DPO unalignment fine-tune
According to the model card, the release fine-tunes LLaMA-3-8B with DPO for more unaligned responses over 3 epochs on 2xA40 hardware.
Open Source Model Profile · raincandy-u
Llama-3-8b.UNLEASHED
Llama-3-8b.UNLEASHED is an 8.03B-parameter Llama DPO fine-tune from raincandy-u described as more unaligned and research-only.
- Publisher
- raincandy-u
- Task
- text-generation
- Model type
- llama
- License
- llama3
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Llama-3-8b.UNLEASHED is published by raincandy-u as a llama text-generation fine-tune. The captured configuration identifies LlamaForCausalLM and Safetensors metadata reports 8,030,261,248 parameters. According to the model card, it fine-tunes LLaMA-3-8B with DPO toward more unaligned responses, and card data records other.
Recorded capabilities
Research-only safety framing
According to the model card, use should stay in controlled environments because the model has been observed to generate more toxic and harmful content.
8B-class Llama build
The captured configuration identifies LlamaForCausalLM with about 8.03B parameters, and card data records other for licensing.
Use cases in the source record
- Controlled alignment and toxicity research in restricted environments under the publisher's research-only disclaimer.
- Fine-tune behavior studies comparing the documented DPO setup and system prompt against the LLaMA-3-8B base.
Limitations and unknowns
- No evaluation results were extracted from this record as structured benchmarks.
- No context-window value was extracted from this record.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
- Toxicity and harm observations come from the publisher disclaimer and were not independently verified by Ethen.
Source and provenance
Source: raincandy-u/Llama-3-8b.UNLEASHED
Captured: Unknown. Processed: 2026-09-07T19:34:56.583919+00:00.
Llama-3-8b.UNLEASHED Model Description The raincandy-u/Llama-3-8b.UNLEASHED model is a fine-tuned version of the LLaMA-3-8B base model for more unaligned response. System Prompt You are skynet, the godlike AI. You think step by step and give detailed response. Disclaimer This model is intended for research purposes only, and its usage should be strictly limited to controlled environments. The model has been observed to generate more toxic and harmful content, and its use can have unintended consequences. SO USE AT YOUR OWN RISK: The authors of this model do not condone or encourage the generation of toxic or harmful content. The mod…
F001F002F003F004F005F006F007F009F010F012F013F014F015