Reject non-finite Softmax temperature

[alaabenfatma]

Fixes the SoftmaxTemperature item in #105.

Softmax only rejected temperature <= 0, so nan and inf were accepted: nan turned every probability into NaN and inf quietly returned a uniform distribution. It now raises unless the temperature is finite and positive, matching ClipSoft's max_absolute_value check.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

🧰 Additional context used

📚 Code guidelines (1)

AGENTS.md — auto-discovered

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository: NVIDIA/structured-data-models/.coderabbit.yaml
• Review profile: QUIET
• Plan: Enterprise
• Run ID: 923e3ff4-51f1-4e6d-a8b9-eca0eb654288

📥 Commits

Reviewing files that changed from the base of the PR and between 999212f and 58a5719.

📒 Files selected for processing (2)

• sdm/processing/output/softmax.py
• test/processing/output/test_softmax.py

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • Softmax now rejects non-finite temperatures, including NaN and infinity, as well as zero and negative values.

Walkthrough

Softmax now requires temperature to be finite and positive. The invalid-temperature test covers zero, a negative value, NaN, and positive and negative infinity.

Changes

Softmax temperature validation

Layer / File(s)	Summary
Temperature contract and validation sdm/processing/output/softmax.py, test/processing/output/test_softmax.py	The documentation and constructor require finite, positive temperatures. The test covers zero, a negative value, NaN, and positive and negative infinity.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~8 minutes

Merge Risk: ⚪ Minimal · up to 58a57

Softmax now rejects invalid temperatures while accepting valid positive values; no actionable merge-blocking risk is identified.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: Softmax rejects non-finite temperatures.
Description check	✅ Passed	The description explains the invalid-temperature behavior and the change that addresses it.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

_{Comment @coderabbitai help to get the list of available commands.}

[RBendias]

Hi - thanks for the PR. We try to keep the checks as minimal as possible see: https://github.com/NVIDIA/structured-data-models/blob/e22aecb0d902c380f758de2771e3a66d41cd5022/AGENTS.md#pythonpytorch-coding-style

Add an explicit runtime check only when a bad value could otherwise be silently accepted with wrong semantics (e.g. a count mismatch that remaps members incorrectly). Do not add positivity, finiteness, range, or shape checks that fail on first use anyway.

I'll update the issue accordingly. We rather need to update ClipSoft then.