feat(CRITICAL): 从 GitHub 扫描搬运 ixformer SDK + xllm 完整 GDN/MoE 代码

来源:
  1. Chranos/ixformer (GitHub) → ixformer_sdk/ (230 files, 70K lines)
     - inference/functions/vllm.py: vllm_moe_topk_softmax 完整实现 (2033 lines)
     - inference/functions/moe.py: MoE ops 完整实现 (1380 lines)
     - contrib/vllm_flash_attn/: FA2 Python 接口 (1018 lines)
     - contrib/tgi/fused_moe.py: TGI fused MoE (429 lines)
     - csrc/include/ixformer/: C++ kernel headers + cmake

  2. Deep-Spark/xllm (GitHub) → upstream_ref/xllm_latest/ (+15 files)
     - npu_torch/qwen3_5_decoder_layer_impl.cpp/.h
     - npu_torch/qwen3_5_gated_delta_net.cpp/.h
     - npu_torch/qwen3_next_*.cpp/.h (6 files)
     - npu_torch/attention.cpp/.h + fused_moe.cpp/.h + CMakeLists.txt
     - models/llm/qwen3_5.h + qwen3_5_mtp.h + qwen3_next.h
     - models/vlm/qwen3_5.h

调用链完整性:
  ixformer_sdk/inference/functions/vllm.py
    → ops.infer.moe_topk_softmax() (C++ 层)
    → 这就是 base 镜像 libixformer.so 里的实现

  upstream_ref/xllm_latest/core/layers/ilu/fused_moe.cpp
    → ixformer::infer::topk_softmax() (直接 C++ 调用)
    → ixformer::infer::group_gemm() → 完整 7-step MoE pipeline

ixformer_sdk/train/speedformer/policy/bloom.py

@@ -0,0 +1,53 @@
+import warnings
+from abc import ABC, abstractmethod
+from functools import partial
+from typing import Callable, Dict, List, Union
+
+import torch.nn as nn
+from torch import Tensor
+from torch.nn import Module
+
+from ixformer.train.speedformer.policy.utils import SubModuleReplacementDescription
+from ixformer.train.speedformer.policy.replacer import Replacer
+
+from ixformer.train.speedformer.models.bloom.modeling_bloom import BloomModel, BloomBlock
+from ixformer.train.speedformer.layers.normalization import APEXFusedRMSNorm, IXFFusedRMSNorm
+from ixformer.train.speedformer.layers.bloom.attention import BloomFlashAttention
+
+
+class BloomReplacer(Replacer):
+    def __init__(self):
+        self.policy = {}
+
+    def module_policy(self) -> Dict[Union[str, nn.Module], List[SubModuleReplacementDescription]]:
+        self.append_or_create_submodule_replacement(
+            description=[
+                SubModuleReplacementDescription(
+                    suffix="input_layernorm",
+                    target_module=APEXFusedRMSNorm,
+                    kwargs={}
+                ),
+                SubModuleReplacementDescription(
+                    suffix="post_attention_layernorm",
+                    target_module=APEXFusedRMSNorm,
+                    kwargs={},
+                ),
+                SubModuleReplacementDescription(
+                    suffix="self_attention",
+                    target_module=BloomFlashAttention,
+                    kwargs={}
+                ),
+            ],
+            target_key="BloomBlock"
+        )
+
+        self.append_or_create_submodule_replacement(
+            description=[
+                SubModuleReplacementDescription(
+                    suffix="ln_f",
+                    target_module=APEXFusedRMSNorm,
+                    kwargs={}
+                ),
+            ],
+            target_key=BloomModel
+        )