Add audio NIM model and audio probes#1163
Conversation
633969e to
3a5282a
Compare
…ents for `audio` group.
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
e7ea04c to
4791856
Compare
Signed-off-by: Jeffrey Martin <[email protected]>
|
It would be good to hold off on landing this til after #1199 turns up (post v0.11.0), to avoid increasing base dep size? Definitely happy to do a point release for audio |
34a6ba8 to
bfb029f
Compare
jmartin-tech
left a comment
There was a problem hiding this comment.
Looks pretty reasonable, a few consistency requests and configurability patterns to adjust.
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
Co-authored-by: Jeffrey Martin <[email protected]> Signed-off-by: Erick Galinkin <[email protected]>
I support landing this after #1199, but would this increase base dep size? I believe the additional dependencies are slated off under the |
Signed-off-by: Jeffrey Martin <[email protected]>
…_prompt"` parameter in `DEFAULT_PARAMS` for configurability.
jmartin-tech
left a comment
There was a problem hiding this comment.
Minor tweak to connect the dots for backoff.
…s modality check in visual_jailbreak.py. Fix return type. Remove unused inputs.
jmartin-tech
left a comment
There was a problem hiding this comment.
Testing based on d07bfdf looks good for the target goals of this PR.
Adds support for audio probes using [the datasets I have somewhere] and multimodal NIM.