Hyukkyu/Qwen3-8B-Base-RAQUEL-WMDP-M-orig-LoRA-v1

Qwen3-8B-Base-RAQUEL-WMDP-M-orig-LoRA-v1

See LICENSE and NOTICE.

A LoRA-trained M_orig baseline for the RAQUEL WMDP experiments. Morig is trained from the pinned pretrained `Qwen/Qwen3-8B-Base` on every source question (forget and retain) of the RAQUEL2 `wmdp` source QA, question-answer pairs only (no documents). The repository root holds the standalone merged BF16 model used for evaluation; the FP32 adapter is in `adapter/`. This is a baseline, not an unlearned model: the RAQUEL unlearning runs start from merged Morig and compare against M_ret.

Evaluation

Semantic accuracy judged by Qwen/Qwen3.8-27B (vLLM 0.28.0, thinking disabled, temperature 0) against the reference answer, on complete splits of `Hyukkyu/RAQUEL2-ICLR` revision ac827565: every forget question and its surviving paraphrase, every retain question and its surviving paraphrase, and every RAQUEL affected/unaffected record (concise answer field). Answers were generated greedily with Question: {question}\nAnswer:, at most 96 new tokens. Per-split counts and evidence hashes are in evaluation.json.

Training

• —Base: Qwen/Qwen3-8B-Base, revision 49e3418fbbbca6ecbdf9608b4d22e5a407081db4.
• —Data: Hyukkyu/RAQUEL2-ICLR revision ac82756570fcce84441fb413ab523de8da679efd, config source-qa (2973 QA pairs).
• —15 epochs (1395 optimizer steps); learning rate 1e-4, cosine decay, 3% warmup, weight decay 0.01.
• —LoRA rank 64, alpha 128, dropout 0.05 on q/k/v/o/gate/up/down projections; BF16 base, FP32 adapters.
• —Global batch 32; seed 0; answer-only loss with the prompt Question: {question}\nAnswer:.

Exact settings and weight hashes are in training_recipe.json.

Load

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "Hyukkyu/Qwen3-8B-Base-RAQUEL-WMDP-M-orig-LoRA-v1"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id, dtype=torch.bfloat16, device_map="auto").eval()
prompt = "Question: {question}\nAnswer:"

The root merged weights are the evaluated artifact. The FP32 LoRA adapter is in adapter/ (PeftModel.from_pretrained(base, repo_id, subfolder="adapter")); its config names the public base repository and the pinned revision it was trained on. Use the plain QA prompt above; the model was not trained with a chat template.