{
  "issues": [
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/84",
      "id": 3925470645,
      "node_id": "I_kwDOP30ofs7p-e21",
      "number": 84,
      "title": "How can I use Hunyuanimage-instrcut-3.0 for VQA ？",
      "user": {
        "login": "zhangquanwei962",
        "id": 61458769,
        "node_id": "MDQ6VXNlcjYxNDU4NzY5",
        "avatar_url": "https://avatars.githubusercontent.com/u/61458769?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/zhangquanwei962",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2026-02-11T09:06:59Z",
      "updated_at": "2026-02-11T09:07:40Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "I'm trying to use Hunyuanimage-instrcut-3.0 for **Visual Question Answering (VQA)** — given an image and a text question, I want the model to output a text answer in the \n```<think>...\n\n</think>\n<answer>...</answer>\n```format, similar to what's other MLLM for text understanding tasks.\n\nBut it will output\n```\nACTUAL PROMPT:\n<|startoftext|>Assistant: <boi><img_size_1024><img_ratio_19><timestep><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><joint_img_sep><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><img><eoi>\n\nUser: 这张图片里有什么？\n\nAssistant:\n================================================================================\n<think>用户给出的原始指令“这张图片里有什么？”是一个开放性问题，需要我仔细观察参考图并识别出其中所有可见的实体和元素。我将逐一分析图片中的各个部分，以确保没有遗漏任何信息。首先，最显眼的是一只小狗，它占据了画面的大部分空间。这只小狗戴着一顶红色的毛线帽和一条红色的毛线围巾。其次，在图片的顶部有文字信息，包括中文“新年快乐汪”和英文“HAPPY NEW YEAR”，以及一些装饰性的图案。小狗所处的环境是室内，背景是一个门口，可以看到门框和门后的部分室内景象。图片的底部是木地板。最后，整个图片被一个白色的边框包围，边框的边缘呈弧形。综合这些观察，我将构建一个详细的改写指令，清晰地列出图片中的所有主要构成部分。</think>这张图片展示了一只可爱的小狗，它戴着一顶红色的毛线帽和一条红色的毛线围巾，正坐在木地板上。小狗的背景是一个门口，门框清晰可见。图片的顶部有“新年快乐汪”和“HAPPY NEW YEAR”的文字，并点缀着一些装饰图案。整个画面被一个白色的边框包围，边框的边缘是圆弧形的。</recaption><answer><boi><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><img_size_1024><img_ratio_19><|endoftext|>\n```\nCan you help me or give me a example?",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/84/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/83",
      "id": 3921913305,
      "node_id": "I_kwDOP30ofs7pw6XZ",
      "number": 83,
      "title": "model_version AttributeError when calling load_tokenizer",
      "user": {
        "login": "jyh2378",
        "id": 33739495,
        "node_id": "MDQ6VXNlcjMzNzM5NDk1",
        "avatar_url": "https://avatars.githubusercontent.com/u/33739495?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/jyh2378",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 1,
      "created_at": "2026-02-10T14:51:35Z",
      "updated_at": "2026-02-12T03:14:44Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "Trying to load the tokenizer with model.load_tokenizer(model_id) results in the following error:\n```python\nAttributeError: 'HunyuanImage3Config' object has no attribute 'model_version'\n```\n\n### Reproduction Code\n```python\nmodel_id = \"./HunyuanImage-3-Instruct-Distil\"\n\nkwargs = dict(\n    attn_implementation=\"sdpa\", \n    trust_remote_code=True,\n    torch_dtype=\"auto\",\n    device_map=\"auto\",\n    moe_impl=\"eager\",\n    moe_drop_tokens=True,\n)\n\nmodel = AutoModelForCausalLM.from_pretrained(model_id, **kwargs)\nmodel.load_tokenizer(model_id)\n```\n\n### Error Trace\n```\nAttributeError: 'HunyuanImage3Config' object has no attribute 'model_version'\n```\n\nAdding a temporary `model_version` value in the `config.json` makes it work, but this seems to be a bug, so I’m reporting it.",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/83/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/82",
      "id": 3920182867,
      "node_id": "I_kwDOP30ofs7pqT5T",
      "number": 82,
      "title": "OOM，HunyuanImage-3.0-Instruct,   8*A100",
      "user": {
        "login": "zkailinzhang",
        "id": 10251153,
        "node_id": "MDQ6VXNlcjEwMjUxMTUz",
        "avatar_url": "https://avatars.githubusercontent.com/u/10251153?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/zkailinzhang",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2026-02-10T08:41:38Z",
      "updated_at": "2026-02-10T08:42:29Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "```\nfrom transformers import AutoModelForCausalLM\n\n# 加载模型\nmodel_id = \"./HunyuanImage-3-Instruct\"\n# 目前我们无法使用 HF 模型 ID `tencent/HunyuanImage-3.0-Instruct` 直接加载模型 \n# 因为名称中包含点号。\nmodel_id = \"/appdata//models/Tencent-Hunyuan/HunyuanImage-3.0\"\nmodel_id = \"/appdata//models/Tencent-Hunyuan/HunyuanImage-3.0-Instruct\"\n\ndef generate_image(image_path, prompt_text,image_paths_gen):\n    kwargs = dict(\n        attn_implementation=\"sdpa\", \n        trust_remote_code=True,\n        torch_dtype=\"auto\",\n        device_map=\"auto\",\n        moe_impl=\"eager\",   # 如果已安装 FlashInfer，可使用 \"flashinfer\"\n        moe_drop_tokens=True,\n    )\n\n    model = AutoModelForCausalLM.from_pretrained(model_id, **kwargs)\n    model.load_tokenizer(model_id)\n\n    # 图像到图像生成 (TI2I)\n    prompt = prompt_text\n\n    input_img1 = \"./assets/demo_instruct_imgs/input_1_0.png\"\n    input_img2 = \"./assets/demo_instruct_imgs/input_1_1.png\"\n    imgs_input = [image_path[0], image_path[1],image_path[2]]\n    # CUDA out of memory.\n    cot_text, samples = model.generate_image(\n        prompt=prompt,\n        image=imgs_input,\n        seed=42,\n        image_size=\"auto\",\n        use_system_prompt=\"en_unified\",\n        bot_task=\"think_recaption\",  # 使用 \"think_recaption\" 进行推理和增强\n        infer_align_image_size=True,  # 将输出图像大小对齐到输入图像大小\n        diff_infer_steps=5, \n        verbose=2\n    )\n\n    # 保存生成的图像\n    samples[0].save(image_paths_gen)\n\n```\n\n日志\n```\nYou are using a model of type hunyuan_image_3_moe to instantiate a model of type Hunyuan. This is not supported for all configurations of models and can yield errors.\n`torch_dtype` is deprecated! Use `dtype` instead!\nLoading checkpoint shards: 100%|███████████████████████████████████████████████████████████████████████████| 32/32 [00:31<00:00,  1.00it/s]\nThe tokenizer class you load from this checkpoint is not the same type as the class this function is called from. It may result in unexpected tokenization. \nThe tokenizer class you load from this checkpoint is 'PreTrainedTokenizerFast'. \nThe class this function is called from is 'HunyuanImage3TokenizerFast'.\nCUDA out of memory. Tried to allocate 20.25 GiB. GPU 0 has a total capacity of 79.32 GiB of which 18.81 GiB is free. Including non-PyTorch memory, this process has 60.50 GiB memory in use. Of the allocated memory 41.17 GiB is allocated by PyTorch, and 18.82 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)\nTraceback (most recent call last):\n  File \"/appdata/zhangkailin/ML/toyota_web_serve.py\", line 147, in ai_model_predict\n    local_gen_img_path = hunyuan_image.pipeline(image_path, prompt)\n                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/appdata/zhangkailin/ML/pack_modules/hunyuan_image.py\", line 48, in pipeline\n    generate_image(image_path, prompt_text, image_paths_gen)\n  File \"/appdata/zhangkailin/ML/pack_modules/hunyuan_image.py\", line 30, in generate_image\n    cot_text, samples = model.generate_image(\n                        ^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/utils/_contextlib.py\", line 120, in decorate_context\n    return func(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/modeling_hunyuan_image_3.py\", line 3296, in generate_image\n    model_inputs = self.prepare_model_inputs(\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/modeling_hunyuan_image_3.py\", line 2765, in prepare_model_inputs\n    cond_vae_images, cond_timesteps, cond_vit_images = self._encode_cond_image(\n                                                       ^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/modeling_hunyuan_image_3.py\", line 2465, in _encode_cond_image\n    cond_t_, cond_vae_image_ = self.vae_encode(\n                               ^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/modeling_hunyuan_image_3.py\", line 2419, in vae_encode\n    vae_encode_result = self.vae.encode(image)\n                        ^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/autoencoder_kl_3d.py\", line 794, in encode\n    h = _encode(x)\n        ^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/autoencoder_kl_3d.py\", line 774, in _encode\n    return self.encoder(x)\n           ^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/module.py\", line 1775, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/module.py\", line 1786, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/accelerate/hooks.py\", line 175, in new_forward\n    output = module._old_forward(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/autoencoder_kl_3d.py\", line 341, in forward\n    h = forward_with_checkpointing(self.down[i_level].block[i_block], h, use_checkpointing=use_checkpointing)\n        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/autoencoder_kl_3d.py\", line 110, in forward_with_checkpointing\n    return module(*inputs)\n           ^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/module.py\", line 1775, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/module.py\", line 1786, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/accelerate/hooks.py\", line 175, in new_forward\n    output = module._old_forward(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/autoencoder_kl_3d.py\", line 198, in forward\n    h = self.conv1(h)\n        ^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/module.py\", line 1775, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/module.py\", line 1786, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/accelerate/hooks.py\", line 175, in new_forward\n    output = module._old_forward(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3_dot_0_hyphen_Instruct/autoencoder_kl_3d.py\", line 147, in forward\n    return super().forward(input)\n           ^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/conv.py\", line 717, in forward\n    return self._conv_forward(input, self.weight, self.bias)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/home/zhangkailin/.conda/envs/ml/lib/python3.12/site-packages/torch/nn/modules/conv.py\", line 712, in _conv_forward\n    return F.conv3d(\n           ^^^^^^^^^\ntorch.OutOfMemoryError: CUDA out of memory. Tried to allocate 20.25 GiB. GPU 0 has a total capacity of 79.32 GiB of which 18.81 GiB is free. Including non-PyTorch memory, this process has 60.50 GiB memory in use. Of the allocated memory 41.17 GiB is allocated by PyTorch, and 18.82 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)\nNone\n```",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/82/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/81",
      "id": 3914288485,
      "node_id": "I_kwDOP30ofs7pT01l",
      "number": 81,
      "title": "TypeError: StaticLayer.lazy_initialization() missing 1 required positional argument: 'value_states'",
      "user": {
        "login": "akk-123",
        "id": 98469560,
        "node_id": "U_kgDOBd6GuA",
        "avatar_url": "https://avatars.githubusercontent.com/u/98469560?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/akk-123",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 1,
      "created_at": "2026-02-09T03:45:19Z",
      "updated_at": "2026-02-13T08:14:34Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "test HunyuanImage-3-Instruct, get error: TypeError: StaticLayer.lazy_initialization() missing 1 required positional argument: 'value_states'",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/81/reactions",
        "total_count": 1,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 1
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/80",
      "id": 3912318592,
      "node_id": "I_kwDOP30ofs7pMT6A",
      "number": 80,
      "title": "vLLM HunyuanImage3Orchestrator uses non-existent tokenizer attrs: special_token_map / all_special_tokens_extended",
      "user": {
        "login": "Xxxxxsun",
        "id": 87520911,
        "node_id": "MDQ6VXNlcjg3NTIwOTEx",
        "avatar_url": "https://avatars.githubusercontent.com/u/87520911?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/Xxxxxsun",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2026-02-08T08:50:30Z",
      "updated_at": "2026-02-12T05:57:26Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "When serving HunyuanImage-3.0 with the included vLLM integration, the server crashes during EngineCore initialization.\n```\nTraceback (most recent call last):                                                                                                                                                                                                                                                                                                  \n(EngineCore_DP0 pid=6139)   File \"/usr/lib/python3.12/multiprocessing/process.py\", line 314, in _bootstrap                                                                                                                                                                                                                                                    \n(EngineCore_DP0 pid=6139)     self.run()                                                                                                                                                                                                                                                                                                                      \n(EngineCore_DP0 pid=6139)   File \"/usr/lib/python3.12/multiprocessing/process.py\", line 108, in run                                                                                                                                                                                                                                                           \n(EngineCore_DP0 pid=6139)     self._target(*self._args, **self._kwargs)                                                                                                                                                                                                                                                                                       \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/v1/engine/core.py\", line 952, in run_engine_core                                                                                                                                                                                                                                                \n(EngineCore_DP0 pid=6139)     raise e                                                                                                                                                                                                                                                                                                                         \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/v1/engine/core.py\", line 939, in run_engine_core                                                                                                                                                                                                                                                \n(EngineCore_DP0 pid=6139)     engine_core = EngineCoreProc(*args, **kwargs)                                                                                                                                                                                                                                                                                   \n(EngineCore_DP0 pid=6139)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                                                                                                                                                   \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/v1/engine/core.py\", line 711, in __init__                                                                                                                                                                                                                                                       \n(EngineCore_DP0 pid=6139)     super().__init__(                                                                                                                                                                                                                                                                                                               \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/v1/engine/core.py\", line 213, in __init__                                                                                                                                                                                                                                                       \n(EngineCore_DP0 pid=6139)     register_hunyuan_image3_engine_core_hooks(self, vllm_config)                                                                                                                                                                                                                                                                    \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/multimodal/tasks/hunyuan_image3_orchestrator.py\", line 756, in register_hunyuan_image3_engine_core_hooks                                                                                                                                                                                        \n(EngineCore_DP0 pid=6139)     orch = HunyuanImage3Orchestrator(                                                                                                                                                                                                                                                                                               \n(EngineCore_DP0 pid=6139)            ^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                                                                                                                                                               \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/multimodal/tasks/hunyuan_image3_orchestrator.py\", line 577, in __init__                                                                                                                                                                                                                         \n(EngineCore_DP0 pid=6139)     self.token_cfg = HunyuanImage3TokenConfig(self.tokenizer)                                                                                                                                                                                                                                                                       \n(EngineCore_DP0 pid=6139)                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                                                                                                                                       \n(EngineCore_DP0 pid=6139)   File \"/HunyuanImage-3.0/vllm/vllm/multimodal/tasks/hunyuan_image3_orchestrator.py\", line 49, in __init__                                                                                                                                                                                                                          \n(EngineCore_DP0 pid=6139)     self.tokenizer.special_token_map[f\"<img_ratio_{i}>\"] for i in range(33)                                                                                                                                                                                                                                                         \n(EngineCore_DP0 pid=6139)     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                                                                                                                                                                                                \n(EngineCore_DP0 pid=6139)   File \"/usr/local/lib/python3.12/dist-packages/transformers/tokenization_utils_base.py\", line 1128, in __getattr__                                                                                                                                                                                                                 \n(EngineCore_DP0 pid=6139)     raise AttributeError(f\"{self.__class__.__name__} has no attribute {key}\")                                                                                                                                                                                                                                                       \n(EngineCore_DP0 pid=6139) AttributeError: HunyuanImage3TokenizerFast has no attribute special_token_map. Did you mean: 'special_tokens_map'?\n```\n\nEnvironment:\n- Python: 3.12\n- vLLM: 0.13.0 (repo fork)\n- transformers: (e.g.) 4.57.3\n- tokenizers: 0.22.x\n- CUDA: 12.8\n\nAfter patching the vLLM side to use special_tokens_map (instead of special_token_map), we hit another attribute mismatch in the tokenizer interface.\nvLLM expects:\n`self.end_answer_token_id = tokenizer.end_answer_token_id\n`\nbut the tokenizer implementation only provides:\n`end_of_answer_token_id`\n\nAnd so on.\n\nIs it possible that the HunyuanImage-3.0 model/tokenizer implementation has been updated, but the corresponding vLLM integration code has not been updated accordingly?\n",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/80/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/79",
      "id": 3909258656,
      "node_id": "I_kwDOP30ofs7pAo2g",
      "number": 79,
      "title": "vllm支持HunyuanImage-3.0-Instruct版本吗",
      "user": {
        "login": "seraphzf",
        "id": 23182782,
        "node_id": "MDQ6VXNlcjIzMTgyNzgy",
        "avatar_url": "https://avatars.githubusercontent.com/u/23182782?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/seraphzf",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2026-02-07T03:28:40Z",
      "updated_at": "2026-02-09T07:54:17Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "按照https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/blob/main/vllm_infer/README.md 页面的Manually Install Dependencies安装，HunyuanImage-3.0-Instruct版本的vllm_server跑不起来",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/79/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/78",
      "id": 3900849298,
      "node_id": "I_kwDOP30ofs7ogjyS",
      "number": 78,
      "title": "后训练主要关注的是图像生成吗？会不会对reasoning 文本造成影响？",
      "user": {
        "login": "zhangquanwei962",
        "id": 61458769,
        "node_id": "MDQ6VXNlcjYxNDU4NzY5",
        "avatar_url": "https://avatars.githubusercontent.com/u/61458769?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/zhangquanwei962",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2026-02-05T10:31:25Z",
      "updated_at": "2026-02-05T10:31:25Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "如题，看起来后训练只关注图像生成部分，这会对思维链啥的造成影响吗？",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/78/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/77",
      "id": 3900123400,
      "node_id": "I_kwDOP30ofs7odykI",
      "number": 77,
      "title": "ValueError: Invalid backend: 80",
      "user": {
        "login": "zkailinzhang",
        "id": 10251153,
        "node_id": "MDQ6VXNlcjEwMjUxMTUz",
        "avatar_url": "https://avatars.githubusercontent.com/u/10251153?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/zkailinzhang",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2026-02-05T07:31:14Z",
      "updated_at": "2026-02-13T02:52:40Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "A100 80g ,\nmoe_impl=\"flashinfer\",\n\nflashinfer/fused_moe/core.py\", line 290, in get_cutlass_fused_moe_module\n    raise ValueError(f\"Invalid backend: {backend}\")\nValueError: Invalid backend: 80",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/77/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/72",
      "id": 3878243449,
      "node_id": "I_kwDOP30ofs7nKUx5",
      "number": 72,
      "title": "vllm 支持的版本是哪个呀？ 一直找不到正确的版本",
      "user": {
        "login": "HXL3153",
        "id": 258172639,
        "node_id": "U_kgDOD2Nm3w",
        "avatar_url": "https://avatars.githubusercontent.com/u/258172639?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/HXL3153",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 1,
      "created_at": "2026-01-31T03:18:46Z",
      "updated_at": "2026-02-01T05:20:17Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": null,
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/72/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/70",
      "id": 3869989929,
      "node_id": "I_kwDOP30ofs7mq1wp",
      "number": 70,
      "title": "fp8 version",
      "user": {
        "login": "aikitoria",
        "id": 151776613,
        "node_id": "U_kgDOCQvtZQ",
        "avatar_url": "https://avatars.githubusercontent.com/u/151776613?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/aikitoria",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2026-01-29T10:42:13Z",
      "updated_at": "2026-02-10T08:15:16Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "It would be great to have an fp8 quantized version of the model that can fit in a single 96GB GPU. An official version using PTQ with the original dataset would be comparably cheap for you to produce and far superior to anything the community can create!",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/70/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/69",
      "id": 3865317843,
      "node_id": "I_kwDOP30ofs7mZBHT",
      "number": 69,
      "title": "HunyuanImage-3.0-Instruct-Distil 能提供 fp4版本吗",
      "user": {
        "login": "wodsoe",
        "id": 40934627,
        "node_id": "MDQ6VXNlcjQwOTM0NjI3",
        "avatar_url": "https://avatars.githubusercontent.com/u/40934627?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/wodsoe",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2026-01-28T12:54:36Z",
      "updated_at": "2026-02-10T08:12:37Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "HunyuanImage-3.0-Instruct-Distil 能提供 fp4版本吗？",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/69/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/63",
      "id": 3704522866,
      "node_id": "I_kwDOP30ofs7czohy",
      "number": 63,
      "title": "Does the vLLM framework you released support other text-to-image models?",
      "user": {
        "login": "JaDonghao",
        "id": 182836634,
        "node_id": "U_kgDOCuXdmg",
        "avatar_url": "https://avatars.githubusercontent.com/u/182836634?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/JaDonghao",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 1,
      "created_at": "2025-12-08T03:49:22Z",
      "updated_at": "2025-12-10T04:08:47Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": null,
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/63/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/62",
      "id": 3666928484,
      "node_id": "I_kwDOP30ofs7akONk",
      "number": 62,
      "title": "AutoTokenizer fails: \"AttributeError: 'dict' object has no attribute 'model_type'\" — missing tokenizer dependency (DeepSeek)",
      "user": {
        "login": "MaargoGysarova",
        "id": 113328619,
        "node_id": "U_kgDOBsFB6w",
        "avatar_url": "https://avatars.githubusercontent.com/u/113328619?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/MaargoGysarova",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 1,
      "created_at": "2025-11-26T10:42:25Z",
      "updated_at": "2026-02-01T05:24:24Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "AttributeError: 'dict' object has no attribute 'model_type'\n",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/62/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/60",
      "id": 3629038988,
      "node_id": "I_kwDOP30ofs7YTr2M",
      "number": 60,
      "title": "out of VRAM - 5 x A100 (80 GB)",
      "user": {
        "login": "mnauf",
        "id": 41832069,
        "node_id": "MDQ6VXNlcjQxODMyMDY5",
        "avatar_url": "https://avatars.githubusercontent.com/u/41832069?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/mnauf",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2025-11-15T18:14:02Z",
      "updated_at": "2026-02-01T06:24:43Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "A memory leak is occurring in the code. First request was served using gradio. In the second request, I got out of VRAM.",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/60/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/59",
      "id": 3615054810,
      "node_id": "I_kwDOP30ofs7XeVva",
      "number": 59,
      "title": "It has been over a month since HunyuanImage-3.0-Instruct was not made open source.",
      "user": {
        "login": "coder4nlp",
        "id": 93675439,
        "node_id": "U_kgDOBZVfrw",
        "avatar_url": "https://avatars.githubusercontent.com/u/93675439?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/coder4nlp",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2025-11-12T05:48:36Z",
      "updated_at": "2026-02-02T09:21:54Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": null,
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/59/reactions",
        "total_count": 1,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 1
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/58",
      "id": 3594754660,
      "node_id": "I_kwDOP30ofs7WQ5pk",
      "number": 58,
      "title": "4x48g 全精度推理方法",
      "user": {
        "login": "Duanyll",
        "id": 24842865,
        "node_id": "MDQ6VXNlcjI0ODQyODY1",
        "avatar_url": "https://avatars.githubusercontent.com/u/24842865?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/Duanyll",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2025-11-06T08:47:12Z",
      "updated_at": "2025-11-06T08:47:12Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "用 vllm 部署，得把 vae 丢到 cpu 上去跑，能跑多快就看 cpu 配置了，我用 9654 vae 解码这一步要 20 秒\n\n参考官方 vllm 部署教程，要魔改一些代码，把 [这里]( https://github.com/kippergong/vllm/blob/b8605f96de7f3f214b26bc413d79df1d164e4d73/vllm/multimodal/tasks/hunyuan_image3_orchestrator.py#L196C1-L241C38) 改成这样\n\n```python\n    @staticmethod\n    def _create_selective_device_map():\n        device_map = {}\n\n        # call vllm transformer_blocks, filter hf weights loading.\n        device_map[\"model.layers\"] = \"cpu\" # Handled by vLLM\n\n        # Order matters here. \n        # See https://github.com/huggingface/transformers/issues/34456#issuecomment-2445714200\n        device_map[\"timestep_emb\"] = \"cuda:0\"\n        device_map[\"patch_embed\"] = \"cuda:0\"\n        device_map[\"time_embed\"] = \"cuda:0\"\n        device_map[\"time_embed_2\"] = \"cuda:0\"\n        device_map[\"final_layer.model\"] = \"cuda:0\"\n        device_map[\"model.wte\"] = \"cuda:0\"\n        device_map[\"model.ln_f\"] = \"cuda:0\"\n        device_map[\"lm_head\"] = \"cuda:0\"\n        device_map[\"vision_model\"] = \"cuda:1\"\n        device_map[\"vision_aligner\"] = \"cuda:1\"\n        # Should be executed with cpu, however accelerate will try to offload to cuda if\n        # we set to 'cpu', so we offload to cuda:2 and then move to cpu later.\n        device_map[\"vae\"] = \"cuda:2\"\n\n        return device_map\n\n    @staticmethod\n    def load_multi_moda(path, infer_mode=\"vllm\") -> tuple[str, Any]:\n        \"\"\"\n        infer_mode: vllm or hf.\n        \"\"\"\n        if infer_mode == \"vllm\":\n            device_map = ModelProvider._create_selective_device_map()\n        else:\n            device_map = \"auto\"\n        kwargs = dict(\n            attn_implementation=\"sdpa\",\n            torch_dtype=\"auto\",\n            device_map=device_map,\n            moe_impl=\"eager\",\n            low_cpu_mem_usage=True,\n        )\n\n        with suppress_transformers_unused_weight_log():\n            model = HunyuanImage3ForCausalMM.from_pretrained(path, **kwargs)\n        # Override getter for model.device so it returns cuda:0\n        model.__class__.device = property(lambda self: torch.device(\"cuda:0\"))\n        # Move VAE to CPU and remove hooks to avoid issues during inference\n        model.vae.to(\"cpu\")\n        from accelerate.hooks import remove_hook_from_module\n        for name, module in model.vae.named_modules():\n            remove_hook_from_module(module)\n        model.load_tokenizer(path)\n        print(\"layer device_map:\")\n        print(model.hf_device_map)\n\n        return model, model.tokenizer\n```\n\n然后在 [这里](https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/blob/62da220178f4b0b7d83e91665a46a20a3ee4f7cd/hunyuan_image_3/hunyuan_image_3_pipeline.py#L865) 改成这样\n\n```python\n        # with torch.autocast(device_type=\"cuda\", dtype=torch.float16, enabled=True):\n        latents = latents.to(device=self.vae.device, dtype=self.vae.dtype)\n        image = self.vae.decode(latents, return_dict=False, generator=generator)[0]\n```\n\n我用的 vllm 启动参数：\n\n```sh\nexport VLLM_ENABLE_HUNYUAN_IMAGE3_TASK=\"1\"\nexport MULTI_MODA_SAVE_PATH=\"/tmp/hunyuan_image3/png/\" # Typo from the original code\n# export NCCL_DEBUG=INFO\nexport NCCL_P2P_DISABLE=1\nexport NCCL_CUMEM_HOST_ENABLE=0\nexport OMP_NUM_THREADS=192 # Set according to your CPU cores\n\n############################################\n# 3. Start vLLM service\n############################################\n# Uncomment the next line for nsys profiling\n# nsys launch --trace-fork-before-exec true --session test -t cuda,cublas,cudnn,nvtx --cuda-graph-trace=node \\\nvllm serve /workspace/HunyuanImage-3.0/models/tencent/HunyuanImage-3.0 \\\n    --trust-remote-code \\\n    --served-model-name vllm_hunyuan_image3 \\\n    --gpu-memory-utilization 0.83 \\\n    --max-model-len 10000 \\\n    --no-enable-prefix-caching \\\n    --no-enable-chunked-prefill \\\n    --max-num-batched-tokens 10000 \\\n    --max-num-seqs 1 \\\n    --enforce-eager \\\n    --trust-request-chat-template \\\n    --tensor-parallel-size 4\n```",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/58/reactions",
        "total_count": 1,
        "+1": 1,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/57",
      "id": 3584354475,
      "node_id": "I_kwDOP30ofs7VpOir",
      "number": 57,
      "title": "Any plan to support diffusers?",
      "user": {
        "login": "DefTruth",
        "id": 31974251,
        "node_id": "MDQ6VXNlcjMxOTc0MjUx",
        "avatar_url": "https://avatars.githubusercontent.com/u/31974251?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/DefTruth",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2025-11-04T01:31:31Z",
      "updated_at": "2025-11-04T01:31:31Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": null,
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/57/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/51",
      "id": 3531751162,
      "node_id": "I_kwDOP30ofs7Sgj76",
      "number": 51,
      "title": "Astronaut test failed",
      "user": {
        "login": "kabachuha",
        "id": 14872007,
        "node_id": "MDQ6VXNlcjE0ODcyMDA3",
        "avatar_url": "https://avatars.githubusercontent.com/u/14872007?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/kabachuha",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2025-10-20T09:39:07Z",
      "updated_at": "2025-10-20T09:39:07Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "The \"astronaut\" test is a classic test probing whether a model has real understanding of instructions and not just mimicking the training data distribution.\n\nThe prompt is \n\n\"\"\"\nA cinematic photo of a horse riding an astronaut on the moon. The horse is above the astronaut.\n\"\"\"\n\nAnd, well, the result, while beautiful, is below\n\n<img width=\"1024\" height=\"1024\" alt=\"Image\" src=\"https://github.com/user-attachments/assets/2fad780b-6277-452e-bb5a-ba01240ff662\" />",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/51/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/50",
      "id": 3531580601,
      "node_id": "I_kwDOP30ofs7Sf6S5",
      "number": 50,
      "title": "Are there any other parallelization schemes available?",
      "user": {
        "login": "lyj-jjj",
        "id": 215845276,
        "node_id": "U_kgDODN2JnA",
        "avatar_url": "https://avatars.githubusercontent.com/u/215845276?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/lyj-jjj",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2025-10-20T08:45:36Z",
      "updated_at": "2025-10-20T08:45:36Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "In hunyuan-image3.0, the default approach uses PP parallelism for transformers. Are there any other parallelization schemes available?",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/50/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/49",
      "id": 3525054609,
      "node_id": "I_kwDOP30ofs7SHBCR",
      "number": 49,
      "title": "Can images be published as soon as possible",
      "user": {
        "login": "dxhark",
        "id": 186033737,
        "node_id": "U_kgDOCxamSQ",
        "avatar_url": "https://avatars.githubusercontent.com/u/186033737?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/dxhark",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 1,
      "created_at": "2025-10-17T08:29:57Z",
      "updated_at": "2025-11-06T07:44:46Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "Can image to image publishing be done as soon as possible? Currently, it is difficult to meet the design application scenarios with only text generated images. If there are some shortcomings in image to image, the community may also work together to study the problem. If it is an autoregressive model, there is no need to worry about consistency issues. Many design requirements do not require consistency but rather diverge in inspiration. We hope to surpass Midjourney in creativity",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/49/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/44",
      "id": 3499641151,
      "node_id": "I_kwDOP30ofs7QmEk_",
      "number": 44,
      "title": "MPS/Metal/Mac support",
      "user": {
        "login": "e1732a364fed",
        "id": 75717694,
        "node_id": "MDQ6VXNlcjc1NzE3Njk0",
        "avatar_url": "https://avatars.githubusercontent.com/u/75717694?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/e1732a364fed",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2025-10-09T15:31:48Z",
      "updated_at": "2025-10-09T15:31:48Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "Please support mac as well. A mac studio has massive RAM which is perfect for hunyuanimage-3.0\n\n求mac的mps支持。现在代码中全是cuda，没法在mac上用。mac studio 的大统一内存用来运行Hunyuanimage 3 不是正好嘛",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/44/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/40",
      "id": 3490464209,
      "node_id": "I_kwDOP30ofs7QDEHR",
      "number": 40,
      "title": "Hanging when using flash attention",
      "user": {
        "login": "jinleic",
        "id": 42644281,
        "node_id": "MDQ6VXNlcjQyNjQ0Mjgx",
        "avatar_url": "https://avatars.githubusercontent.com/u/42644281?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/jinleic",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 0,
      "created_at": "2025-10-07T08:21:20Z",
      "updated_at": "2025-10-07T08:21:20Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "```\n    image = model.generate_image(prompt=prompt, stream=True)\n            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/root/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3/hunyuan.py\", line 2644, in generate_image\n    outputs = self._generate(**model_inputs, **kwargs, verbose=verbose)\n              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/root/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3/hunyuan.py\", line 2559, in _generate\n    return super().generate(\n           ^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/utils/_contextlib.py\", line 116, in decorate_context\n    return func(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/transformers/generation/utils.py\", line 2564, in generate\n    result = decoding_method(\n             ^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/transformers/generation/utils.py\", line 2784, in _sample\n    outputs = self(**model_inputs, return_dict=True)\n              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1751, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1762, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/accelerate/hooks.py\", line 175, in new_forward\n    output = module._old_forward(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/root/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3/hunyuan.py\", line 2019, in forward\n    outputs = self.model(\n              ^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1751, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1762, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/root/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3/hunyuan.py\", line 1671, in forward\n    layer_outputs = decoder_layer(\n                    ^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1751, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1762, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/accelerate/hooks.py\", line 175, in new_forward\n    output = module._old_forward(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/root/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3/hunyuan.py\", line 1497, in forward\n    hidden_states = self.mlp(hidden_states)\n                    ^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1751, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py\", line 1762, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/accelerate/hooks.py\", line 175, in new_forward\n    output = module._old_forward(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/root/.cache/huggingface/modules/transformers_modules/HunyuanImage_hyphen_3/hunyuan.py\", line 1126, in forward\n    _ = flashinfer.fused_moe.cutlass_fused_moe(     # noqa\n        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/flashinfer/fused_moe/core.py\", line 888, in cutlass_fused_moe\n    return get_cutlass_fused_moe_module(device_arch).cutlass_fused_moe(\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/flashinfer/fused_moe/core.py\", line 404, in get_cutlass_fused_moe_module\n    ).build_and_load(class_name=\"FusedMoeRunner\")\n      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/site-packages/flashinfer/jit/core.py\", line 160, in build_and_load\n    self.build(verbose, need_lock=False)\n  File \"/opt/conda/lib/python3.11/site-packages/flashinfer/jit/core.py\", line 140, in build\n    run_ninja(jit_env.FLASHINFER_JIT_DIR, self.ninja_path, verbose)\n  File \"/opt/conda/lib/python3.11/site-packages/flashinfer/jit/cpp_ext.py\", line 258, in run_ninja\n    subprocess.run(\n  File \"/opt/conda/lib/python3.11/subprocess.py\", line 550, in run\n    stdout, stderr = process.communicate(input, timeout=timeout)\n                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/opt/conda/lib/python3.11/subprocess.py\", line 1196, in communicate\n    stdout = self.stdout.read()\n             ^^^^^^^^^^^^^^^^^^\nKeyboardInterrupt\n\n```\n\n\nThe command I'm using:\n\n```\ndocker pull ghcr.io/pytorch/pytorch:2.7.1-cuda12.8-cudnn9-devel\n\ndocker run -itd --shm-size 64g --gpus all -v /data/.cache:/root/.cache -v /data:/root/data --ipc=host --network=host --name hunyuan-nvcc ghcr.io/pytorch/pytorch:2.7.1-cuda12.8-cudnn9-devel /bin/bash\n\ndocker exec -it hunyuan-nvcc bash\n\napt update && apt install -y vim && apt install -y git && apt install -y fish\n\ngit clone https://github.com/Tencent-Hunyuan/HunyuanImage-3.0.git\n\ncd HunyuanImage-3.0\n\n# 1. First install PyTorch (CUDA 12.8 Version)\npip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu128\n\n# 2. Then install tencentcloud-sdk\npip install -i https://mirrors.tencent.com/pypi/simple/ --upgrade tencentcloud-sdk-python\n\n# 3. Then install other dependencies\npip install -r requirements.txt\n\n# FlashAttention for faster attention computation\npip install flash-attn==2.8.3 --no-build-isolation\n\n# FlashInfer for optimized moe inference. v0.3.1 is tested.\npip install flashinfer-python\n\nhf download tencent/HunyuanImage-3.0 --local-dir ~/.cache/HunyuanImage-3\n```\n\nthe PYTHON file I'm using:\n\n\n```\nfrom transformers import AutoModelForCausalLM\n\n# Load the model\nmodel_id = \"/root/.cache/HunyuanImage-3\"\n# Currently we can not load the model using HF model_id `tencent/HunyuanImage-3.0` directly \n# due to the dot in the name.\n\nkwargs = dict(\n    attn_implementation=\"flash_attention_2\",     # Use \"flash_attention_2\" if FlashAttention is installed\n    trust_remote_code=True,\n    torch_dtype=\"auto\",\n    device_map=\"auto\",\n    moe_impl=\"flashinfer\",   # Use \"flashinfer\" if FlashInfer is installed\n)\n\nmodel = AutoModelForCausalLM.from_pretrained(model_id, **kwargs)\nmodel.load_tokenizer(model_id)\n\n# generate the image\nprompt = \"A brown and white dog is running on the grass\"\nimage = model.generate_image(prompt=prompt, stream=True)\nimage.save(\"image.png\")\n```",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/40/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/39",
      "id": 3490174743,
      "node_id": "I_kwDOP30ofs7QB9cX",
      "number": 39,
      "title": "🆘 OOM Error during inference with 4x80GB A800 GPUs",
      "user": {
        "login": "chen-yy20",
        "id": 84724215,
        "node_id": "MDQ6VXNlcjg0NzI0MjE1",
        "avatar_url": "https://avatars.githubusercontent.com/u/84724215?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/chen-yy20",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 9,
      "created_at": "2025-10-07T06:44:45Z",
      "updated_at": "2026-02-10T08:49:32Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "Hi, I'm encountering an Out of Memory (OOM) error while running inference using HunyuanImage-3 model, even with 4x80GB A800 GPUs (and also tried with 8xA800). The error occurs during step 3 of the inference process.\n\n### Environment:\n```\nHardware: 4x NVIDIA A800-SXM4-80GB GPUs\nCUDA Version: 12.4\nDriver Version: 550.54.15\n```\n\n### Steps to reproduce:\n\nFollowing the readme instructions\nRunning the command (disabled Deepseek rewrite):\n`python3 run_image_gen.py --model-id ../models/hunyuanimage --verbose 1 --prompt \"A brown and white dog is running on the grass\"`\n\n### Memory usage before OOM:\n\n```\nstep 0 | GPU Memory: 34375MiB / 81222MiB\nstep 1 | GPU Memory: 34598MiB / 81222MiB\nstep 2 | GPU Memory: 34598MiB / 81222MiB\n```\n### Error message:\n```\ntorch.OutOfMemoryError: CUDA out of memory. Tried to allocate 31.02 GiB. GPU 1 has a total capacity of 79.32 GiB of which 5.11 GiB is free. Including non-PyTorch memory, this process has 74.20 GiB memory in use. Of the allocated memory 54.21 GiB is allocated by PyTorch, and 19.50 GiB is reserved by PyTorch but unallocated.\n```\n\nThe error occurs in the MoE gating function:\n\nFile \".../hunyuan_image_3/hunyuan.py\", line 319, in topkgating\n    exp_counts_capacity = torch.sum(dispatch_mask)\n\n\n### Additional Information:\n\nThe model initially loads successfully and begins inference\nOOM occurs consistently at step 3\nEven with 8xA800, the same issue persists\nThe error suggests trying PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True, but this didn't resolve the issue\n\n---\nWould appreciate any guidance on:\n\n* Proper configuration for multi-GPU inference\n* Recommended memory management settings\n* Whether there are known workarounds for this specific OOM issue\n\nLet me know if you need any additional information.",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/39/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/38",
      "id": 3482081472,
      "node_id": "I_kwDOP30ofs7PjFjA",
      "number": 38,
      "title": "好奇怪...怎么输入二次元风格都没法出像样的二次元,",
      "user": {
        "login": "xdong189",
        "id": 113279608,
        "node_id": "U_kgDOBsCCeA",
        "avatar_url": "https://avatars.githubusercontent.com/u/113279608?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/xdong189",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2025-10-03T18:14:58Z",
      "updated_at": "2025-10-17T08:54:29Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "好像对风格不敏感.,\n还有就是怎么细节又多又乱...是不是数据集都是AI跑出来的...\n美学倒还是可以,不过出来的东西噪点也很多,细节很乱\n拿来做项目的话要改的就非常多了..",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/38/reactions",
        "total_count": 3,
        "+1": 2,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 1,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/34",
      "id": 3469904013,
      "node_id": "I_kwDOP30ofs7O0oiN",
      "number": 34,
      "title": "Does the model have a quantized version? 240GB of VRAM isn't meant for me.",
      "user": {
        "login": "jerometan",
        "id": 1586546,
        "node_id": "MDQ6VXNlcjE1ODY1NDY=",
        "avatar_url": "https://avatars.githubusercontent.com/u/1586546?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/jerometan",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2025-09-30T14:56:17Z",
      "updated_at": "2026-02-10T08:14:12Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "Does the model have a quantized version? 240GB of VRAM isn't meant for me.\n\n",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/34/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/33",
      "id": 3469664231,
      "node_id": "I_kwDOP30ofs7Ozt_n",
      "number": 33,
      "title": "vLLM or Comfy UI support when?",
      "user": {
        "login": "nitinh12",
        "id": 99498126,
        "node_id": "U_kgDOBe44jg",
        "avatar_url": "https://avatars.githubusercontent.com/u/99498126?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/nitinh12",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 6,
      "created_at": "2025-09-30T14:04:55Z",
      "updated_at": "2026-02-10T08:14:35Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": null,
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/33/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/31",
      "id": 3468670593,
      "node_id": "I_kwDOP30ofs7Ov7aB",
      "number": 31,
      "title": "我觉得你们的moe架构和ktransformer加速框架很适合",
      "user": {
        "login": "Lagrebanana",
        "id": 178156007,
        "node_id": "U_kgDOCp5x5w",
        "avatar_url": "https://avatars.githubusercontent.com/u/178156007?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/Lagrebanana",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 4,
      "created_at": "2025-09-30T10:23:36Z",
      "updated_at": "2025-10-29T06:23:00Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": null,
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/31/reactions",
        "total_count": 0,
        "+1": 0,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/29",
      "id": 3468103220,
      "node_id": "I_kwDOP30ofs7Otw40",
      "number": 29,
      "title": "Bitsandbytes quantization and 4bit model upload",
      "user": {
        "login": "kabachuha",
        "id": 14872007,
        "node_id": "MDQ6VXNlcjE0ODcyMDA3",
        "avatar_url": "https://avatars.githubusercontent.com/u/14872007?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/kabachuha",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 6,
      "created_at": "2025-09-30T08:12:54Z",
      "updated_at": "2025-10-18T21:13:49Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "Hi!\n\nWould it be possible to add 8/4 bit quantization with bitsandbytes and accelerate, just like Bytedance's Bagel does in https://github.com/ByteDance-Seed/Bagel/blob/7026cfa0a4df274460d0b0b990117398a4ec6fca/app.py#L115-L122?\n\nOn top of that, quantization will create a folder with the quantized 4bit model and it is simple to upload it to Huggingface for people to download quickly",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/29/reactions",
        "total_count": 4,
        "+1": 4,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/28",
      "id": 3467924087,
      "node_id": "I_kwDOP30ofs7OtFJ3",
      "number": 28,
      "title": "表达一些我的看法",
      "user": {
        "login": "quanfei55",
        "id": 231609927,
        "node_id": "U_kgDODc4WRw",
        "avatar_url": "https://avatars.githubusercontent.com/u/231609927?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/quanfei55",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 4,
      "created_at": "2025-09-30T07:16:53Z",
      "updated_at": "2025-10-07T20:39:11Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "以下是个人的一点感受，当时2.1出来的时候用了当时还没有什么太大感觉，这么快3.0出来又不断在试，感受就是模型生图效果可圈可点，分辨率不高的话拿无损放大一下就不错了，这个模型可能不是顶尖的，但一定优秀，反正我认为完全不差。得益于这样的参数量，这个模型没有致命的短板，而且在非常多复杂任务上都能得心应手，比如理解你的意思并做出准确的标注着详细信息的海报。很多图对提示词理解很好，很多抽象的词汇表达的意境也能表达准确。我生图当中很多时候会用来做绘本，写实非写实的都有，它生成的构图和千问很多地方不一样。两个模型各有千秋，混元的相同的提示词这个画风构图和千问挺不一样，完全给了我更多的一个不一样的选择，在出图结果上多了一个模型，带来的出图结果是多了更多的可能性，这点对我挺重要（不然好点就千问一个选择，那就只有类似的构图类似的画风了，这样太单调了）。测了这么多张图，我感觉像这种多模态的模型有着很高的上限，就是希望画面内容再多点加上人像这块得再提升点。\n总之很感谢混元开源这样的模型，很多这种大体积大参数的模型都是闭源的，开源的有很多都是小的模型，那种小的基本都好不到哪里去，模型如果不大我都看不上。很多人骂这个模型太大跑不动，我也跑不动，但就算跑不动开源也比那种纯闭源在我心里要好的多得多得多得多，像即梦banana那种纯闭源是真的绝望，这种模型随时给你降智，什么时候模型没了下架都不知道，这种模型永远是用一天算一天的状态用着太憋屈了，基本不太想用。看到这样的大模型开源了都能被骂，我很担心以后再也不会这样大的模型开源了，混元3.0很可能是最后一个也是最大的图像开源模型了，各位且用且珍惜吧。图像这块有这样大的一个开源是极其不容易的。就说这么多吧，总之很感谢混元，这次这个很好，这种诚意的开源已经完全不想挑刺了，比起其他有些抠抠搜搜藏着掖着憋了半天开源一个小模型混元好的不是一星半点，混元是直接把自己线上最大最好的图像模型直接拿出来开源，光凭这点说什么也要支持，就这种程度的模型再多的不足我都可以忽略了，更何况完全不差，是个六边形战士。另外我很期待图像编辑，我也很重视图像编辑，希望混元能出一个强大的如同banana的编辑模型，再次感谢你们。",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/28/reactions",
        "total_count": 11,
        "+1": 8,
        "-1": 0,
        "laugh": 0,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 3,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/5",
      "id": 3461388481,
      "node_id": "I_kwDOP30ofs7OUJjB",
      "number": 5,
      "title": "Reward Distribution Alignment (ReDA) method paper release?",
      "user": {
        "login": "JoshonSmith",
        "id": 69639577,
        "node_id": "MDQ6VXNlcjY5NjM5NTc3",
        "avatar_url": "https://avatars.githubusercontent.com/u/69639577?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/JoshonSmith",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "labels": {},
      "state": "open",
      "locked": false,
      "assignee": null,
      "assignees": {},
      "milestone": null,
      "comments": 2,
      "created_at": "2025-09-28T08:48:47Z",
      "updated_at": "2025-10-10T12:25:17Z",
      "closed_at": null,
      "author_association": "NONE",
      "type": null,
      "active_lock_reason": null,
      "sub_issues_summary": {
        "total": 0,
        "completed": 0,
        "percent_completed": 0
      },
      "issue_dependencies_summary": {
        "blocked_by": 0,
        "total_blocked_by": 0,
        "blocking": 0,
        "total_blocking": 0
      },
      "body": "great work!",
      "closed_by": null,
      "reactions": {
        "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/5/reactions",
        "total_count": 1,
        "+1": 0,
        "-1": 0,
        "laugh": 1,
        "hooray": 0,
        "confused": 0,
        "heart": 0,
        "rocket": 0,
        "eyes": 0
      },
      "performed_via_github_app": null,
      "state_reason": null,
      "pinned_comment": null,
      "linked_prs": []
    }
  ],
  "pulls": [
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/86",
      "id": 3288518846,
      "node_id": "PR_kwDOP30ofs7EAtC-",
      "number": 86,
      "state": "open",
      "locked": false,
      "title": "Fix progressive slowdown in text streaming across generations",
      "user": {
        "login": "auroter",
        "id": 7332587,
        "node_id": "MDQ6VXNlcjczMzI1ODc=",
        "avatar_url": "https://avatars.githubusercontent.com/u/7332587?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/auroter",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "body": "## Summary\n\nConsecutive generations in the Gradio web UI get progressively slower in the text streaming (thinking/recaption) phase. The model generates tokens at a constant rate (~24 tok/s), but each `yield history` sends the full chatbot history to Gradio for serialization. Two compounding issues cause this:\n\n- **~200 yields per generation**: Every single text token triggers a `yield history`, and serialization cost grows with history size (O(tokens × images))\n- **PIL re-encoding on every yield**: `gr.Image(pil_image, type=\"pil\")` objects in history get re-encoded to PNG by Gradio on every yield — even for unchanged old messages (Gradio creates a new component instance each time in `chatbot.py:557`). This affects both generated images and user-uploaded images.\n\nA 1024px PNG encode takes ~100ms. With N prior images and ~200 yields, this adds N×20 seconds of overhead per generation, making text streaming visibly degrade from fast to ~1 word/second after just 2-3 generations.\n\n### Changes (single file: `app/run_chatbot.py`)\n\n- **Throttle yields to 50ms intervals** (~20-40 per generation instead of ~200). Text is still buffered and flushed before image events.\n- **Store file paths in `gr.Image` instead of PIL objects.** Generated images are saved to temp files first; user-uploaded images are already file paths from Gradio's `MultimodalTextbox`, so dropping `type=\"pil\"` avoids an unnecessary PIL round-trip. File path serialization is nearly free vs. re-encoding.\n- **Close the file descriptor from `tempfile.mkstemp()`** to prevent fd exhaustion after many generations.\n\nTogether these reduce per-generation overhead from O(tokens × images) to roughly O(1).\n\n### Measured results\n\n| Generation | Model tok/s | Avg yield time (before fix) | Avg yield time (after fix) |\n|---|---|---|---|\n| 1st | 24.0 | 2ms | 2ms |\n| 2nd (1 image in history) | 24.6 | ~500ms+ (unthrottled) | 72ms |\n| 3rd (2 images in history) | 24.5 | ~1000ms+ (unthrottled) | 129ms |\n\nModel speed is constant — the slowdown was entirely in the Gradio serialization layer.\n\n## Test plan\n\n- [ ] Run 3+ consecutive text-to-image generations with `single_round` + `think` mode\n- [ ] Confirm text streaming speed stays consistent across all generations\n- [ ] Confirm images still display correctly in the chatbot\n- [ ] Confirm `--image-cache-dir` still saves images correctly\n- [ ] Test with `unlimited` context mode to verify full history still works\n- [ ] Test image-to-image (user uploads an image) to confirm uploaded images display correctly\n\n🤖 Generated with [Claude Code](https://claude.com/claude-code)",
      "created_at": "2026-02-16T06:04:07Z",
      "updated_at": "2026-02-16T06:21:07Z",
      "closed_at": null,
      "merged_at": null,
      "merge_commit_sha": "844174b35aabfa352a7b148ffe1b65575720fe70",
      "assignee": null,
      "assignees": {},
      "requested_reviewers": {},
      "requested_teams": {},
      "labels": {},
      "milestone": null,
      "draft": false,
      "head": {
        "label": "auroter:fix/streaming-slowdown",
        "ref": "fix/streaming-slowdown",
        "sha": "31bc2809f64118342f2d8c96d6e9b007c4cd0edb",
        "user": {
          "login": "auroter",
          "id": 7332587,
          "node_id": "MDQ6VXNlcjczMzI1ODc=",
          "avatar_url": "https://avatars.githubusercontent.com/u/7332587?v=4",
          "gravatar_id": "",
          "url": "https://api.github.com/users/auroter",
          "type": "User",
          "user_view_type": "public",
          "site_admin": false
        },
        "repo": {
          "id": 1147076906,
          "node_id": "R_kgDORF8BKg",
          "name": "HunyuanImage-3.0",
          "full_name": "auroter/HunyuanImage-3.0",
          "private": false,
          "owner": {
            "login": "auroter",
            "id": 7332587,
            "node_id": "MDQ6VXNlcjczMzI1ODc=",
            "avatar_url": "https://avatars.githubusercontent.com/u/7332587?v=4",
            "gravatar_id": "",
            "url": "https://api.github.com/users/auroter",
            "type": "User",
            "user_view_type": "public",
            "site_admin": false
          },
          "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
          "fork": true,
          "url": "https://api.github.com/repos/auroter/HunyuanImage-3.0",
          "created_at": "2026-02-01T06:33:32Z",
          "updated_at": "2026-02-01T06:33:32Z",
          "pushed_at": "2026-02-16T06:44:55Z",
          "homepage": "https://hunyuan.tencent.com/image",
          "size": 93479,
          "stargazers_count": 0,
          "watchers_count": 0,
          "language": null,
          "has_issues": false,
          "has_projects": true,
          "has_downloads": true,
          "has_wiki": true,
          "has_pages": false,
          "has_discussions": false,
          "forks_count": 0,
          "archived": false,
          "disabled": false,
          "open_issues_count": 0,
          "license": {
            "key": "other",
            "name": "Other",
            "spdx_id": "NOASSERTION",
            "url": null,
            "node_id": "MDc6TGljZW5zZTA="
          },
          "allow_forking": true,
          "is_template": false,
          "web_commit_signoff_required": false,
          "has_pull_requests": true,
          "pull_request_creation_policy": "all",
          "topics": {},
          "visibility": "public",
          "forks": 0,
          "open_issues": 0,
          "watchers": 0,
          "default_branch": "main"
        }
      },
      "base": {
        "label": "Tencent-Hunyuan:main",
        "ref": "main",
        "sha": "d280425cf453a153e5846c725af58de39c10b09f",
        "user": {
          "login": "Tencent-Hunyuan",
          "id": 210980732,
          "node_id": "O_kgDODJNPfA",
          "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
          "gravatar_id": "",
          "url": "https://api.github.com/users/Tencent-Hunyuan",
          "type": "Organization",
          "user_view_type": "public",
          "site_admin": false
        },
        "repo": {
          "id": 1065166974,
          "node_id": "R_kgDOP30ofg",
          "name": "HunyuanImage-3.0",
          "full_name": "Tencent-Hunyuan/HunyuanImage-3.0",
          "private": false,
          "owner": {
            "login": "Tencent-Hunyuan",
            "id": 210980732,
            "node_id": "O_kgDODJNPfA",
            "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
            "gravatar_id": "",
            "url": "https://api.github.com/users/Tencent-Hunyuan",
            "type": "Organization",
            "user_view_type": "public",
            "site_admin": false
          },
          "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
          "fork": false,
          "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0",
          "created_at": "2025-09-27T07:18:47Z",
          "updated_at": "2026-02-16T16:28:09Z",
          "pushed_at": "2026-02-03T02:40:53Z",
          "homepage": "https://hunyuan.tencent.com/image",
          "size": 93477,
          "stargazers_count": 2854,
          "watchers_count": 2854,
          "language": "Python",
          "has_issues": true,
          "has_projects": true,
          "has_downloads": true,
          "has_wiki": true,
          "has_pages": false,
          "has_discussions": false,
          "forks_count": 142,
          "archived": false,
          "disabled": false,
          "open_issues_count": 33,
          "license": {
            "key": "other",
            "name": "Other",
            "spdx_id": "NOASSERTION",
            "url": null,
            "node_id": "MDc6TGljZW5zZTA="
          },
          "allow_forking": true,
          "is_template": false,
          "web_commit_signoff_required": false,
          "has_pull_requests": true,
          "pull_request_creation_policy": "all",
          "topics": {
            "0": "image-generation",
            "1": "native-multimodal-model"
          },
          "visibility": "public",
          "forks": 142,
          "open_issues": 33,
          "watchers": 2854,
          "default_branch": "main"
        }
      },
      "_links": {
        "self": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/86"
        },
        "html": {
          "href": "https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/pull/86"
        },
        "issue": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/86"
        },
        "comments": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/86/comments"
        },
        "review_comments": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/86/comments"
        },
        "review_comment": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/comments{/number}"
        },
        "commits": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/86/commits"
        },
        "statuses": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/statuses/31bc2809f64118342f2d8c96d6e9b007c4cd0edb"
        }
      },
      "author_association": "NONE",
      "auto_merge": null,
      "active_lock_reason": null,
      "linked_issues": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/74",
      "id": 3232654918,
      "node_id": "PR_kwDOP30ofs7ArmZG",
      "number": 74,
      "state": "open",
      "locked": false,
      "title": "Instruct gradio UI",
      "user": {
        "login": "auroter",
        "id": 7332587,
        "node_id": "MDQ6VXNlcjczMzI1ODc=",
        "avatar_url": "https://avatars.githubusercontent.com/u/7332587?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/auroter",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "body": "Fixed this so that I could use HunyuanImage-3.0-Instruct locally with the Gradio UI.",
      "created_at": "2026-02-01T06:35:27Z",
      "updated_at": "2026-02-16T06:44:56Z",
      "closed_at": null,
      "merged_at": null,
      "merge_commit_sha": "f2fac302ac90469a761998081add4ca3827c46bf",
      "assignee": null,
      "assignees": {},
      "requested_reviewers": {},
      "requested_teams": {},
      "labels": {},
      "milestone": null,
      "draft": false,
      "head": {
        "label": "auroter:instruct-gradio-ui",
        "ref": "instruct-gradio-ui",
        "sha": "290a56890f407ca81e02ae643d43b2040ead7828",
        "user": {
          "login": "auroter",
          "id": 7332587,
          "node_id": "MDQ6VXNlcjczMzI1ODc=",
          "avatar_url": "https://avatars.githubusercontent.com/u/7332587?v=4",
          "gravatar_id": "",
          "url": "https://api.github.com/users/auroter",
          "type": "User",
          "user_view_type": "public",
          "site_admin": false
        },
        "repo": {
          "id": 1147076906,
          "node_id": "R_kgDORF8BKg",
          "name": "HunyuanImage-3.0",
          "full_name": "auroter/HunyuanImage-3.0",
          "private": false,
          "owner": {
            "login": "auroter",
            "id": 7332587,
            "node_id": "MDQ6VXNlcjczMzI1ODc=",
            "avatar_url": "https://avatars.githubusercontent.com/u/7332587?v=4",
            "gravatar_id": "",
            "url": "https://api.github.com/users/auroter",
            "type": "User",
            "user_view_type": "public",
            "site_admin": false
          },
          "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
          "fork": true,
          "url": "https://api.github.com/repos/auroter/HunyuanImage-3.0",
          "created_at": "2026-02-01T06:33:32Z",
          "updated_at": "2026-02-01T06:33:32Z",
          "pushed_at": "2026-02-16T06:44:55Z",
          "homepage": "https://hunyuan.tencent.com/image",
          "size": 93479,
          "stargazers_count": 0,
          "watchers_count": 0,
          "language": null,
          "has_issues": false,
          "has_projects": true,
          "has_downloads": true,
          "has_wiki": true,
          "has_pages": false,
          "has_discussions": false,
          "forks_count": 0,
          "archived": false,
          "disabled": false,
          "open_issues_count": 0,
          "license": {
            "key": "other",
            "name": "Other",
            "spdx_id": "NOASSERTION",
            "url": null,
            "node_id": "MDc6TGljZW5zZTA="
          },
          "allow_forking": true,
          "is_template": false,
          "web_commit_signoff_required": false,
          "has_pull_requests": true,
          "pull_request_creation_policy": "all",
          "topics": {},
          "visibility": "public",
          "forks": 0,
          "open_issues": 0,
          "watchers": 0,
          "default_branch": "main"
        }
      },
      "base": {
        "label": "Tencent-Hunyuan:main",
        "ref": "main",
        "sha": "1ea1526fe6d953ef44712b31014b60137bab0894",
        "user": {
          "login": "Tencent-Hunyuan",
          "id": 210980732,
          "node_id": "O_kgDODJNPfA",
          "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
          "gravatar_id": "",
          "url": "https://api.github.com/users/Tencent-Hunyuan",
          "type": "Organization",
          "user_view_type": "public",
          "site_admin": false
        },
        "repo": {
          "id": 1065166974,
          "node_id": "R_kgDOP30ofg",
          "name": "HunyuanImage-3.0",
          "full_name": "Tencent-Hunyuan/HunyuanImage-3.0",
          "private": false,
          "owner": {
            "login": "Tencent-Hunyuan",
            "id": 210980732,
            "node_id": "O_kgDODJNPfA",
            "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
            "gravatar_id": "",
            "url": "https://api.github.com/users/Tencent-Hunyuan",
            "type": "Organization",
            "user_view_type": "public",
            "site_admin": false
          },
          "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
          "fork": false,
          "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0",
          "created_at": "2025-09-27T07:18:47Z",
          "updated_at": "2026-02-16T16:28:09Z",
          "pushed_at": "2026-02-03T02:40:53Z",
          "homepage": "https://hunyuan.tencent.com/image",
          "size": 93477,
          "stargazers_count": 2854,
          "watchers_count": 2854,
          "language": "Python",
          "has_issues": true,
          "has_projects": true,
          "has_downloads": true,
          "has_wiki": true,
          "has_pages": false,
          "has_discussions": false,
          "forks_count": 142,
          "archived": false,
          "disabled": false,
          "open_issues_count": 33,
          "license": {
            "key": "other",
            "name": "Other",
            "spdx_id": "NOASSERTION",
            "url": null,
            "node_id": "MDc6TGljZW5zZTA="
          },
          "allow_forking": true,
          "is_template": false,
          "web_commit_signoff_required": false,
          "has_pull_requests": true,
          "pull_request_creation_policy": "all",
          "topics": {
            "0": "image-generation",
            "1": "native-multimodal-model"
          },
          "visibility": "public",
          "forks": 142,
          "open_issues": 33,
          "watchers": 2854,
          "default_branch": "main"
        }
      },
      "_links": {
        "self": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/74"
        },
        "html": {
          "href": "https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/pull/74"
        },
        "issue": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/74"
        },
        "comments": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/74/comments"
        },
        "review_comments": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/74/comments"
        },
        "review_comment": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/comments{/number}"
        },
        "commits": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/74/commits"
        },
        "statuses": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/statuses/290a56890f407ca81e02ae643d43b2040ead7828"
        }
      },
      "author_association": "NONE",
      "auto_merge": null,
      "active_lock_reason": null,
      "linked_issues": []
    },
    {
      "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/64",
      "id": 3089499444,
      "node_id": "PR_kwDOP30ofs64JgU0",
      "number": 64,
      "state": "open",
      "locked": false,
      "title": "Add community training pipeline to README",
      "user": {
        "login": "YaNgZhAnG-V5",
        "id": 65215487,
        "node_id": "MDQ6VXNlcjY1MjE1NDg3",
        "avatar_url": "https://avatars.githubusercontent.com/u/65215487?v=4",
        "gravatar_id": "",
        "url": "https://api.github.com/users/YaNgZhAnG-V5",
        "type": "User",
        "user_view_type": "public",
        "site_admin": false
      },
      "body": "Hi HunyuanImage Team,\r\n\r\nWe have been working with HunyuanImage 3.0 and have developed a robust [LoRA finetuning pipeline](https://github.com/PhotonAISG/hunyuan-image3-finetune) that we believe would be a great addition to your Community section.\r\n\r\nKey features of our project:\r\n1. Full support for native multi-modal training.\r\n2. Flexible training for various LoRA adapter types.\r\n\r\nOur goal is to help the community push the boundaries of unified understanding and generation. We would love to open a PR to link this repository in your documentation so others can benefit from it.\r\n\r\nLooking forward to your thoughts!",
      "created_at": "2025-12-10T14:07:41Z",
      "updated_at": "2025-12-10T14:07:41Z",
      "closed_at": null,
      "merged_at": null,
      "merge_commit_sha": "3da3582b55e8def75bc519bc440129bb321ec675",
      "assignee": null,
      "assignees": {},
      "requested_reviewers": {},
      "requested_teams": {},
      "labels": {},
      "milestone": null,
      "draft": false,
      "head": {
        "label": "PhotonAISG:main",
        "ref": "main",
        "sha": "ad741b9e13541bbde8757c396647aa64618c92ee",
        "user": {
          "login": "PhotonAISG",
          "id": 199608454,
          "node_id": "O_kgDOC-XIhg",
          "avatar_url": "https://avatars.githubusercontent.com/u/199608454?v=4",
          "gravatar_id": "",
          "url": "https://api.github.com/users/PhotonAISG",
          "type": "Organization",
          "user_view_type": "public",
          "site_admin": false
        },
        "repo": {
          "id": 1113630302,
          "node_id": "R_kgDOQmCmXg",
          "name": "HunyuanImage-3.0",
          "full_name": "PhotonAISG/HunyuanImage-3.0",
          "private": false,
          "owner": {
            "login": "PhotonAISG",
            "id": 199608454,
            "node_id": "O_kgDOC-XIhg",
            "avatar_url": "https://avatars.githubusercontent.com/u/199608454?v=4",
            "gravatar_id": "",
            "url": "https://api.github.com/users/PhotonAISG",
            "type": "Organization",
            "user_view_type": "public",
            "site_admin": false
          },
          "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
          "fork": true,
          "url": "https://api.github.com/repos/PhotonAISG/HunyuanImage-3.0",
          "created_at": "2025-12-10T08:43:56Z",
          "updated_at": "2025-12-10T09:21:53Z",
          "pushed_at": "2025-12-10T09:21:48Z",
          "homepage": "https://hunyuan.tencent.com/image",
          "size": 34748,
          "stargazers_count": 0,
          "watchers_count": 0,
          "language": "Python",
          "has_issues": false,
          "has_projects": true,
          "has_downloads": true,
          "has_wiki": true,
          "has_pages": false,
          "has_discussions": false,
          "forks_count": 0,
          "archived": false,
          "disabled": false,
          "open_issues_count": 0,
          "license": {
            "key": "other",
            "name": "Other",
            "spdx_id": "NOASSERTION",
            "url": null,
            "node_id": "MDc6TGljZW5zZTA="
          },
          "allow_forking": true,
          "is_template": false,
          "web_commit_signoff_required": false,
          "has_pull_requests": true,
          "pull_request_creation_policy": "all",
          "topics": {},
          "visibility": "public",
          "forks": 0,
          "open_issues": 0,
          "watchers": 0,
          "default_branch": "main"
        }
      },
      "base": {
        "label": "Tencent-Hunyuan:main",
        "ref": "main",
        "sha": "62da220178f4b0b7d83e91665a46a20a3ee4f7cd",
        "user": {
          "login": "Tencent-Hunyuan",
          "id": 210980732,
          "node_id": "O_kgDODJNPfA",
          "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
          "gravatar_id": "",
          "url": "https://api.github.com/users/Tencent-Hunyuan",
          "type": "Organization",
          "user_view_type": "public",
          "site_admin": false
        },
        "repo": {
          "id": 1065166974,
          "node_id": "R_kgDOP30ofg",
          "name": "HunyuanImage-3.0",
          "full_name": "Tencent-Hunyuan/HunyuanImage-3.0",
          "private": false,
          "owner": {
            "login": "Tencent-Hunyuan",
            "id": 210980732,
            "node_id": "O_kgDODJNPfA",
            "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
            "gravatar_id": "",
            "url": "https://api.github.com/users/Tencent-Hunyuan",
            "type": "Organization",
            "user_view_type": "public",
            "site_admin": false
          },
          "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
          "fork": false,
          "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0",
          "created_at": "2025-09-27T07:18:47Z",
          "updated_at": "2026-02-16T16:28:09Z",
          "pushed_at": "2026-02-03T02:40:53Z",
          "homepage": "https://hunyuan.tencent.com/image",
          "size": 93477,
          "stargazers_count": 2854,
          "watchers_count": 2854,
          "language": "Python",
          "has_issues": true,
          "has_projects": true,
          "has_downloads": true,
          "has_wiki": true,
          "has_pages": false,
          "has_discussions": false,
          "forks_count": 142,
          "archived": false,
          "disabled": false,
          "open_issues_count": 33,
          "license": {
            "key": "other",
            "name": "Other",
            "spdx_id": "NOASSERTION",
            "url": null,
            "node_id": "MDc6TGljZW5zZTA="
          },
          "allow_forking": true,
          "is_template": false,
          "web_commit_signoff_required": false,
          "has_pull_requests": true,
          "pull_request_creation_policy": "all",
          "topics": {
            "0": "image-generation",
            "1": "native-multimodal-model"
          },
          "visibility": "public",
          "forks": 142,
          "open_issues": 33,
          "watchers": 2854,
          "default_branch": "main"
        }
      },
      "_links": {
        "self": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/64"
        },
        "html": {
          "href": "https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/pull/64"
        },
        "issue": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/64"
        },
        "comments": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/issues/64/comments"
        },
        "review_comments": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/64/comments"
        },
        "review_comment": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/comments{/number}"
        },
        "commits": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/pulls/64/commits"
        },
        "statuses": {
          "href": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0/statuses/ad741b9e13541bbde8757c396647aa64618c92ee"
        }
      },
      "author_association": "NONE",
      "auto_merge": null,
      "active_lock_reason": null,
      "linked_issues": []
    }
  ],
  "discussions": [],
  "details": {
    "id": 1065166974,
    "node_id": "R_kgDOP30ofg",
    "name": "HunyuanImage-3.0",
    "full_name": "Tencent-Hunyuan/HunyuanImage-3.0",
    "private": false,
    "owner": {
      "login": "Tencent-Hunyuan",
      "id": 210980732,
      "node_id": "O_kgDODJNPfA",
      "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
      "gravatar_id": "",
      "url": "https://api.github.com/users/Tencent-Hunyuan",
      "type": "Organization",
      "user_view_type": "public",
      "site_admin": false
    },
    "description": "HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
    "fork": false,
    "url": "https://api.github.com/repos/Tencent-Hunyuan/HunyuanImage-3.0",
    "created_at": "2025-09-27T07:18:47Z",
    "updated_at": "2026-02-16T16:28:09Z",
    "pushed_at": "2026-02-03T02:40:53Z",
    "homepage": "https://hunyuan.tencent.com/image",
    "size": 93477,
    "stargazers_count": 2854,
    "watchers_count": 2854,
    "language": "Python",
    "has_issues": true,
    "has_projects": true,
    "has_downloads": true,
    "has_wiki": true,
    "has_pages": false,
    "has_discussions": false,
    "forks_count": 142,
    "archived": false,
    "disabled": false,
    "open_issues_count": 33,
    "license": {
      "key": "other",
      "name": "Other",
      "spdx_id": "NOASSERTION",
      "url": null,
      "node_id": "MDc6TGljZW5zZTA="
    },
    "allow_forking": true,
    "is_template": false,
    "web_commit_signoff_required": false,
    "has_pull_requests": true,
    "pull_request_creation_policy": "all",
    "topics": {
      "0": "image-generation",
      "1": "native-multimodal-model"
    },
    "visibility": "public",
    "forks": 142,
    "open_issues": 33,
    "watchers": 2854,
    "default_branch": "main",
    "permissions": {
      "admin": false,
      "maintain": false,
      "push": false,
      "triage": false,
      "pull": true
    },
    "temp_clone_token": "",
    "custom_properties": {},
    "organization": {
      "login": "Tencent-Hunyuan",
      "id": 210980732,
      "node_id": "O_kgDODJNPfA",
      "avatar_url": "https://avatars.githubusercontent.com/u/210980732?v=4",
      "gravatar_id": "",
      "url": "https://api.github.com/users/Tencent-Hunyuan",
      "type": "Organization",
      "user_view_type": "public",
      "site_admin": false
    },
    "network_count": 142,
    "subscribers_count": 16
  },
  "lastFetched": 1771293353097
}