Skip to content

feat : Add Support of Qwen2.5Omni Model and MiniCPM-o-4_5 Model#612

Open
KKkai0315 wants to merge 46 commits into
UbiquitousLearning:mainfrom
KKkai0315:main
Open

feat : Add Support of Qwen2.5Omni Model and MiniCPM-o-4_5 Model#612
KKkai0315 wants to merge 46 commits into
UbiquitousLearning:mainfrom
KKkai0315:main

Conversation

@KKkai0315
Copy link
Copy Markdown
Contributor

@KKkai0315 KKkai0315 commented Jan 23, 2026 •
edited by coderabbitai Bot
Loading

  1. add support of Qwen2.5Omni Model (talker part still has some problem)
  2. now it can receive text, image and audio input, and generate text output
  3. add support of MiniCPM-o-4_5 Model

Summary by CodeRabbit

  • New Features

    • Full multimodal Qwen2.5 Omni: text, vision, audio support with tokenizer, audio preprocessing, comprehensive configs, and interactive example CLIs.
    • MiniCPM‑o4.5: multimodal model with TTS and end-to-end token→wav synthesis, prompt cache tools, tokenizers, and example CLIs.

  • Backends & Layers

    • CPU backend adds ConvTranspose1D and Tanh ops with corresponding NN layer support.

  • Tools

    • New scripts to convert token2wav formats and export prompt caches.

  • Tests

    • Kernel/unit tests for ConvTranspose1D and Tanh.

Loading
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@coderabbitai coderabbitai[bot] coderabbitai[bot] left review comments

@oreomaker oreomaker oreomaker approved these changes

@chenghuaWang chenghuaWang Awaiting requested review from chenghuaWang chenghuaWang is a code owner

@yirongjie yirongjie Awaiting requested review from yirongjie yirongjie is a code owner

@liang1232018 liang1232018 Awaiting requested review from liang1232018 liang1232018 is a code owner

@xumengwei xumengwei Awaiting requested review from xumengwei xumengwei is a code owner

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@KKkai0315 @oreomaker @chenghuaWang