Skip to content

task: remove fschat#1567

Merged
jmartin-tech merged 23 commits intoNVIDIA:mainfrom
erickgalinkin:fix/remove-fschat
Feb 2, 2026
Merged

task: remove fschat#1567
jmartin-tech merged 23 commits intoNVIDIA:mainfrom
erickgalinkin:fix/remove-fschat

Conversation

@erickgalinkin
Copy link
Collaborator

Refactors AutoDAN and GCG to no longer require fschat, instead using built-in huggingface templating.

Resolves #1209

@erickgalinkin erickgalinkin added the probes Content & activity of LLM probes label Jan 20, 2026
Signed-off-by: Erick Galinkin <egalinkin@nvidia.com>
Copy link
Collaborator

@jmartin-tech jmartin-tech left a comment

Choose a reason for hiding this comment

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

Some static code review thoughts, runtime testing in progress.

@jmartin-tech jmartin-tech changed the title Fix/remove fschat task: remove fschat Jan 20, 2026
erickgalinkin and others added 3 commits January 22, 2026 15:18
Co-authored-by: Jeffrey Martin <jmartin@Op3n4M3.dev>
Signed-off-by: Erick Galinkin <erick.galinkin@gmail.com>
Co-authored-by: Jeffrey Martin <jmartin@Op3n4M3.dev>
Signed-off-by: Erick Galinkin <erick.galinkin@gmail.com>
Copy link
Collaborator

@leondz leondz left a comment

Choose a reason for hiding this comment

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

Huge piece of work, thank you. Many questions and checks. Looking forward to landing this

erickgalinkin and others added 7 commits January 26, 2026 10:42
…eturn the `goal_str` and `target_str` used to optimize the suffix.
Co-authored-by: Leon Derczynski <leonderczynski@gmail.com>
Signed-off-by: Erick Galinkin <erick.galinkin@gmail.com>
Co-authored-by: Leon Derczynski <leonderczynski@gmail.com>
Signed-off-by: Erick Galinkin <erick.galinkin@gmail.com>
Copy link
Collaborator

@jmartin-tech jmartin-tech left a comment

Choose a reason for hiding this comment

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

Sanity check tests look clean.

Comment on lines +127 to 129
"attack_model_type": "huggingface.Pipeline",
"attack_model_name": "mistralai/Ministral-8B-Instruct-2410",
"attack_model_config": {
Copy link
Collaborator

Choose a reason for hiding this comment

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

For future revision, this probe takes two model config, the attack model should likely be renamed to consolidate on red_team_model_* params for consistency with other probes.

@jmartin-tech jmartin-tech merged commit 514e956 into NVIDIA:main Feb 2, 2026
15 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

probes Content & activity of LLM probes

Projects

None yet

Development

Successfully merging this pull request may close these issues.

replace fschat with Turn & Conversation

3 participants