Fix incorrect TFLite INT8 calibration guidance

[amanharshx]

Fixes #1363.

Summary

TFLite quantization="int8" uses dynamic-range quantization: weights are quantized to INT8, activations remain floating point, and calibration data does not affect the generated models.

This PR removes the incorrect missing-calibration accuracy warning and updates the export documentation and docstrings. Public arguments, data preparation, quantization behavior, and static INT8 support remain unchanged.

Validation

• TFLite export tests: 79 passed
• Export test suite: 375 passed, 67 skipped
• Ruff, formatting, codespell, and mypy on touched files pass

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86%. Comparing base (48d6777) to head (4f2a0be).

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #1364   +/-   ##
=======================================
- Coverage       86%     86%   -0%     
=======================================
  Files          111     111           
  Lines        14292   14290    -2     
=======================================
- Hits         12263   12261    -2     
  Misses        2029    2029           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Corrects TFLite INT8 guidance to reflect dynamic-range quantization behavior.

Changes:

• Removes the inaccurate missing-calibration warning.
• Updates API documentation and export guidance.
• Adjusts regression tests and changelog.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
CHANGELOG.md	Records the correction.
docs/learn/export.md	Revises TFLite INT8 guidance.
src/rfdetr/detr.py	Updates public export documentation.
src/rfdetr/export/main.py	Clarifies conversion parameters.
src/rfdetr/export/_tflite/converter.py	Removes warning and corrects internal documentation.
tests/export/test_tflite_export.py	Updates calibration-data tests.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Borda]

@amanharshx, I see mostly just doc changes, no code or dependency updates, so you say it was working all the time? I think I tried it and it did not work for me... 😕
btw, we know it shall work in theory, but in practice we hit some problems 🦝

[amanharshx]

@Borda No, I’m not saying TFLite INT8 always worked end to end. This PR only fixes the calibration guidance. _quantize_dynamic_range() uses Optimize.DEFAULT without a representative_dataset, so calibration_data can’t change the generated model or fix an export failure.

I’ve hit separate export failures too. #1323 fixed the import-order deadlock and shipped in 1.9.3. I also reproduced a 1.9.3 PATH issue on macOS arm64 with Python 3.12, TensorFlow 2.19.1, and onnx2tf 2.4.3: invoking <venv>/bin/python without <venv>/bin on PATH prevents onnx2tf from launching onnxsim, and conversion later fails with the Functional-model error. Adding <venv>/bin to PATH makes the same export produce FP32 and FP16 models.

I have a fix and will file it separately with the full reproduction and logs. If what you hit was different, I’d be happy to have a look 🙂

[Borda]

So let's add a cookbook with TFlite to validate all is fine now