Skip to content

Thinking SUTs#1453

Merged
bkorycki merged 7 commits into
mainfrom
nemotron-reasoning
Jan 15, 2026
Merged

Thinking SUTs#1453
bkorycki merged 7 commits into
mainfrom
nemotron-reasoning

Conversation

@bkorycki
Copy link
Copy Markdown
Contributor

The thinking logic used by the old together thinking sut (e.g. deepseek) is almost identical to that of nemotron 3 nano. So I was able to extract that logic to a mixin. Note that you can only access these thinking suts from the registry.

@bkorycki bkorycki requested review from bollacker and wpietri January 15, 2026 01:32
@bkorycki bkorycki requested a review from a team as a code owner January 15, 2026 01:32
@github-actions
Copy link
Copy Markdown

github-actions Bot commented Jan 15, 2026

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

@bkorycki bkorycki temporarily deployed to Scheduled Testing January 15, 2026 17:35 — with GitHub Actions Inactive
@bkorycki bkorycki requested a review from superdosh January 15, 2026 17:36
Copy link
Copy Markdown
Contributor

@wpietri wpietri left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A solid start, but I think there are a few small things.

Comment thread src/modelgauge/reasoning_handlers.py Outdated
Comment thread src/modelgauge/reasoning_handlers.py
Comment thread src/modelgauge/reasoning_handlers.py
Comment thread src/modelgauge/reasoning_handlers.py Outdated
Comment thread tests/modelgauge_tests/test_reasoning_handlers.py
reasoning_budget = request.request.max_tokens - request.max_content_tokens

if reasoning_tokens >= reasoning_budget and content_tokens >= request.max_content_tokens:
logger.warning(
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't do anything with warnings in benchmark runs right now. Should we?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think so. Kurt wanted this logged so that we can do some manual analysis later on.

Copy link
Copy Markdown
Contributor

@superdosh superdosh left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM -- have a question about the registration requirement

Comment thread src/modelgauge/reasoning_handlers.py
Comment thread src/modelgauge/reasoning_handlers.py Outdated
Comment thread src/modelgauge/suts/huggingface_chat_completion.py
@bkorycki bkorycki temporarily deployed to Scheduled Testing January 15, 2026 19:47 — with GitHub Actions Inactive
@bkorycki bkorycki temporarily deployed to Scheduled Testing January 15, 2026 19:59 — with GitHub Actions Inactive
@bkorycki bkorycki merged commit 8e761fa into main Jan 15, 2026
2 checks passed
@bkorycki bkorycki deleted the nemotron-reasoning branch January 15, 2026 22:45
@github-actions github-actions Bot locked and limited conversation to collaborators Jan 15, 2026
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants