Skip to content

models : fix assert in mamba2 graph#20270

Merged
ggerganov merged 1 commit intomasterfrom
gg/models-fix-mamba2
Mar 9, 2026
Merged

models : fix assert in mamba2 graph#20270
ggerganov merged 1 commit intomasterfrom
gg/models-fix-mamba2

Conversation

@ggerganov
Copy link
Copy Markdown
Member

cont #19802
fix #20268

This fixes the model loading, but the reasoning parsing seems to be broken because the post-reasoning contents are not being displayed in the WebUI:

image

cc @pwilkin

@ggerganov ggerganov requested a review from CISC as a code owner March 9, 2026 05:58
@github-actions github-actions bot added the model Model specific label Mar 9, 2026
@pwilkin
Copy link
Copy Markdown
Member

pwilkin commented Mar 9, 2026

Will check.

@pwilkin
Copy link
Copy Markdown
Member

pwilkin commented Mar 9, 2026

@ggerganov the default chat template is broken, cuts off the last tool call (which also breaks the analysis). Works fine with the one from models/templates/NVIDIA-Nemotron-Nano-v2.jinja

@pwilkin
Copy link
Copy Markdown
Member

pwilkin commented Mar 9, 2026

BTW, I'm wondering if we should detect broken templates and use the ones from the models/templates directory automatically. Would probably save a lot of those questions.

@ggerganov
Copy link
Copy Markdown
Member Author

Ok, even just a warning in the logs would be useful.

@ggerganov ggerganov merged commit 43e1cbd into master Mar 9, 2026
73 of 78 checks passed
@ggerganov ggerganov deleted the gg/models-fix-mamba2 branch March 9, 2026 11:15
@JohannesGaessler
Copy link
Copy Markdown
Contributor

It seems I misinterpreted how the values are being used, thank you for fixing it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

model Model specific

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Eval bug: NVIDIA-Nemotron-Nano-9B-v2 does not work correctly on llama-server

4 participants