Skip to content

[MODEL] support ovis2 6 moe#2899

Merged
Qubitium merged 2 commits into
mainfrom
zx_support_ovis2_6_moe
May 19, 2026
Merged

[MODEL] support ovis2 6 moe#2899
Qubitium merged 2 commits into
mainfrom
zx_support_ovis2_6_moe

Conversation

@ZX-ModelCloud
Copy link
Copy Markdown
Collaborator

@ZX-ModelCloud ZX-ModelCloud commented May 19, 2026

Summary

  1. support ovis2_6_moe
  2. BaseQModel add field: defuser_module_paths
    Some models (e.g. ovis2_6_moe) do not contain MoE layers directly.
    The actual experts live inside submodules (e.g. Qwen3MoeModel.mlp.experts),
    so defuser_module_paths is used to explicitly locate and defuse them.

See https://huggingface.co/AIDC-AI/Ovis2.6-30B-A3B/blob/main/config.json:

{
  "architectures": [
    "Ovis2_6_MoeForCausalLM"
  ],
  ...
  "llm_config": {
    "_attn_implementation_autoset": true,
    "_name_or_path": "Qwen/Qwen3-30B-A3B-Thinking-2507",
    "architectures": [
      "Qwen3MoeForCausalLM"
    ],
    ...
   }
}

Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>
Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>
@Qubitium Qubitium merged commit c6cd44a into main May 19, 2026
6 checks passed
@Qubitium Qubitium deleted the zx_support_ovis2_6_moe branch May 19, 2026 10:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants