[Cherry-pick][Optimization] enable trtllm_all_reduce fusion kernel in glm model

[BingooYang]

Motivation

FD接入trtllm_allreduce_fusion算子

Modifications

1. FD新增flashinfer allreduce fusion算子接入
2. 更改GLM-Air-4.5模型组网结构接入trtllm_allreduce_fusion算子（默认不开启）
3. 新增命令行参数--enable-flashinfer-allreduce-fusion，通过该参数来使能trtllm_allreduce_fusion
4. 新增trtllm_allreduce_fusion算子单测
5. 将def has_flashinfer()函数挪动到utils.py中

Usage or Command

H卡和B卡本地测试均通过
python -m fastdeploy.entrypoints.openai.api_server --model /root/paddlejob/workspace/bingoo/model/GLM-4.5-Air --tensor-parallel-size 4 --port 8185 --max-num-batched-tokens 2048 --enable-flashinfer-allreduce-fusion

Accuracy Tests

python -m paddle.distributed.launch --gpus=0,1 ./FastDeploy/tests/layers/test_rms_allreduce_fusion.py

Checklist

• Add at least a tag in the PR title.
  • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
  • You can add new tags based on the PR content, but the semantics must be clear.
• Format your code, run pre-commit before commit.
• Add unit tests. Please write the reason in this PR if no unit tests.
• Provide accuracy results.
• If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

[paddle-bot]

Thanks for your contribution!

[fastdeploy-bot]

📋 Review 摘要

PR 概述：为 GLM-Air-4.5 模型启用 trtllm_all_reduce fusion kernel，通过新增 flashinfer 融合算子提升性能。

变更范围：model_executor/layers/、model_executor/models/glm4_moe.py、config/、engine/

影响面 Tag：[Optimization] [Models] [OP]

📝 PR 规范检查

PR 符合规范：

• 标题包含有效 Tag [Optimization]
• Motivation 和 Modifications 描述清晰
• Checklist 项目完整
• 已提供 Usage 和 Accuracy Tests

问题

级别	文件	概述
🔴 Bug	normalization.py:249-252	fusion 返回 None 时会导致 AttributeError
🟡 建议	normalization.py:249, linear.py:941, flashinfer_comm_fusion.py:87,118	max_token_num 硬编码 2048，建议从配置读取

总体评价

PR 实现了 trtllm_all_reduce fusion kernel 的接入，测试覆盖了主要路径。但存在一个关键的 bug：当 flashinfer 不可用时，fusion 函数返回 (None, None)，但调用方没有正确处理这种情况，会导致运行时错误。建议修复后合并。

[fastdeploy-bot]

🔴 Bug 当 flashinfer_allreduce_residual_rmsnorm 返回 (None, None) 时（flashinfer 不可用或 workspace 未初始化），代码会尝试访问 norm_out[0].astype()，导致 AttributeError: 'NoneType' object has no attribute 'astype'。

建议修复方式：

# enable trtllm all reduce fusion
elif self.enable_all_reduce_fusion and x.shape[0] <= 2048:
    norm_out = flashinfer_allreduce_residual_rmsnorm(
        fd_config=self.fd_config, input_tensor=x, residual=residual_input, weight=self.weight, eps=self.eps
    )
    # Check if fusion succeeded, fallback to normal path if not
    if norm_out[0] is None or norm_out[1] is None:
        norm_out = self.norm_func(
            x,
            norm_weight=self.weight,
            norm_bias=None,
            epsilon=self.eps,
            begin_norm_axis=self.begin_norm_axis,
            bias=self.bias,
            residual=residual_input,
            quant_scale=(-1 if self.quant_scale is None else self.quant_scale),
            quant_round_type=self.quant_round_type,
            quant_max_bound=self.quant_max_bound,
            quant_min_bound=self.quant_min_bound,
        )

[fastdeploy-bot]

🟡 建议 max_token_num 在多处硬编码为 2048，限制了配置灵活性。建议从 FDConfig 中读取此参数。

影响位置：

1. linear.py:941 - out.shape[0] <= 2048
2. normalization.py:249 - x.shape[0] <= 2048
3. flashinfer_comm_fusion.py:87 - max_token_num: int = 2048 (默认参数)
4. flashinfer_comm_fusion.py:118 - max_token_num: int = 2048 (默认参数)

建议在 FDConfig 中添加 flashinfer_allreduce_max_token_num 字段，统一配置。