# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # This file was automatically generated from src/transformers/models/interns2_preview/modular_interns2_preview.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_interns2_preview.py file directly. One of our CI enforces this. # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # Copyright 2025 The Qwen Team 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 transformers.configuration_utils import PreTrainedConfig, layer_type_validation from transformers.modeling_rope_utils import RopeParameters class InternS2PreviewVisionConfig(PreTrainedConfig): model_type = "intern_s2_preview" base_config_key = "vision_config" def __init__( self, depth=27, hidden_size=1152, hidden_act="gelu_pytorch_tanh", intermediate_size=4304, num_heads=16, in_channels=3, patch_size=16, spatial_merge_size=2, temporal_patch_size=2, out_hidden_size=3584, num_position_embeddings=2304, initializer_range=0.02, **kwargs, ): super().__init__(**kwargs) self.depth = depth self.hidden_size = hidden_size self.hidden_act = hidden_act self.intermediate_size = intermediate_size self.num_heads = num_heads self.in_channels = in_channels self.patch_size = patch_size self.spatial_merge_size = spatial_merge_size self.temporal_patch_size = temporal_patch_size self.out_hidden_size = out_hidden_size self.num_position_embeddings = num_position_embeddings self.initializer_range = initializer_range class InternS2PreviewTextConfig(PreTrainedConfig): r""" This is the configuration class to store the configuration of a [`InternS2PreviewTextModel`]. It is used to instantiate a Qwen3.5-MoE model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of Qwen3.5-35B-A3B-Instruct [Qwen/Qwen3.5-35B-A3B-Instruct](https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Instruct). 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 248320): Vocabulary size of the model. Defines the number of different tokens that can be represented by the `inputs_ids`. hidden_size (`int`, *optional*, defaults to 2048): Dimension of the hidden representations. num_hidden_layers (`int`, *optional*, defaults to 40): Number of hidden layers in the Transformer encoder. num_attention_heads (`int`, *optional*, defaults to 16): Number of attention heads for each attention layer in the Transformer encoder. num_key_value_heads (`int`, *optional*, defaults to 2): 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 `32`. hidden_act (`str`, *optional*, defaults to `"silu"`): The non-linear activation function in the decoder. max_position_embeddings (`int`, *optional*, defaults to 32768): 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`. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether the model's input and output word embeddings should be tied. rope_parameters (`RopeParameters`, *optional*): Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain a value for `rope_theta` and optionally parameters used for scaling in case you want to use RoPE with longer `max_position_embeddings`. 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. head_dim (`int`, *optional*, defaults to 256): Projection weights dimension in multi-head attention. linear_conv_kernel_dim (`int`, *optional*, defaults to 4): Kernel size of the convolution used in linear attention layers. linear_key_head_dim (`int`, *optional*, defaults to 128): Dimension of each key head in linear attention. linear_value_head_dim (`int`, *optional*, defaults to 128): Dimension of each value head in linear attention. linear_num_key_heads (`int`, *optional*, defaults to 16): Number of key heads used in linear attention layers. linear_num_value_heads (`int`, *optional*, defaults to 32): Number of value heads used in linear attention layers. moe_intermediate_size (`int`, *optional*, defaults to 512): Intermediate size of the routed expert. shared_expert_intermediate_size (`int`, *optional*, defaults to 512): Intermediate size of the shared expert. num_experts_per_tok (`int`, *optional*, defaults to 8): Number of selected experts. num_experts (`int`, *optional*, defaults to 256): Number of routed experts. output_router_logits (`bool`, *optional*, defaults to `False`): Whether or not the router logits should be returned by the model. Enabling this will also allow the model to output the auxiliary loss, including load balancing loss and router z-loss. router_aux_loss_coef (`float`, *optional*, defaults to 0.001): The aux loss factor for the total loss. layer_types (`list[str]`, *optional*): Types of each layer (attention or linear). pad_token_id (`int`, *optional*): Padding token id. bos_token_id (`int`, *optional*): Beginning of stream token id. eos_token_id (`int`, *optional*): End of stream token id. ```python >>> from transformers import InternS2PreviewTextModel, InternS2PreviewTextConfig >>> # Initializing a Qwen3.5-MoE style configuration >>> configuration = InternS2PreviewTextConfig() >>> # Initializing a model from the Qwen3.5-35B-A3B style configuration >>> model = InternS2PreviewTextModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` """ # NOTE: `model_type` is kept as `qwen3_5_moe_text` because transformers hardcodes weight-renaming logic keyed # on model_type (e.g. `model_dtype`); reusing the parent's value ensures correct weight loading via # `AutoModelForCausalLM.from_pretrained`. model_type = "qwen3_5_moe_text" keys_to_ignore_at_inference = ["past_key_values"] base_model_tp_plan = { "layers.*.self_attn.q_proj": "colwise", "layers.*.self_attn.k_proj": "colwise", "layers.*.self_attn.v_proj": "colwise", "layers.*.self_attn.o_proj": "rowwise", "layers.*.mlp.experts.gate_up_proj": "packed_colwise", "layers.*.mlp.experts.down_proj": "rowwise", "layers.*.mlp.shared_expert.gate_proj": "colwise", "layers.*.mlp.shared_expert.up_proj": "colwise", "layers.*.mlp.shared_expert.down_proj": "rowwise", } base_model_pp_plan = { "embed_tokens": (["input_ids"], ["inputs_embeds"]), "layers": (["hidden_states", "attention_mask"], ["hidden_states"]), "norm": (["hidden_states"], ["hidden_states"]), } base_config_key = "text_config" def __init__( self, vocab_size=248320, hidden_size=2048, num_hidden_layers=40, num_attention_heads=16, num_key_value_heads=2, hidden_act="silu", max_position_embeddings=32768, initializer_range=0.02, rms_norm_eps=1e-6, use_cache=True, tie_word_embeddings=False, rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None, attention_bias=False, attention_dropout=0.0, head_dim=256, linear_conv_kernel_dim=4, linear_key_head_dim=128, linear_value_head_dim=128, linear_num_key_heads=16, linear_num_value_heads=32, moe_intermediate_size=512, shared_expert_intermediate_size=512, num_experts_per_tok=8, num_experts=256, output_router_logits=False, router_aux_loss_coef=0.001, layer_types=None, pad_token_id: int | None = None, bos_token_id: int | None = None, eos_token_id: int | None = None, **kwargs, ): kwargs["ignore_keys_at_rope_validation"] = {"mrope_section", "mrope_interleaved"} self.pad_token_id = pad_token_id self.bos_token_id = bos_token_id self.eos_token_id = eos_token_id self.tie_word_embeddings = tie_word_embeddings self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.num_hidden_layers = num_hidden_layers self.num_attention_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.attention_bias = attention_bias self.attention_dropout = attention_dropout self.head_dim = head_dim self.rope_parameters = rope_parameters kwargs.setdefault("partial_rotary_factor", 0.25) # assign default for BC self.layer_types = layer_types if self.layer_types is None: interval_pattern = kwargs.get("full_attention_interval", 4) self.layer_types = [ "linear_attention" if bool((i + 1) % interval_pattern) else "full_attention" for i in range(self.num_hidden_layers) ] layer_type_validation(self.layer_types, self.num_hidden_layers) # linear attention part self.linear_conv_kernel_dim = linear_conv_kernel_dim self.linear_key_head_dim = linear_key_head_dim self.linear_value_head_dim = linear_value_head_dim self.linear_num_key_heads = linear_num_key_heads self.linear_num_value_heads = linear_num_value_heads self.moe_intermediate_size = moe_intermediate_size self.shared_expert_intermediate_size = shared_expert_intermediate_size self.num_experts_per_tok = num_experts_per_tok self.num_experts = num_experts self.output_router_logits = output_router_logits self.router_aux_loss_coef = router_aux_loss_coef super().__init__(**kwargs) class InternS2PreviewTimeSeriesConfig(PreTrainedConfig): r""" This is the configuration class to store the configuration of a [`InternS2PreviewTimeSeriesModel`]. It is used to instantiate a InternS2PreviewTimeSeries model according to the specified arguments, defining the model architecture. Args: ts_adapt_in_dim (`int`, *optional*, defaults to 256): The input dimension of the time series adapter. ts_adapt_out_dim (`int`, *optional*, defaults to 1024): The output dimension of the time series adapter. ts_hidden_dim (`int`, *optional*, defaults to 1024): The hidden dimension of the time series model. ts_cnn_channels (`list[int]`, *optional*, defaults to [1, 32, 64, 128, 128]): The channels of the time series CNN. ts_cnn_kernel_sizes (`list[int]`, *optional*, defaults to [3, 5, 5, 5]): The kernel sizes of the time series CNN. ts_cnn_strides (`list[int]`, *optional*, defaults to [2, 4, 4, 5]): The strides of the time series CNN. ts_cnn_paddings (`list[int]`, *optional*, defaults to [1, 2, 2, 2]): The paddings of the time series CNN. ts_concat_subsampling_in_channels (`int`, *optional*, defaults to 128): The input channels of the time series concat subsampling. ts_concat_subsampling_concat_size (`int`, *optional*, defaults to 2): The concat size of the time series concat subsampling. **super_kwargs: Additional keyword arguments passed along to the base class `WhisperConfig`. """ model_type = "interns2_preview_time_series" base_config_key = "ts_config" def __init__( self, activation_dropout: float = 0.0, activation_function: str = "gelu", attention_dropout: float = 0.0, d_model: int = 768, dropout: float = 0.0, encoder_attention_heads: int = 8, encoder_ffn_dim: int = 3072, encoder_layerdrop: float = 0.0, encoder_layers: int = 17, max_source_positions: int = 1500, num_mel_bins: int = 80, out_hidden_size: int = 2048, scale_embedding: bool = False, ts_adapt_in_dim: int = 256, ts_adapt_out_dim: int = 1024, ts_hidden_dim: int = 1024, **super_kwargs, ): super().__init__(**super_kwargs) self.auto_map = { "AutoConfig": "configuration_interns2_preview.InternS2PreviewTimeSeriesConfig", "AutoModel": "modeling_interns2_preview.InternS2PreviewTimeSeriesModel", } self.activation_dropout = activation_dropout self.activation_function = activation_function self.attention_dropout = attention_dropout self.d_model = d_model self.dropout = dropout self.encoder_attention_heads = encoder_attention_heads self.encoder_ffn_dim = encoder_ffn_dim self.encoder_layerdrop = encoder_layerdrop self.encoder_layers = encoder_layers self.max_source_positions = max_source_positions self.num_mel_bins = num_mel_bins self.out_hidden_size = out_hidden_size self.scale_embedding = scale_embedding self.ts_adapt_in_dim = ts_adapt_in_dim self.ts_adapt_out_dim = ts_adapt_out_dim self.ts_hidden_dim = ts_hidden_dim assert self.ts_adapt_out_dim == self.ts_hidden_dim, "ts_adapt_out_dim should be equal to ts_hidden_dim" class InternS2PreviewConfig(PreTrainedConfig): r""" This is the configuration class to store the configuration of a [`InternS2PreviewModel`]. It is used to instantiate a Qwen3.5-MoE model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of Qwen3.5-35B-A3B-Instruct [Qwen/Qwen3.5-35B-A3B-Instruct](https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Instruct). Configuration objects inherit from [`PreTrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PreTrainedConfig`] for more information. Args: text_config (`Union[PreTrainedConfig, dict]`, *optional*, defaults to `Qwen3_5TextConfig`): The config object or dictionary of the text backbone. vision_config (`Union[PreTrainedConfig, dict]`, *optional*, defaults to `Qwen3_5VisionConfig`): The config object or dictionary of the vision backbone. image_token_id (`int`, *optional*, defaults to 248056): The image token index to encode the image prompt. video_token_id (`int`, *optional*, defaults to 248057): The video token index to encode the image prompt. vision_start_token_id (`int`, *optional*, defaults to 248053): The start token index to encode the image prompt. vision_end_token_id (`int`, *optional*, defaults to 248054): The end token index to encode the image prompt. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether to tie the word embeddings. ```python >>> from transformers import InternS2PreviewForConditionalGeneration, InternS2PreviewConfig >>> # Initializing a Qwen3.5-MoE style configuration >>> configuration = InternS2PreviewConfig() >>> # Initializing a model from the Qwen3.5-35B-A3B style configuration >>> model = InternS2PreviewForConditionalGeneration(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```""" model_type = "intern_s2_preview" sub_configs = { "vision_config": InternS2PreviewVisionConfig, "text_config": InternS2PreviewTextConfig, "ts_config": InternS2PreviewTimeSeriesConfig, } keys_to_ignore_at_inference = ["past_key_values"] def __init__( self, text_config=None, vision_config=None, image_token_id=248056, video_token_id=248057, vision_start_token_id=248053, vision_end_token_id=248054, tie_word_embeddings=False, ts_config=None, ts_token_id=248093, ts_start_id=248091, ts_end_id=248092, **kwargs, ): if isinstance(ts_config, dict): self.ts_config = self.sub_configs["ts_config"](**ts_config) elif ts_config is None: self.ts_config = self.sub_configs["ts_config"]() self.ts_token_id = ts_token_id self.ts_start_id = ts_start_id self.ts_end_id = ts_end_id if isinstance(vision_config, dict): self.vision_config = self.sub_configs["vision_config"](**vision_config) elif vision_config is None: self.vision_config = self.sub_configs["vision_config"]() if isinstance(text_config, dict): self.text_config = self.sub_configs["text_config"](**text_config) elif text_config is None: self.text_config = self.sub_configs["text_config"]() self.image_token_id = image_token_id self.video_token_id = video_token_id self.vision_start_token_id = vision_start_token_id self.vision_end_token_id = vision_end_token_id self.tie_word_embeddings = tie_word_embeddings super().__init__(**kwargs) self.auto_map = { "AutoConfig": "configuration_interns2_preview.InternS2PreviewConfig", "AutoModelForCausalLM": "modeling_interns2_preview.InternS2PreviewForCausalLM", "AutoModel": "modeling_interns2_preview.InternS2PreviewModel", "AutoModelForImageTextToText": "modeling_interns2_preview.InternS2PreviewForConditionalGeneration", "AutoModelForMultimodalLM": "modeling_interns2_preview.InternS2PreviewForConditionalGeneration", } self.architectures = ["InternS2PreviewForConditionalGeneration"] __all__ = ["InternS2PreviewConfig", "InternS2PreviewTextConfig"]