# Copyright 2026 OpenMOSS 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. """Local transformer used by MossTTSRealtime for RVQ codebook decoding.""" from __future__ import annotations from typing import Optional, Union import torch import torch.nn as nn from transformers.activations import ACT2FN from transformers.cache_utils import Cache, StaticCache from transformers.generation import GenerationMixin from transformers.modeling_flash_attention_utils import FlashAttentionKwargs from transformers.modeling_layers import GradientCheckpointingLayer from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel from transformers.masking_utils import create_causal_mask from transformers.processing_utils import Unpack from transformers.loss.loss_utils import ForCausalLMLoss from transformers.utils import TransformersKwargs, logging from .configuration_mossttsrealtime import MossTTSRealtimeLocalTransformerConfig logger = logging.get_logger(__name__) class MossTTSRealtimeLocalTransformerRMSNorm(nn.Module): def __init__(self, hidden_size, eps=1e-6) -> None: super().__init__() self.weight = nn.Parameter(torch.ones(hidden_size)) self.variance_epsilon = eps def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: 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 MossTTSRealtimeLocalTransformerMLP(nn.Module): def __init__(self, config: MossTTSRealtimeLocalTransformerConfig): super().__init__() self.config = config self.hidden_size = config.hidden_size self.intermediate_size = config.intermediate_size self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False) self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False) self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False) self.act_fn = ACT2FN[config.hidden_act] def forward(self, x): down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x)) return down_proj def rotate_half(x): 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, unsqueeze_dim=1): 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: 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: Unpack[TransformersKwargs], ): 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 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 class MossTTSRealtimeLocalTransformerAttention(nn.Module): def __init__(self, config: MossTTSRealtimeLocalTransformerConfig, layer_idx: int): super().__init__() self.config = config self.layer_idx = layer_idx self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads) self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads self.scaling = self.head_dim**-0.5 self.attention_dropout = config.attention_dropout self.is_causal = True self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias) self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias) self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias) self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias) self.q_norm = MossTTSRealtimeLocalTransformerRMSNorm(self.head_dim, eps=config.rms_norm_eps) self.k_norm = MossTTSRealtimeLocalTransformerRMSNorm(self.head_dim, eps=config.rms_norm_eps) self.sliding_window = None def forward( self, hidden_states: torch.Tensor, position_embeddings: tuple[torch.Tensor, torch.Tensor], attention_mask: Optional[torch.Tensor], past_key_values: Optional[Cache] = None, cache_position: Optional[torch.LongTensor] = None, **kwargs: Unpack[FlashAttentionKwargs], ) -> tuple[torch.Tensor, Optional[torch.Tensor]]: input_shape = hidden_states.shape[:-1] hidden_shape = (*input_shape, -1, self.head_dim) query_states = self.q_norm(self.q_proj(hidden_states).view(hidden_shape)).transpose(1, 2) key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2) value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2) cos, sin = position_embeddings query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin) if past_key_values is not None: cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position} key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs) attention_interface = eager_attention_forward if self.config._attn_implementation != "eager": 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, sliding_window=self.sliding_window, **kwargs, ) attn_output = attn_output.reshape(*input_shape, -1).contiguous() attn_output = self.o_proj(attn_output) return attn_output, attn_weights class MossTTSRealtimeLocalTransformerDecoderLayer(GradientCheckpointingLayer): def __init__(self, config: MossTTSRealtimeLocalTransformerConfig, layer_idx: int): super().__init__() self.hidden_size = config.hidden_size self.self_attn = MossTTSRealtimeLocalTransformerAttention(config=config, layer_idx=layer_idx) self.mlp = MossTTSRealtimeLocalTransformerMLP(config) self.input_layernorm = MossTTSRealtimeLocalTransformerRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.post_attention_layernorm = MossTTSRealtimeLocalTransformerRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.attention_type = "full_attention" def forward( self, hidden_states: torch.Tensor, attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_values: Optional[Cache] = None, use_cache: Optional[bool] = False, cache_position: Optional[torch.LongTensor] = None, position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None, **kwargs: Unpack[TransformersKwargs], ) -> torch.Tensor: residual = hidden_states hidden_states = self.input_layernorm(hidden_states) hidden_states, _ = self.self_attn( hidden_states=hidden_states, attention_mask=attention_mask, position_ids=position_ids, past_key_values=past_key_values, use_cache=use_cache, cache_position=cache_position, position_embeddings=position_embeddings, **kwargs, ) hidden_states = residual + hidden_states residual = hidden_states hidden_states = self.post_attention_layernorm(hidden_states) hidden_states = self.mlp(hidden_states) hidden_states = residual + hidden_states return hidden_states class MossTTSRealtimeLocalTransformerPreTrainedModel(PreTrainedModel): config_class = MossTTSRealtimeLocalTransformerConfig config: MossTTSRealtimeLocalTransformerConfig base_model_prefix = "local_transformer" supports_gradient_checkpointing = True _no_split_modules = ["MossTTSRealtimeLocalTransformerDecoderLayer"] _skip_keys_device_placement = ["past_key_values"] _supports_sdpa = True _supports_flex_attn = True _supports_flash_attn = True _can_compile_fullgraph = True _supports_attention_backend = True _can_record_outputs = { "hidden_states": MossTTSRealtimeLocalTransformerDecoderLayer, "attentions": MossTTSRealtimeLocalTransformerAttention, } class MossTTSRealtimeLocalTransformerRotaryEmbedding(nn.Module): inv_freq: torch.Tensor def __init__(self, config: MossTTSRealtimeLocalTransformerConfig, device=None): super().__init__() self.config = config self.rope_type = getattr(config, "rope_type", "linear") self.max_seq_len_cached = config.max_position_embeddings self.original_max_seq_len = config.max_position_embeddings 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 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): 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) class MossTTSRealtimeLocalTransformer(MossTTSRealtimeLocalTransformerPreTrainedModel): def __init__(self, config: MossTTSRealtimeLocalTransformerConfig): super().__init__(config) self.padding_idx = config.pad_token_id self.embed_tokens = nn.ModuleList( [nn.Embedding(config.audio_vocab_size, config.hidden_size, config.audio_pad_token) for _ in range(config.rvq - 1)] ) self.layers = nn.ModuleList( [MossTTSRealtimeLocalTransformerDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)] ) self.norm = MossTTSRealtimeLocalTransformerRMSNorm(config.hidden_size, eps=config.rms_norm_eps) self.rotary_emb = MossTTSRealtimeLocalTransformerRotaryEmbedding(config=config) self.gradient_checkpointing = False self.has_sliding_layers = None self.post_init() def forward( self, input_ids: Optional[torch.LongTensor] = None, backbone_last_hidden_state: Optional[torch.FloatTensor] = None, attention_mask: Optional[torch.Tensor] = None, position_ids: Optional[torch.LongTensor] = None, past_key_values: Optional[Cache] = None, inputs_embeds: Optional[torch.FloatTensor] = None, use_cache: Optional[bool] = None, cache_position: Optional[torch.LongTensor] = None, codebook_idx: Optional[int] = None, **kwargs: Unpack[TransformersKwargs], ) -> BaseModelOutputWithPast: if position_ids is not None and not torch.compiler.is_compiling(): position_ids = None if (input_ids is None) == (inputs_embeds is None): raise ValueError("You must specify exactly one of input_ids or inputs_embeds.") if use_cache and past_key_values is None: device = inputs_embeds.device if inputs_embeds is not None else input_ids.device past_key_values = StaticCache(config=self.config, max_cache_len=self.config.rvq, device=device) if cache_position is None: past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0 inputs_seq_length = inputs_embeds.shape[1] if inputs_embeds is not None else input_ids.shape[1] device = inputs_embeds.device if inputs_embeds is not None else input_ids.device cache_position = torch.arange(past_seen_tokens, past_seen_tokens + inputs_seq_length, device=device) if inputs_embeds is None: if codebook_idx is not None: if codebook_idx <= 0: raise ValueError(f"`codebook_idx` must be in [1, {len(self.embed_tokens)}], got {codebook_idx}.") if codebook_idx > len(self.embed_tokens): raise ValueError(f"`codebook_idx` must be in [1, {len(self.embed_tokens)}], got {codebook_idx}.") if input_ids.ndim == 1: input_ids = input_ids.unsqueeze(1) token_emb = self.embed_tokens[codebook_idx - 1](input_ids[:, 0]).unsqueeze(1) # [B,1,H] inputs_embeds = token_emb else: if input_ids.shape[1] != cache_position.shape[0]: raise ValueError( "`input_ids` and `cache_position` must align in sequence length: " f"got {input_ids.shape[1]} and {cache_position.shape[0]}." ) codebook_idxs = torch.clamp(cache_position - 1, min=0, max=len(self.embed_tokens) - 1) inputs_embeds = torch.stack( [ self.embed_tokens[codebook_idx](input_ids[:, seq_idx]) for seq_idx, codebook_idx in enumerate(codebook_idxs.tolist()) ], dim=1, ) input_ids_are_first_codebook = bool(cache_position[0] == 0) if backbone_last_hidden_state is not None: inputs_embeds[:, 0, :] = backbone_last_hidden_state[:, 0, :] else: if not torch.compiler.is_compiling() and input_ids_are_first_codebook: logger.warning( "When the first codebook token is provided, `backbone_last_hidden_state` should also be provided for correct inference." ) causal_mask = create_causal_mask( config=self.config, input_embeds=inputs_embeds, attention_mask=attention_mask, cache_position=cache_position, past_key_values=past_key_values, position_ids=position_ids, ) hidden_states = inputs_embeds position_ids = cache_position.unsqueeze(0) position_embeddings = self.rotary_emb(hidden_states, position_ids) for decoder_layer in self.layers[: self.config.num_hidden_layers]: hidden_states = decoder_layer( hidden_states, attention_mask=causal_mask, position_ids=position_ids, past_key_values=past_key_values, use_cache=use_cache, cache_position=cache_position, position_embeddings=position_embeddings, **kwargs, ) hidden_states = self.norm(hidden_states) return BaseModelOutputWithPast( last_hidden_state=hidden_states, past_key_values=past_key_values if use_cache else None, ) class MossTTSRealtimeLocalTransformerForCausalLM(MossTTSRealtimeLocalTransformerPreTrainedModel, GenerationMixin): _tied_weights_keys = None _tp_plan = None _pp_plan = None def __init__(self, config): super().__init__(config) self.model = MossTTSRealtimeLocalTransformer(config) self.audio_vocab_size = self.config.audio_vocab_size self.local_lm_heads = nn.ModuleList( [nn.Linear(config.hidden_size, config.audio_vocab_size, bias=False) for _ in range(config.rvq)] ) self.post_init() def forward( self, input_ids: Optional[torch.LongTensor] = None, backbone_last_hidden_state: Optional[torch.FloatTensor] = 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, cache_position: Optional[torch.LongTensor] = None, codebook_idx: Optional[int] = None, logits_to_keep: Union[int, torch.Tensor] = 0, **kwargs: Unpack[TransformersKwargs], ) -> Union[tuple, CausalLMOutputWithPast]: outputs = self.model( input_ids=input_ids, backbone_last_hidden_state=backbone_last_hidden_state, inputs_embeds=inputs_embeds, attention_mask=attention_mask, position_ids=position_ids, past_key_values=past_key_values, use_cache=use_cache, cache_position=cache_position, codebook_idx=codebook_idx, **kwargs, ) hidden_states = outputs.last_hidden_state if isinstance(logits_to_keep, int): if logits_to_keep == 0: slice_indices = slice(0, None) else: slice_indices = slice(-logits_to_keep, None) else: slice_indices = logits_to_keep hs = hidden_states[:, slice_indices, :] if cache_position is not None: if codebook_idx is None: raise ValueError("`codebook_idx` must be provided when `cache_position` is provided.") logits = self.local_lm_heads[codebook_idx](hs[:, 0, :]).unsqueeze(1) else: if hs.shape[1] > len(self.local_lm_heads): raise ValueError( f"Cannot project {hs.shape[1]} codebooks with only {len(self.local_lm_heads)} LM heads." ) logits_list = [] for i in range(hs.shape[1]): logits_list.append(self.local_lm_heads[i](hs[:, i, :])) logits = torch.stack(logits_list, dim=1) logits = logits.contiguous() loss = None if labels is not None: loss = ForCausalLMLoss(logits, None, self.audio_vocab_size, shift_labels=labels.contiguous()) return CausalLMOutputWithPast( loss=loss, logits=logits, past_key_values=outputs.past_key_values, hidden_states=outputs.hidden_states, attentions=outputs.attentions, ) __all__ = [ "MossTTSRealtimeLocalTransformer", "MossTTSRealtimeLocalTransformerAttention", "MossTTSRealtimeLocalTransformerConfig", "MossTTSRealtimeLocalTransformerDecoderLayer", "MossTTSRealtimeLocalTransformerForCausalLM", "MossTTSRealtimeLocalTransformerPreTrainedModel", "MossTTSRealtimeLocalTransformerRMSNorm", "MossTTSRealtimeLocalTransformerRotaryEmbedding", ]