[data] fix qwen3omni moe model (#9501)

Co-authored-by: frozenleaves <frozen@Mac.local>

src/llamafactory/model/model_utils/moe.py

@@ -14,9 +14,13 @@
 
 from typing import TYPE_CHECKING, Union
 
+import torch
+from torch import nn
+from torch.nn import functional as F
 from transformers.integrations import is_deepspeed_zero3_enabled
 
 from ...extras.misc import check_version
+from ...extras.packages import is_transformers_version_greater_than
 
 
 if TYPE_CHECKING:
@@ -25,6 +29,9 @@ if TYPE_CHECKING:
 
     from ...hparams import ModelArguments
 
+if is_transformers_version_greater_than("4.57.0"):
+    from transformers.models.qwen3_omni_moe import modeling_qwen3_omni_moe
+
 
 def _set_z3_leaf_modules(model: "PreTrainedModel", leaf_modules: list[Union["nn.Module", str]]) -> None:
     check_version("deepspeed>=0.13.0")
@@ -175,3 +182,66 @@ def configure_moe(config: "PretrainedConfig", model_args: "ModelArguments", is_t
 
     elif model_type == "jetmoe":
         setattr(config, "aux_loss_coef", model_args.moe_aux_loss_coef)
+
+
+class Qwen3OmniMoeThinkerTextSparseMoeBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.num_experts = config.num_experts
+        self.top_k = config.num_experts_per_tok
+        self.norm_topk_prob = config.norm_topk_prob
+
+        # gating
+        self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False)
+        self.experts = nn.ModuleList(
+            [
+                modeling_qwen3_omni_moe.Qwen3OmniMoeThinkerTextMLP(
+                    config, intermediate_size=config.moe_intermediate_size
+                )
+                for _ in range(self.num_experts)
+            ]
+        )
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+        # router_logits: (batch * sequence_length, n_experts)
+        router_logits = self.gate(hidden_states)
+
+        # Calculate the routing weights for all experts
+        routing_weights = F.softmax(router_logits, dim=1, dtype=torch.float)
+
+        # Retain the weight of the top_k and reset the rest of the expert rights to 0 (instead of retaining only top_k experts)
+        top_k_weights, top_k_indices = torch.topk(routing_weights, self.top_k, dim=-1)
+        # Initialize the all-zero weight matrix (same shape as all experts)
+        full_routing_weights = torch.zeros_like(routing_weights)
+        # Only the weight of top_k experts is retained, and the weight of the rest of the experts remains at 0
+        full_routing_weights.scatter_(1, top_k_indices, top_k_weights)
+
+        # Normalized top_k weights (keep the original logic consistent)
+        if self.norm_topk_prob:
+            # Calculate the sum of the weights top_k each row (for normalization)
+            top_k_sum = full_routing_weights.sum(dim=-1, keepdim=True)
+            # Avoid dividing by zero
+            top_k_sum = torch.clamp(top_k_sum, min=1e-9)
+            full_routing_weights /= top_k_sum
+
+        # Convert back to the input data type
+        full_routing_weights = full_routing_weights.to(hidden_states.dtype)
+
+        final_hidden_states = torch.zeros(
+            (batch_size * sequence_length, hidden_dim), dtype=hidden_states.dtype, device=hidden_states.device
+        )
+
+        # Go through all the experts (not just the selected ones)
+        for expert_idx in range(self.num_experts):
+            expert_layer = self.experts[expert_idx]
+            # Get the weight of the current expert (inactive expert has a weight of 0 here)
+            expert_weights = full_routing_weights[:, expert_idx, None]  # shape: (batch*seq, 1)
+            # All samples participate in the calculations of the current expert, the weight may be equal to 0
+            current_hidden_states = expert_layer(hidden_states) * expert_weights
+            # Add-up to all expert outputs (experts with a weight of 0 do not affect the result)
+            final_hidden_states += current_hidden_states
+
+        final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
+        return final_hidden_states, router_logits

src/llamafactory/model/patcher.py

@@ -43,10 +43,20 @@ if TYPE_CHECKING:
 
     from ..hparams import ModelArguments
 
+if is_transformers_version_greater_than("4.57.0"):
+    from transformers.models.qwen3_omni_moe import modeling_qwen3_omni_moe
+
 
 logger = logging.get_logger(__name__)
 
 
+def patch_qwen3_omni_moe_thinker_text_sparse_moe_block():
+    if is_transformers_version_greater_than("4.57.0"):
+        from .model_utils.moe import Qwen3OmniMoeThinkerTextSparseMoeBlock
+
+        modeling_qwen3_omni_moe.Qwen3OmniMoeThinkerTextSparseMoeBlock = Qwen3OmniMoeThinkerTextSparseMoeBlock
+
+
 def patch_tokenizer(tokenizer: "PreTrainedTokenizer", model_args: "ModelArguments") -> None:
     if "PreTrainedTokenizerBase" not in str(tokenizer._pad.__func__):
         tokenizer._pad = MethodType(PreTrainedTokenizerBase._pad, tokenizer)
@@ -136,6 +146,9 @@ def patch_config(
     if getattr(config, "model_type", None) == "internlm3" and not is_transformers_version_greater_than("4.47.1"):
         raise RuntimeError("InternLM3 model requires transformers>=4.47.1, please upgrade it.")
 
+    if getattr(config, "model_type", None) == "qwen3_omni_moe":
+        patch_qwen3_omni_moe_thinker_text_sparse_moe_block()
+
     # deepspeed zero3 is not compatible with low_cpu_mem_usage
     init_kwargs["low_cpu_mem_usage"] = model_args.low_cpu_mem_usage and (not is_deepspeed_zero3_enabled())