tt-metal: [Bounty $500-1500] mobilenet-v2 bring up using C++ APIs of TTNN (N150 and N300)

Bounty: Bring up mobileNet-v2 via TTNN C++ APIs

Description

Mobilenetv2 current implementation is using python TTNN APIs.

This bounty involves implementing Mobilenet-v2 model using C++ TTNN APIs. The final program must be linked against ttnncpp instead of ttnn, as the objective is to get rid of all Python linkage.

Target Model

  • Model: mobilenet-v2
  • Model Input Target: input shape/resolution (224, 224, 3)
  • Model Hardware Target: Wormhole (N150 or N300)
  • Target FPS: 444 Device FPS (all ops running on device) This can be broken down into two stages based on achieved performance. Easy and Medium.

Difficulty Ratings:

  • Easy ($500): Implement the model via C++ TTNN APIs achieving ~222 FPS. If certain op APIs are missing in C++ and/or not working as expected, issues can be filed with engineering to add support.

  • Medium ($1500): Optimize the model in C++ TTNN APIs to match the perf of the python version of the implementation ~333 FPS. Expect all ops to run on device/via C++ TTNN APIs.

Success Criteria

1. Functional Bring-Up

  • the model to run on N150 and/or N300 with no errors.

2. Performance Validation

  • Clear documentation of the throughput and latency.
  • Achieve at least the baseline targets for the selected difficulty rating.

3. Accuracy Validation

4. Documentation

  • Clear instructions for building, installing dependencies, and running on N150 and/or N300.
  • Any relevant parameter tuning or known issues noted.

Bring-up Checklist

  • Model functionality
  • Post commit CI passes
  • Model regression CI testing passes (if applicable)
  • Device performance regression CI testing passes (if applicable)
  • New/Existing tests provide coverage for changes
  • Demo script to show the inference output

Notes

  • Work may involve multiple pull requests across relevant repositories (e.g., TT-metal, or others).
  • If additional patches or dependencies are required, please include a short guide for maintainers.

About this issue

  • Original URL
  • State: open
  • Created 2 years ago
  • Comments: 18 (9 by maintainers)

Most upvoted comments

@sabira-mcw can you please provide some feedback here?