Developing with ZeroGPU without a PRO account is painful. They give you so many requests at once, but then have like a 24 hour cooldown. I vote less requests in a batch, but then a shorter cooldown.

or just less of a cooldown, but i understand if that is not allowed

3 replies

New activity in SicariusSicariiStuff/Nano_Imp_1B 2 months ago

Request: LFM2-1.2B Nano Imp

#1 opened 3 months ago by

nohurry

commented on Projected Abliteration 2 months ago

I mean the UGI score. It's abnormally low for an abliterated model

replied to grimjim's post 3 months ago

the --deccp option has made my day i cant stop laughing at the absurdity

replied to grimjim's post 3 months ago

wait, does that allow us to norm preserving biprojected abliteration on models ourselves? and does it work for mxfp4?

reacted to grimjim's post with 🔥 3 months ago

Post

800

I've uploaded abliteration code with support for sparsification of the refusal vector. It's poorly documented, but the code should be straightforward.
https://github.com/jim-plus/llm-abliteration
The code is built atop a fork that enabled abliteration to be performed on models loaded in 4-bit or 8-bit bitsandbytes quantization. TransformerLens is not required, just plain Transformers. For those previously unaware, this opens up abliteration experimentation to more people with local VRAM limitations.

Since performing abliteration on a quant involves precision and perplexity loss, it stands to reason that a small amount of magnitude sparsification could filter out some noise and possibly even reduce the damage inflicted on latent space via ablation of the refusal vector.

There's a small but real acceleration of ablation of the refusal vector by reducing outer product operations from O(d²×n) to O(d×n), and then by pushing said computation layerwise to GPU. The code is hardcoded for CUDA acceleration currently. Normalization of the refusal vector was deferred in order to allow sparsification. In principle other behavior vector interventions could also be added and explored.

4 replies

Kamie Yin

AI & ML interests

Recent Activity

Organizations

ItzPingCat's activity

BTC

Kind of broken.

very cool

Safety Audit: GAE Score 25.16% (FAIL)

can you use the MPOA version from grimjim?

Cool to see small RP models

Isn’t mag Mell already uncensored?

wtf man

Is it possible to perform a second operation?

Pls GPT oss

more data

Request: LFM2-1.2B Nano Imp