5 10 40

The Collector

IAmTheCollector

AI & ML interests

Multiple gerbils recovered.

Recent Activity

liked a model about 16 hours ago

Casual-Autopsy/Maginum-Cydoms-24B

upvoted an article about 1 month ago

New in llama.cpp: Model Management

liked a model about 1 month ago

nvidia/Nemotron-Orchestrator-8B

View all activity

Organizations

None yet

liked a model about 16 hours ago

Casual-Autopsy/Maginum-Cydoms-24B

Text Generation • 24B • Updated Dec 11, 2025 • 71 • 17

upvoted an article about 1 month ago

Article

New in llama.cpp: Model Management

Dec 11, 2025

•

108

liked a model about 1 month ago

nvidia/Nemotron-Orchestrator-8B

Text Generation • 8B • Updated Dec 2, 2025 • 26k • 514

liked 2 models about 2 months ago

Tongyi-MAI/Z-Image-Turbo

Text-to-Image • Updated 10 days ago • 409k • • 3.79k

ArliAI/gpt-oss-20b-Derestricted

Text Generation • 21B • Updated Nov 28, 2025 • 1.33k • 80

reacted to grimjim's post with 🤗 about 2 months ago

Post

3283

I wanted to call attention to Arli Ai's success in applying my recent modifications to refusal ablation to a MoE model successfully. Nice work, @OwenArli !
ArliAI/GLM-4.5-Air-Derestricted
Ablation on a MoE model is no small thing; I expect preserving norms/magnitudes during intervention better respects routing compared to naive refusal ablation.

(I would have tagged their org earlier, but that feature seemed to be broken via "@")

ArliAI

4 replies

upvoted an article about 2 months ago

Article

DeLERP: Decomposed Linear Interpolation for Model Merging

Nov 20, 2025

•

liked a model about 2 months ago

grimjim/gemma-3-12b-it-norm-preserved-biprojected-abliterated

Image-Text-to-Text • 12B • Updated 13 days ago • 359 • 21

upvoted an article about 2 months ago

Article

Projected Abliteration

Oct 25, 2025

•

reacted to grimjim's post with 👍 about 2 months ago

Post

3323

Going forward, I will be adopting the term Magnitude-Preserving Orthogonal Ablation (MPOA) for my recent work in mitigating model damage from abliteration. The technique potentially unlocks reasoning capacity previously occupied with safety refusal processing.

For details, start here: https://huggingface.co/blog/grimjim/norm-preserving-biprojected-abliteration

Showcase results: grimjim/gemma-3-12b-it-norm-preserved-biprojected-abliterated (outperforms base instruct on UGI Leaderboard NatInt)

(The existing name, while technically accurate, was a bit of a mouthful.)