights = None + else: + hidden_states, self_attn_weights, _ = self.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, + position_embeddings=position_embeddings, + **kwargs, + ) + + hidden_states = residual + hidden_states * self.residual_multiplier + + # Fully Connected + residual = hidden_states + hidden_states = self.post_attention_layernorm(hidden_states) + moe_hidden_states, router_logits = self.block_sparse_moe(hidden_states) + + hidden_states = moe_hidden_states + self.shared_mlp(hidden_states) + hidden_states = residual + hidden_states * self.residual_multiplier + + outputs = (hidden_states,) + + if output_attentions: + outputs += (self_attn_weights,) + + if use_cache: + outputs += (past_key_value,) + + if output_router_logits: + outputs += (router_logits,) + + return outputs + + +GRANITEMOEHYBRID_START_DOCSTRING = r""" + 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](https://pytorch.org/docs/stable/nn.html#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. + + Parameters: + 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. +""" + + +@add_start_docstrings( + "The bare GraniteMoeHybrid Model outputting raw hidden-states without any specific head on top.", + GRANITEMOEHYBRID_START_DOCSTRING, +) +class GraniteMoeHybridPreTrainedModel(GraniteMoeSharedPreTrainedModel): + config_class = GraniteMoeHybridConfig + _no_split_modules = ["GraniteMoeHybridDecoderLayer"] + _is_stateful = True + + def _init_weights(self, module): + super()._init_weights() + # Initialize Mamba modules + if isinstance(module, (nn.Conv1d)): + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) + if module.bias is not None: + module.bias.data.zero_() + elif isinstance(module, GraniteMoeHybridMambaLayer): + module.dt_bias.data.fill_(1.0) + module.A_log.data = torch.log(torch.arange(1, module.num_heads + 1)) + module.D.data.fill_(1.0) + elif isinstance(module, GraniteMoeHybridRMSNormGated): + module.weight.data.fill_(1.0) + + +GRANITEMOEHYBRID_INPUTS_DOCSTRING = r""" + Args: + input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`): + Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide + it. + + Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and + [`PreTrainedTokenizer.__call__`] for details. + + [What are input IDs?](../glossary#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?](../glossary#attention-mask) + + Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and + [`PreTrainedTokenizer.__call__`] for details. + + If `past_key_values` is used, optionally only the last `input_ids` have to be input (see + `past_key_values`). + + If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`] + and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more + information on the default strategy. + + - 1 indicates the head is **not masked**, + - 0 indicates the head is **masked**. + 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?](../glossary#position-ids) + past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *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`. + + Two formats are allowed: + - a [`~cache_utils.Cache`] instance; + - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of + shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy + cache format. + + The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the + legacy cache format will be returned. + + If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't + have their past key value states given to this model) of shape `(batch_size, 1)` 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. + 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`). + output_attentions (`bool`, *optional*): + Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned + tensors for more detail. + output_hidden_states (`bool`, *optional*): + Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for + more detail. + return_dict (`bool`, *optional*): + Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. + 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. +""" + + +@add_start_docstrings( + "The bare GraniteMoeHybrid Model outputting raw hidden-states without any specific head on top.", + GRANITEMOEHYBRID_START_DOCSTRING, +) +class GraniteMoeHybridModel(GraniteMoeSharedModel): + """ + Transformer decoder consisting of *config.num_hidden_layers* layers. + Each layer is a [`GraniteMoeHybridDecoderLayer`] + + Args: + config: GraniteMoeHybridConfig + """ + + def __init__(self, config: GraniteMoeHybridConfig): + super().__init__(config) + self.layers = nn.ModuleList( + [GraniteMoeHybridDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] + ) + + @can_return_tuple + @add_start_docstrings_to_model_forward(GRANITEMOEHYBRID_INPUTS_DOCSTRING) + def forward( + self, + input_ids: torch.LongTensor = None, + attention_mask: Optional[torch.Tensor] = None, + position_ids: Optional[torch.LongTensor] = None, + past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None, + inputs_embeds: Optional[torch.FloatTensor] = None, + use_cache: Optional[bool] = None, + output_attentions: Optional[bool] = None, + output_hidden_states: Optional[bool] = None, + output_router_logits: Optional[bool] = None, + return_dict: Optional[bool] = None, + cache_position: Optional[torch.LongTensor] = None, + ) -> Union[Tuple, BaseModelOutputWithPast]: + 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 must specify exactly one of input_ids or inputs_embeds") + + 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.embed_tokens(input_ids) + + inputs_embeds = inputs_embeds * self.embedding_multiplier + + ## overwritten because `HybridMambaAttentionDynamicCache` is needed + if use_cache and past_key_values is None: + logger.warning_once( + "GraniteMoeHybrid requires an initialized `HybridMambaAttentionDynamicCache` to return a cache. " + "Because one was not provided, no cache will be returned." + ) + + if cache_position is None: + past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 + cache_position = torch.arange( + past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device + ) + if position_ids is None: + position_ids = cache_position.unsqueeze(0) + + causal_mask = self._update_causal_mask( + attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions + ) + mamba_mask = self._update_mamba_mask(attention_mask, cache_position) + + # embed positions + hidden_states = inputs_embeds + + position_embeddings = None + # create position embeddings to be shared across the decoder layers + if self.rotary_emb is not None: + position_embeddings = self.rotary_emb(hidden_states, position_ids) + + # decoder layers + all_hidden_states = () if output_hidden_states else None + all_self_attns = () if output_attentions else None + all_router_logits = () if output_router_logits else None + next_decoder_cache = None + + for decoder_layer in self.layers: + # Depending on the layer type we opt for 2D base attention mask (Mamba) or 4D causal mask (Attention) + layer_mask = mamba_mask if decoder_layer.layer_type == "mamba" else causal_mask + + if output_hidden_states: + all_hidden_states += (hidden_states,) + + layer_outputs = decoder_layer( + hidden_states, + attention_mask=layer_mask, + past_key_value=past_key_values, + output_attentions=output_attentions, + use_cache=use_cache, + cache_position=cache_position, + output_router_logits=output_router_logits, + position_embeddings=position_embeddings, + ) + + hidden_states = layer_outputs[0] + + if use_cache: + next_decoder_cache = layer_outputs[2 if output_attentions else 1] + + if output_attentions: + if layer_outputs[1] is not None: + # append attentions only of attention layers. Mamba layers return `None` as the attention weights + all_self_attns += (layer_outputs[1],) + + if output_router_logits: + if layer_outputs[-1] is not None: + # append router logits only of expert layers. Regular MLP layers return `None` as the router logits + all_router_logits += (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 = next_decoder_cache if use_cache else None + + return MoeModelOutputWithPast( + last_hidden_state=hidden_states, + past_key_values=next_cache, + hidden_states=all_hidden_states, + attentions=all_self_attns, + router_logits=all_router_logits, + ) + + def _update_mamba_mask(self, attention_mask, cache_position): + """ + No need for zeroing states when + 1. Cached forward + 2. Attending to all inputs + """ + mamba_mask = attention_mask + if cache_position[0] > 0 or (attention_mask is not None and torch.all(attention_mask == 1)): + mamba_mask = None + return mamba_mask + + +class GraniteMoeHybridForCausalLM(GraniteMoeSharedForCausalLM): + _tied_weights_keys = ["lm_head.weight"] + + def __init__(self, config: GraniteMoeHybridConfig): + super().__init__(config) + self.model = GraniteMoeHybridModel(config) + # Initialize weights and apply final processing + self.post_init() + + def prepare_inputs_for_generation( + self, + input_ids, + past_key_values=None, + attention_mask=None, + inputs_embeds=None, + cache_position=None, + position_ids=None, + use_cache=True, + **kwargs, + ): + # Overwritten -- has a unique cache type, `HybridMambaAttentionDynamicCache` + + empty_past_kv = past_key_values is None + + # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens + # Exception 1: when passing input_embeds, input_ids may be missing entries + # Exception 2: some generation methods do special slicing of input_ids, so we don't need to do it here + # Exception 3: with synced GPUs cache_position may go out of bounds, but we only want dummy token in that case. + # (we can't check exception 3 while compiling) + if not empty_past_kv: + if ( + inputs_embeds is not None # Exception 1 + or cache_position[-1] >= input_ids.shape[1] # Exception 3 + ): + input_ids = input_ids[:, -cache_position.shape[0] :] + elif input_ids.shape[1] != cache_position.shape[0]: # Default case (the "else", a no op, is Exception 2) + input_ids = input_ids[:, cache_position] + else: + past_key_values = HybridMambaAttentionDynamicCache( + self.config, input_ids.shape[0], self.dtype, device=self.device + ) + + 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.masked_fill_(attention_mask == 0, 1) + if not empty_past_kv: + position_ids = position_ids[:, -input_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 empty_past_kv: + model_inputs = {"inputs_embeds": inputs_embeds} + else: + model_inputs = {"input_ids": input_ids.contiguous()} # `contiguous()` needed for compilation use cases + + model_inputs.update( + { + "position_ids": position_ids, + "past_key_values": past_key_values, + "use_cache": use_cache, + "attention_mask": attention_mask, + "cache_position": cache_position, + } + ) + return model_inputs + + def _supports_default_dynamic_cache(self) -> bool: + """ + Function overwritten as this class uses its own `HybridMambaAttentionDynamicCache` + and do not need to initialize the Cache in advance in order to save memory + (because no back and forth `to_legacy_cache` and `from_legacy_cache` will be performed + for `HybridMambaAttentionDynamicCache`). + """ + return False + + +__all__ = ["GraniteMoeHybridForCausalLM", "GraniteMoeHybridModel", "GraniteMoeHybridPreTrainedModel"] diff --git a/src/transformers/models/granitemoeshared/configuration_granitemoeshared.py b/src/transformers/models/granitemoeshared/configuration_granitemoeshared.py index 32b55c69f37..2f81cad8932 100644 --- a/src/transformers/models/granitemoeshared/configuration_granitemoeshared.py +++ b/src/transformers/models/granitemoeshared/configuration_granitemoeshared.py @@ -169,6 +169,8 @@ class GraniteMoeSharedConfig(PretrainedConfig): self.use_cache = use_cache self.rope_theta = rope_theta self.rope_scaling = rope_scaling + # this model has rope embedding type, hardcoded for BC + self.position_embedding_type = "rope" self.attention_bias = attention_bias self.attention_dropout = attention_dropout diff --git a/src/transformers/models/granitemoeshared/modeling_granitemoeshared.py b/src/transformers/models/granitemoeshared/modeling_granitemoeshared.py index 251962c979e..21233195e25 100644 --- a/src/transformers/models/granitemoeshared/modeling_granitemoeshared.py +++ b/src/transformers/models/granitemoeshared/modeling_granitemoeshared.py @@ -19,7 +19,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -from typing import List, Optional, Tuple, Union +from typing import Callable, List, Optional, Tuple, Union import torch import torch.nn.functional as F @@ -29,10 +29,10 @@ from ...activations import ACT2FN from ...cache_utils import Cache, DynamicCache from ...generation import GenerationMixin from ...modeling_attn_mask_utils import AttentionMaskConverter -from ...modeling_flash_attention_utils import _flash_attention_forward, flash_attn_supports_top_left_mask +from ...modeling_layers import GradientCheckpointingLayer from ...modeling_outputs import BaseModelOutputWithPast, MoeCausalLMOutputWithPast, MoeModelOutputWithPast from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update -from ...modeling_utils import PreTrainedModel +from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel from ...utils import ( add_start_docstrings, add_start_docstrings_to_model_forward, @@ -300,6 +300,33 @@ def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor: return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim) +def eager_attention_forward( + module: nn.Module, + query: torch.Tensor, + key: torch.Tensor, + value: torch.Tensor, + attention_mask: Optional[torch.Tensor], + scaling: float, + dropout: float = 0.0, + **kwargs, +): + key_states = repeat_kv(key, module.num_key_value_groups) + value_states = repeat_kv(value, module.num_key_value_groups) + + attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling + if attention_mask is not None: + causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] + attn_weights = attn_weights + causal_mask + + # upcast attention to fp32 + attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype) + attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training) + attn_output = torch.matmul(attn_weights, value_states) + attn_output = attn_output.transpose(1, 2).contiguous() + + return attn_output, attn_weights + + # copied from transformers.models.granite.modeling_granite.GraniteAttention with Granite->GraniteMoeShared # no longer copied after attention refactors class GraniteMoeSharedAttention(nn.Module): @@ -343,10 +370,9 @@ class GraniteMoeSharedAttention(nn.Module): attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_value: Optional[Cache] = None, - output_attentions: bool = False, use_cache: bool = False, cache_position: Optional[torch.LongTensor] = None, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # None or rope embeddings **kwargs, ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: bsz, q_len, _ = hidden_states.size() @@ -359,262 +385,48 @@ class GraniteMoeSharedAttention(nn.Module): key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - cos, sin = position_embeddings - query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) + cos, sin = position_embeddings if position_embeddings is not None else (None, None) + if position_embeddings is not None: + query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) if past_key_value is not None: # sin and cos are specific to RoPE models; cache_position needed for the static cache cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - key_states = repeat_kv(key_states, self.num_key_value_groups) - value_states = repeat_kv(value_states, self.num_key_value_groups) - - attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) * self.scaling - - if attention_mask is not None: # no matter the length, we just slice it - causal_mask = attention_mask[:, :, :, : key_states.shape[-2]] - attn_weights = attn_weights + causal_mask - - # upcast attention to fp32 - attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype) - attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training) - attn_output = torch.matmul(attn_weights, value_states) - - if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim): - raise ValueError( - f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is" - f" {attn_output.size()}" - ) - - attn_output = attn_output.transpose(1, 2).contiguous() - - attn_output = attn_output.view(bsz, q_len, -1) - attn_output = self.o_proj(attn_output) - - if not output_attentions: - attn_weights = None - - return attn_output, attn_weights, past_key_value - - -# NO LONGER EXIST Copied from transformers.models.granite.modeling_granite.GraniteFlashAttention2 with Granite->GraniteMoeShared -# TODO cyril: modular -class GraniteMoeSharedFlashAttention2(GraniteMoeSharedAttention): - """ - GraniteMoeShared flash attention module. This module inherits from `GraniteMoeSharedAttention` as the weights of the module stays - untouched. The only required change would be on the forward pass where it needs to correctly call the public API of - flash attention and deal with padding tokens in case the input contains any of them. - """ - - def __init__(self, *args, **kwargs): - super().__init__(*args, **kwargs) - - # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1. - # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignment, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0. - # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left). - self._flash_attn_uses_top_left_mask = flash_attn_supports_top_left_mask() - - def forward( - self, - hidden_states: torch.Tensor, - attention_mask: Optional[torch.LongTensor] = None, - position_ids: Optional[torch.LongTensor] = None, - past_key_value: Optional[Cache] = None, - output_attentions: bool = False, - use_cache: bool = False, - cache_position: Optional[torch.LongTensor] = None, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC - ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: - output_attentions = False - - bsz, q_len, _ = hidden_states.size() - - query_states = self.q_proj(hidden_states) - key_states = self.k_proj(hidden_states) - value_states = self.v_proj(hidden_states) - - # Flash attention requires the input to have the shape - # batch_size x seq_length x head_dim x hidden_dim - # therefore we just need to keep the original shape - query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) - key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - - cos, sin = position_embeddings - query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) - - if past_key_value is not None: - # sin and cos are specific to RoPE models; cache_position needed for the static cache - cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} - key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - - # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache - # to be able to avoid many of these transpose/reshape/view. - query_states = query_states.transpose(1, 2) - key_states = key_states.transpose(1, 2) - value_states = value_states.transpose(1, 2) - - dropout_rate = self.attention_dropout if self.training else 0.0 - - # In PEFT, usually we cast the layer norms in float32 for training stability reasons - # therefore the input hidden states gets silently casted in float32. Hence, we need - # cast them back in the correct dtype just to be sure everything works as expected. - # This might slowdown training & inference so it is recommended to not cast the LayerNorms - # in fp32. (GraniteMoeSharedRMSNorm handles it correctly) - - input_dtype = query_states.dtype - if input_dtype == torch.float32: - if torch.is_autocast_enabled(): - target_dtype = torch.get_autocast_gpu_dtype() - # Handle the case where the model is quantized - elif hasattr(self.config, "_pre_quantization_dtype"): - target_dtype = self.config._pre_quantization_dtype + attention_interface: Callable = eager_attention_forward + if self.config._attn_implementation != "eager": + if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False): + logger.warning_once( + "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to " + 'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' + ) else: - target_dtype = self.q_proj.weight.dtype + attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation] - logger.warning_once( - f"The input hidden states seems to be silently casted in float32, this might be related to" - f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in" - f" {target_dtype}." - ) - - query_states = query_states.to(target_dtype) - key_states = key_states.to(target_dtype) - value_states = value_states.to(target_dtype) - - attn_output = _flash_attention_forward( + attn_output, attn_weights = attention_interface( + self, query_states, key_states, value_states, attention_mask, - q_len, - position_ids=position_ids, - dropout=dropout_rate, - softmax_scale=self.scaling, - sliding_window=getattr(self, "sliding_window", None), - use_top_left_mask=self._flash_attn_uses_top_left_mask, - is_causal=self.is_causal, + dropout=0.0 if not self.training else self.attention_dropout, + scaling=self.scaling, + **kwargs, ) - attn_output = attn_output.reshape(bsz, q_len, -1).contiguous() + attn_output = attn_output.view(bsz, q_len, -1) attn_output = self.o_proj(attn_output) - if not output_attentions: - attn_weights = None - return attn_output, attn_weights, past_key_value -# NO LONGER EXIST Copied from transformers.models.granite.modeling_granite.GraniteSdpaAttention with Granite->GraniteMoeShared -# TODO cyril: modular -class GraniteMoeSharedSdpaAttention(GraniteMoeSharedAttention): - """ - GraniteMoeShared attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from - `GraniteMoeSharedAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to - SDPA API. - """ - - # Adapted from GraniteMoeSharedAttention.forward - def forward( - self, - hidden_states: torch.Tensor, - attention_mask: Optional[torch.Tensor] = None, - position_ids: Optional[torch.LongTensor] = None, - past_key_value: Optional[Cache] = None, - output_attentions: bool = False, - use_cache: bool = False, - cache_position: Optional[torch.LongTensor] = None, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC - **kwargs, - ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]: - if output_attentions: - # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented. - logger.warning_once( - "GraniteMoeSharedModel is using GraniteMoeSharedSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, " - 'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.' - ) - return super().forward( - hidden_states=hidden_states, - attention_mask=attention_mask, - position_ids=position_ids, - past_key_value=past_key_value, - output_attentions=output_attentions, - use_cache=use_cache, - cache_position=cache_position, - position_embeddings=position_embeddings, - ) - - bsz, q_len, _ = hidden_states.size() - - query_states = self.q_proj(hidden_states) - key_states = self.k_proj(hidden_states) - value_states = self.v_proj(hidden_states) - - query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2) - key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2) - - cos, sin = position_embeddings - query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) - - if past_key_value is not None: - # sin and cos are specific to RoPE models; cache_position needed for the static cache - cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} - key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs) - - key_states = repeat_kv(key_states, self.num_key_value_groups) - value_states = repeat_kv(value_states, self.num_key_value_groups) - - causal_mask = attention_mask - if attention_mask is not None: - causal_mask = causal_mask[:, :, :, : key_states.shape[-2]] - - # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask, - # Reference: https://github.com/pytorch/pytorch/issues/112577. - if query_states.device.type == "cuda" and causal_mask is not None: - query_states = query_states.contiguous() - key_states = key_states.contiguous() - value_states = value_states.contiguous() - - # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment - # in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling. - is_causal = True if causal_mask is None and q_len > 1 else False - - attn_output = torch.nn.functional.scaled_dot_product_attention( - query_states, - key_states, - value_states, - attn_mask=causal_mask, - dropout_p=self.attention_dropout if self.training else 0.0, - is_causal=is_causal, - scale=self.scaling, - ) - - attn_output = attn_output.transpose(1, 2).contiguous() - attn_output = attn_output.view(bsz, q_len, -1) - - attn_output = self.o_proj(attn_output) - - return attn_output, None, past_key_value - - -GRANITEMOESHARED_ATTENTION_CLASSES = { - "eager": GraniteMoeSharedAttention, - "flash_attention_2": GraniteMoeSharedFlashAttention2, - "sdpa": GraniteMoeSharedSdpaAttention, -} - - -class GraniteMoeSharedDecoderLayer(nn.Module): +class GraniteMoeSharedDecoderLayer(GradientCheckpointingLayer): def __init__(self, config: GraniteMoeSharedConfig, layer_idx: int): super().__init__() self.hidden_size = config.hidden_size - self.self_attn = GRANITEMOESHARED_ATTENTION_CLASSES[config._attn_implementation]( - config=config, layer_idx=layer_idx - ) - + self.self_attn = GraniteMoeSharedAttention(config=config, layer_idx=layer_idx) self.block_sparse_moe = GraniteMoeSharedMoE(config) self.input_layernorm = GraniteMoeSharedRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.post_attention_layernorm = GraniteMoeSharedRMSNorm(config.hidden_size, eps=config.rms_norm_eps) @@ -632,7 +444,7 @@ class GraniteMoeSharedDecoderLayer(nn.Module): use_cache: Optional[bool] = False, cache_position: Optional[torch.LongTensor] = None, output_router_logits: Optional[bool] = False, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, **kwargs, ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: """ @@ -739,13 +551,12 @@ class GraniteMoeSharedPreTrainedModel(PreTrainedModel): _supports_static_cache = False # MoE models don't work with torch.compile (`torch.where(condition)` not supported) def _init_weights(self, module): - std = self.config.initializer_range if isinstance(module, nn.Linear): - module.weight.data.normal_(mean=0.0, std=std) + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) if module.bias is not None: module.bias.data.zero_() elif isinstance(module, nn.Embedding): - module.weight.data.normal_(mean=0.0, std=std) + module.weight.data.normal_(mean=0.0, std=self.config.initializer_range) if module.padding_idx is not None: module.weight.data[module.padding_idx].zero_() elif isinstance(module, GraniteMoeSharedRMSNorm): @@ -893,8 +704,8 @@ class GraniteMoeSharedModel(GraniteMoeSharedPreTrainedModel): self.max_position_embeddings = config.max_position_embeddings self.rope_theta = config.rope_theta - # rope - self.rotary_emb = GraniteMoeSharedRotaryEmbedding(config) + self.position_embedding_type = config.position_embedding_type + self.rotary_emb = GraniteMoeSharedRotaryEmbedding(config) if self.position_embedding_type == "rope" else None # Initialize weights and apply final processing self.post_init() @@ -965,8 +776,10 @@ class GraniteMoeSharedModel(GraniteMoeSharedPreTrainedModel): # embed positions hidden_states = inputs_embeds + position_embeddings = None # create position embeddings to be shared across the decoder layers - position_embeddings = self.rotary_emb(hidden_states, position_ids) + if self.rotary_emb is not None: + position_embeddings = self.rotary_emb(hidden_states, position_ids) # decoder layers all_hidden_states = () if output_hidden_states else None @@ -978,31 +791,17 @@ class GraniteMoeSharedModel(GraniteMoeSharedPreTrainedModel): 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, - output_router_logits, - position_embeddings, - ) - 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, - output_router_logits=output_router_logits, - position_embeddings=position_embeddings, - ) + 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, + output_router_logits=output_router_logits, + position_embeddings=position_embeddings, + ) hidden_states = layer_outputs[0] @@ -1291,6 +1090,7 @@ class GraniteMoeSharedForCausalLM(GraniteMoeSharedPreTrainedModel, GenerationMix output_router_logits: Optional[bool] = None, return_dict: Optional[bool] = None, cache_position: Optional[torch.LongTensor] = None, + logits_to_keep: Union[int, torch.Tensor] = 0, **kwargs, ) -> Union[Tuple, MoeCausalLMOutputWithPast]: r""" @@ -1299,6 +1099,13 @@ class GraniteMoeSharedForCausalLM(GraniteMoeSharedPreTrainedModel, GenerationMix 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]`. + logits_to_keep (`int` or `torch.Tensor`, *optional*): + 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: Example: @@ -1341,8 +1148,10 @@ class GraniteMoeSharedForCausalLM(GraniteMoeSharedPreTrainedModel, GenerationMix cache_position=cache_position, ) + # Only compute necessary logits hidden_states = outputs[0] - logits = self.lm_head(hidden_states) + slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep + logits = self.lm_head(hidden_states[:, slice_indices, :]) logits = logits / self.config.logits_scaling loss = None diff --git a/src/transformers/models/granitemoeshared/modular_granitemoeshared.py b/src/transformers/models/granitemoeshared/modular_granitemoeshared.py index d275eb5f455..8aa2d64aa56 100644 --- a/src/transformers/models/granitemoeshared/modular_granitemoeshared.py +++ b/src/transformers/models/granitemoeshared/modular_granitemoeshared.py @@ -16,7 +16,6 @@ from typing import Optional, Tuple import torch -import torch.utils.checkpoint from torch import nn from ...activations import ACT2FN @@ -78,7 +77,7 @@ class GraniteMoeSharedDecoderLayer(GraniteMoeDecoderLayer): use_cache: Optional[bool] = False, cache_position: Optional[torch.LongTensor] = None, output_router_logits: Optional[bool] = False, - position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, # necessary, but kept here for BC + position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None, **kwargs, ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: """ diff --git a/tests/generation/test_utils.py b/tests/generation/test_utils.py index 46a1b90001b..bb4965cb767 100644 --- a/tests/generation/test_utils.py +++ b/tests/generation/test_utils.py @@ -2491,6 +2491,7 @@ class GenerationTesterMixin: "bamba", "ctrl", "fsmt", + "granitemoehybrid", "gptbigcode", "mega", "reformer", diff --git a/tests/models/bamba/test_modeling_bamba.py b/tests/models/bamba/test_modeling_bamba.py index 2159f427bd2..76d593343fd 100644 --- a/tests/models/bamba/test_modeling_bamba.py +++ b/tests/models/bamba/test_modeling_bamba.py @@ -47,6 +47,11 @@ if is_torch_available(): class BambaModelTester: + config_class = BambaConfig + if is_torch_available(): + model_class = BambaModel + for_causal_lm_class = BambaForCausalLM + def __init__( self, parent, @@ -118,6 +123,7 @@ class BambaModelTester: if self.use_labels: token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels) + self._update_layer_configs() config = self.get_config() return config, input_ids, input_mask, token_labels @@ -133,10 +139,12 @@ class BambaModelTester: inputs_dict = {"input_ids": input_ids, "attention_mask": input_mask} return config, inputs_dict - def get_config(self): + def _update_layer_configs(self): + """Configures hidden layers and attn layer indices if they are not set.""" # Fix for SDPA tests, force at least 4 layers if self.num_hidden_layers < 4: self.num_hidden_layers = 4 + if self.attn_layer_indices is None: d = [x for x in range(2, self.num_hidden_layers) if self.num_hidden_layers % x == 0] if len(d) == 0: @@ -144,7 +152,8 @@ class BambaModelTester: d = d[-1] # get the largest divisor self.attn_layer_indices = [x + 1 for x in range(0, self.num_hidden_layers, d)] - return BambaConfig( + def get_config(self, **kwargs): + return self.config_class( vocab_size=self.vocab_size, hidden_size=self.hidden_size, num_hidden_layers=self.num_hidden_layers, @@ -164,6 +173,7 @@ class BambaModelTester: mamba_d_conv=self.mamba_d_conv, mamba_expand=self.mamba_expand, mamba_chunk_size=self.mamba_chunk_size, + **kwargs, ) def create_and_check_model( @@ -173,7 +183,7 @@ class BambaModelTester: input_mask, token_labels, ): - model = BambaModel(config=config) + model = self.model_class(config=config) model.to(torch_device) model.eval() result = model(input_ids, attention_mask=input_mask) @@ -187,7 +197,7 @@ class BambaModelTester: input_mask, token_labels, ): - model = BambaForCausalLM(config=config) + model = self.for_causal_lm_class(config=config) model.to(torch_device) model.eval() result = model(input_ids, attention_mask=input_mask, labels=token_labels) @@ -205,7 +215,7 @@ class BambaModelTester: ): # config.is_decoder = True # config.add_cross_attention = True - model = BambaForCausalLM(config=config) + model = self.for_causal_lm_class(config=config) model.to(torch_device) model.eval() @@ -258,6 +268,7 @@ class BambaModelTester: @require_torch class BambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase): + model_tester_class = BambaModelTester all_model_classes = (BambaModel, BambaForCausalLM) if is_torch_available() else () pipeline_model_mapping = ( { @@ -276,8 +287,8 @@ class BambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi model_split_percents = [0.5, 0.7, 0.8] def setUp(self): - self.model_tester = BambaModelTester(self) - self.config_tester = ConfigTester(self, config_class=BambaConfig, hidden_size=64) + self.model_tester = self.model_tester_class(self) + self.config_tester = ConfigTester(self, config_class=self.model_tester.config_class, hidden_size=64) def test_config(self): self.config_tester.run_common_tests() diff --git a/tests/models/granitemoehybrid/__init__.py b/tests/models/granitemoehybrid/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/models/granitemoehybrid/test_modeling_granitemoehybrid.py b/tests/models/granitemoehybrid/test_modeling_granitemoehybrid.py new file mode 100644 index 00000000000..3f4f45017d0 --- /dev/null +++ b/tests/models/granitemoehybrid/test_modeling_granitemoehybrid.py @@ -0,0 +1,164 @@ +# coding=utf-8 +# Copyright 2024 The HuggingFace Inc. team. All rights reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +"""Testing suite for the PyTorch GraniteMoeHybrid model.""" + +import unittest + +import pytest + +from transformers import ( + AutoTokenizer, + GraniteMoeHybridConfig, + is_torch_available, +) +from transformers.testing_utils import ( + require_torch, + require_torch_gpu, + slow, + torch_device, +) + +from ...generation.test_utils import GenerationTesterMixin +from ...models.bamba.test_modeling_bamba import BambaModelTest, BambaModelTester + + +if is_torch_available(): + import torch + + from transformers import ( + GraniteMoeHybridForCausalLM, + GraniteMoeHybridModel, + ) + + +class GraniteMoeHybridModelTester(BambaModelTester): + config_class = GraniteMoeHybridConfig + if is_torch_available(): + model_class = GraniteMoeHybridModel + for_causal_lm_class = GraniteMoeHybridForCausalLM + + def __init__( + self, + parent, + use_cache=False, + shared_intermediate_size=174, + layer_types=None, + ): + super().__init__(parent) + self.shared_intermediate_size = shared_intermediate_size + self.layer_types = layer_types + self.use_cache = use_cache + + def _update_layer_configs(self): + super()._update_layer_configs() + # GraniteMoeHybrid uses layer_types instead of attn_layer_indices + self.layer_types = ["mamba"] * self.num_hidden_layers + for idx in self.attn_layer_indices: + self.layer_types[idx] = "attention" + + def get_config(self): + return super().get_config( + shared_intermediate_size=self.shared_intermediate_size, + layer_types=self.layer_types, + ) + + +@require_torch +class GraniteMoeHybridModelTest(BambaModelTest, GenerationTesterMixin, unittest.TestCase): + model_tester_class = GraniteMoeHybridModelTester + all_model_classes = ( + ( + GraniteMoeHybridModel, + GraniteMoeHybridForCausalLM, + ) + if is_torch_available() + else () + ) + pipeline_model_mapping = ( + { + "feature-extraction": GraniteMoeHybridModel, + "text-generation": GraniteMoeHybridForCausalLM, + } + if is_torch_available() + else {} + ) + + def test_config_requires_mamba_or_attention_layers(self): + """Ensure we can't create a config with disallowed layers.""" + with pytest.raises(ValueError): + GraniteMoeHybridConfig(layer_types=["not allowed!"]) + + +# TODO (@alex-jw-brooks) - update this once the model(s) are out +@unittest.skip(reason="GraniteMoeHybrid models are not yet released") +@require_torch_gpu +class GraniteMoeHybridIntegrationTest(unittest.TestCase): + # This variable is used to determine which CUDA device are we using for our runners (A10 or T4) + # Depending on the hardware we get different logits / generations + cuda_compute_capability_major_version = None + + @classmethod + def setUpClass(cls): + if is_torch_available() and torch.cuda.is_available(): + # 8 is for A100 / A10 and 7 for T4 + cls.cuda_compute_capability_major_version = torch.cuda.get_device_capability()[0] + + @slow + def test_model_logits(self): + input_ids = [31390, 631, 4162, 30, 322, 25342, 432, 1875, 43826, 10066, 688, 225] + + model = GraniteMoeHybridForCausalLM.from_pretrained("ibm-granite/granite-4.0-tiny", device_map="auto") + + with torch.no_grad(): + out = model(torch.tensor([input_ids]).to(torch_device)) + + # fmt: off + # Expected mean on dim = -1 + EXPECTED_MEAN = torch.tensor([ + [-2.9711, -2.2554, -1.0814, -1.6123, -0.8780, -1.0685, -0.6368, -1.9732, -3.3548, -2.6895, -2.3062, -2.6338] + ]) + + torch.testing.assert_close(EXPECTED_MEAN.to(torch_device), out.logits.float().mean(-1), rtol=1e-2, atol=1e-2) + + # slicing logits[0, 0, 0:15] + EXPECTED_SLICE = torch.tensor([ + [4.0662, 5.9547, 3.5803, 3.1306, 4.3211, 3.8902, 4.6438, 8.5434, 7.5865, 5.1623, 5.2240, 9.2982, 5.9094, 6.8834, 5.7551], + ]) + # fmt: on + + self.assertTrue( + torch.allclose( + EXPECTED_SLICE.to(torch_device), + out.logits[0, 0, :15].float(), + atol=1e-3, + rtol=1e-3, + ) + ) + + @slow + def test_model_generation(self): + EXPECTED_TEXT_COMPLETION = ( + "Simply put, the theory of relativity states that 1) time is relative, and 2) space is relative. The first" + ) + prompt = "Simply put, the theory of relativity states that " + tokenizer = AutoTokenizer.from_pretrained("ibm-granite/granite-4.0-tiny") + model = GraniteMoeHybridForCausalLM.from_pretrained("ibm-granite/granite-4.0-tiny", device_map="auto") + model_inputs = tokenizer(prompt, return_tensors="pt").to(model.device) + + # greedy generation outputs + generated_ids = model.generate(**model_inputs, max_new_tokens=16, do_sample=False) + text = tokenizer.decode(generated_ids[0], skip_special_tokens=True) + + self.assertEqual(EXPECTED_TEXT_COMPLETION, text) diff --git a/utils/check_config_docstrings.py b/utils/check_config_docstrings.py index 7de82aff81a..4a5d0395a20 100644 --- a/utils/check_config_docstrings.py +++ b/utils/check_config_docstrings.py @@ -47,6 +47,7 @@ CONFIG_CLASSES_TO_IGNORE_FOR_DOCSTRING_CHECKPOINT_CHECK = { "LlamaConfig", "GraniteConfig", "GraniteMoeConfig", + "GraniteMoeHybridConfig", "Qwen3MoeConfig", "GraniteSpeechConfig", }