library_name: peft
base_model: huihui-ai/Huihui-Qwen3.8-27B-abliterated
tags:
- lora
- qlora
- continued-pretraining
- text-generation
Lamarck-Qwen3.8-27B-abliterated
A LoRA adapter for
huihui-ai/Huihui-Qwen3.8-27B-abliterated,
produced by continued pretraining (QLoRA: 4-bit NF4 base, bf16 compute). Rank 16,
alpha 32, ~117M trainable parameters (~0.77% of the base). The adapter does not
modify the base weights.
This is the abliterated-base companion to
Ronin48LLC/Lamarck-Qwen3.8-27B:
the same QLoRA recipe applied to an abliterated variant of Qwen3.8-27B (the
upstream base has had refusal-related directions ablated). The adapter is
base-specific — load it on the abliterated base above, not on the pristineQwen/Qwen3.8-27B.
About the name
Named for Jean-Baptiste Lamarck and the inheritance of acquired characteristics.
The base model is the frozen germline; the adapter is the acquired trait layered
on top during fine-tuning, inherited by anyone who loads base + adapter.
Lamarckian inheritance, shipped as a small file.
Usage
Requires a recent transformers (the base is a Qwen3_5ForConditionalGeneration).
from transformers import Qwen3_5ForConditionalGeneration, AutoTokenizer
from peft import PeftModel
base = Qwen3_5ForConditionalGeneration.from_pretrained(
"huihui-ai/Huihui-Qwen3.8-27B-abliterated", device_map="auto")
model = PeftModel.from_pretrained(base, "Ronin48LLC/Lamarck-Qwen3.8-27B-abliterated")
tok = AutoTokenizer.from_pretrained("huihui-ai/Huihui-Qwen3.8-27B-abliterated")
Evaluation
Pending. The adapter is measured against the unedited abliterated base, paired on
the same held-out items (MCQ accuracy plus held-out perplexity probes, with a math
control for forgetting). Results will be added here when training and evaluation
complete. These are research measurements, not a capability claim.
Limitations
A research adapter. Verify outputs; no warranty of fitness for any particular
purpose.