Implement max_symbols_per_frame for GigaAM2

[nshmyrev]

Accurate decoding requires max_symbols_per_frame since the model uses char tokens instead of BPE.

See here:

https://github.com/salute-developers/GigaAM/blob/main/gigaam/decoding.py#L103

WER improves significantly, 16.9 -> 14.2%

Summary by CodeRabbit

• New Features

  • Enhanced decoding to support multiple symbol predictions per encoder frame, allowing for more flexible and accurate output.

• Improvements

  • Updated internal handling of encoder output lengths for better compatibility and performance.

[coderabbitai]

Walkthrough

The decoding process in the offline transducer greedy search logic was updated to support multiple symbol predictions per encoder frame, limited by a new maximum symbol count. The code now loops joiner calls per frame until a blank is predicted or the symbol cap is reached. A minor type change was also made in the decoder method.

Changes

File(s)	Change Summary
.../offline-transducer-greedy-search-nemo-decoder.cc	Modified decoding logic to allow multiple symbol predictions per encoder frame with a capped loop; updated type of encoder output length pointer in Decode method.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Poem

In the meadow of code, the joiner hops twice,
Predicting more symbols, oh isn't that nice?
No longer just one, but up to ten,
Each frame's a new warren, again and again.
With type tweaks and loops, our search is now keen,
A rabbit's delight—so swift and serene! 🐇

Note

⚡️ Unit Test Generation is now available in beta!

Learn more here, or try it out under "Finishing Touches" below.

✨ Finishing Touches

• 📝 Generate Docstrings

🧪 Generate unit tests

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

• Review comments: Directly reply to a review comment made by CodeRabbit. Example:
  • I pushed a fix in commit <commit_id>, please review it.
  • Explain this complex logic.
  • Open a follow-up GitHub issue for this discussion.
• Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
  • @coderabbitai explain this code block.
  • @coderabbitai modularize this function.
• PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
  • @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
  • @coderabbitai read src/utils.ts and explain its main purpose.
  • @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
  • @coderabbitai help me debug CodeRabbit configuration file.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

• @coderabbitai pause to pause the reviews on a PR.
• @coderabbitai resume to resume the paused reviews.
• @coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
• @coderabbitai full review to do a full review from scratch and review all the files again.
• @coderabbitai summary to regenerate the summary of the PR.
• @coderabbitai generate docstrings to generate docstrings for this PR.
• @coderabbitai generate sequence diagram to generate a sequence diagram of the changes in this PR.
• @coderabbitai generate unit tests to generate unit tests for this PR.
• @coderabbitai resolve resolve all the CodeRabbit review comments.
• @coderabbitai configuration to show the current CodeRabbit configuration for the repository.
• @coderabbitai help to get help.

Other keywords and placeholders

• Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
• Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
• Add @coderabbitai anywhere in the PR title to generate the title automatically.

CodeRabbit Configuration File (.coderabbit.yaml)

• You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
• Please see the configuration documentation for more information.
• If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Documentation and Community

• Visit our Documentation for detailed information on how to use CodeRabbit.
• Join our Discord Community to get help, request features, and share feedback.
• Follow us on X/Twitter for updates and announcements.

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (1)

sherpa-onnx/csrc/offline-transducer-greedy-search-nemo-decoder.cc (1)

48-48: Consider making max_symbols_per_frame configurable.

The hardcoded value of 10 aligns with the PR objectives for GigaAM2, but consider making this a configurable parameter (e.g., via constructor or model configuration) to support different models that might require different limits.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between dfe00f7 and 091d98f.

📒 Files selected for processing (1)

• sherpa-onnx/csrc/offline-transducer-greedy-search-nemo-decoder.cc (3 hunks)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (10)

• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest
• GitHub Check: ubuntu-latest

🔇 Additional comments (2)

sherpa-onnx/csrc/offline-transducer-greedy-search-nemo-decoder.cc (2)

64-91: LGTM! The multi-symbol per frame logic is correctly implemented.

The new loop structure properly:

• Limits symbol predictions per frame to prevent over-generation
• Maintains decoder state consistency across predictions
• Handles blank token termination correctly
• Accumulates tokens and timestamps appropriately

This implementation aligns well with the PR objectives for improving character-based model decoding.

---

107-107: Verify encoder_out_length tensor data type matches int32

Please confirm that the ONNX model’s second output really uses 32-bit integers. A mismatch between the template argument in

const int32_t* p_length = encoder_out_length.GetTensorData<int32_t>();

and the actual tensor element type will lead to crashes or incorrect length values.

Suggested actions:

• Add a runtime assertion before reading the data:

  auto& info = encoder_out_length.GetTensorTypeAndShapeInfo();
  if (info.GetElementType() != ONNX_TENSOR_ELEMENT_DATA_TYPE_INT32) {
    throw std::runtime_error("Expected encoder_out_length to have INT32 elements");
  }

• If the model actually emits INT64, change the template argument (and any downstream logic) to int64_t.

Location:

• sherpa-onnx/csrc/offline-transducer-greedy-search-nemo-decoder.cc, line 107

[csukuangfj]

Thank you for your contribution!