Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

Add features for Qwen2_moe support

未关闭
  • 哪里不对？
#164 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
22/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
停滞
技术栈
python
领域
machine-learning

调研方向

The issue names no files or tests, so start by locating where the Qwen2 MoE model config and decoder layers are defined in the repository. Add the parameters decoder_sparse_step, moe_intermediate_size, shared_expert_intermediate_size, norm_topk_prob and mlp_only_layers, then use the layer choice to switch between dense MLP and sparse MoE blocks. Done means these options are honored, covered by tests, and documented.

由索引模型根据 Issue 内容生成。

描述

enhancement

🎯 Goal (What & Why)

The following additional parameters need to be added and their functionality implemented to support Qwen2_moe models:

  • decoder_sparse_step (int, optional, defaults to 1):
    The frequency of the MoE layer.
  • moe_intermediate_size (int, optional, defaults to 1408):
    Intermediate size of the routed expert.
  • shared_expert_intermediate_size (int, optional, defaults to 5632):
    Intermediate size of the shared expert.
  • norm_topk_prob (bool, optional, defaults to False):
    Whether to normalize the topk probabilities.
  • mlp_only_layers (List[int], optional, defaults to []):
    Indicate which layers use dense MLP rather than Sparse Moe blocks
    The list contains layer index, from 0 to num_layers-1 if we have num_layers layers
    If mlp_only_layers is empty, decoder_sparse_step is used to determine the sparsity.

Additionally add_linear_bias=only_attn_qkv needs to be supported then using MLP MoE blocks.

This is needed for #135

🚀 Execution Plan

**Step 1: Changes in MLP Moe
  • Implement custom intermediate sizes for shared and non shared Experts
  • add support for norm_topk_prob
**Step 2: Changes in layer architecture
  • Implement decoder_sparse_step and mlp_only_layers

📌 Acceptance Criteria (Must-Haves for Completion)

  • The feature must be functional and tested.
  • The implementation must be documented in practical terms.
  • The PR must include a performance/impact summary.
  • No refactors unless directly necessary for feature completion.

🛠️ Project Management

  • Assign the project to the Fast-LLM project.
  • Set the Estimate field (in days) in the GitHub project.
  • Use the Size field to categorize the PR size (Small/Medium/Large).
  • Assign an owner when opening the issue.
主要语言
Python
星标
337
派生
52
30 天内没有已合并 PR

环境准备

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

ServiceNow/Fast-LLM 的其他 Issue

查看 ServiceNow/Fast-LLM 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。