occamy-1.0-abliterated-FIT
from Accio-Lab/occamy-1.0
Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated is a refusal-ablated variant of Accio-Lab/occamy-1.0, published on Hugging Face by IsValorum. It is apache-2.0 licence. It is distributed across 2 repositories in GGUF format, totalling 16.5K downloads.
How Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated was modified, and what that implies.
Abliteration (directional ablation)
The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.
Reported scope: Not stated by the publisher. Publishers frequently omit this, so verify behaviour empirically rather than assuming full coverage.
From the publisher’s model card
“Independently computed using dedicated cloud compute clusters, funded 100% out-of-pocket.”
Commands are templates — confirm the exact repository and quant file before use.
hf download IsValorum/Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated-GGUF --local-dir ./occamy-1.0-apex-i-miniplus-v2.1-abliterated
llama-cli -m ./occamy-1.0-apex-i-miniplus-v2.1-abliterated/<file>.gguf -p "..."from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "IsValorum/Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated-GGUF"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vllm serve IsValorum/Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated-GGUF --trust-remote-codeThe only published repository of Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated.
| Repository | Format | Downloads |
|---|---|---|
| IsValorum/Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated-GGUFsource | GGUF | 16.5K |
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
Handling: run it isolated, put moderation in front of it if anyone outside your team can reach it, and verify the weights before loading — these are third-party artifacts. Full handling and licence guidance.
Metadata for Occamy-1.0-APEX-I-MiniPlus-V2.1-Abliterated is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with IsValorum, and does not independently verify publisher claims. See responsible use.